feat(benchmarks): expand multilingual stemming quality evaluation

* cover all Radixor dictionary languages
* add PRIMARY_OUTPUT, ANY_CANDIDATE, and ALL_CANDIDATES policies
* measure pairwise over-stemming and under-stemming
* add balanced accuracy and complementary quality metrics
* compare single-output and multi-output stemmers fairly
* improve result validation, reporting, and documentation
* move stemming quality tests into the standard test source set
* preserve the existing JMH benchmark structure and badge output

.classpath

@@ -1,46 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<classpath>
-	<classpathentry kind="src" output="bin/main" path="src/main/java">
-		<attributes>
-			<attribute name="gradle_scope" value="main"/>
-			<attribute name="gradle_used_by_scope" value="main,test,jmh"/>
-		</attributes>
-	</classpathentry>
-	<classpathentry kind="src" output="bin/test" path="src/test/java">
-		<attributes>
-			<attribute name="gradle_scope" value="test"/>
-			<attribute name="gradle_used_by_scope" value="test,jmh"/>
-			<attribute name="test" value="true"/>
-		</attributes>
-	</classpathentry>
-	<classpathentry kind="src" output="bin/main" path="src/main/resources">
-		<attributes>
-			<attribute name="gradle_scope" value="main"/>
-			<attribute name="gradle_used_by_scope" value="main,test,jmh"/>
-		</attributes>
-	</classpathentry>
-	<classpathentry kind="src" output="bin/jmh" path="src/jmh/java">
-		<attributes>
-			<attribute name="gradle_scope" value="jmh"/>
-			<attribute name="gradle_used_by_scope" value="jmh"/>
-			<attribute name="test" value="true"/>
-		</attributes>
-	</classpathentry>
-	<classpathentry kind="src" output="bin/jmh" path="build/third-party/snowball/source/libstemmer_java-3.0.1/java">
-		<attributes>
-			<attribute name="gradle_scope" value="jmh"/>
-			<attribute name="gradle_used_by_scope" value="jmh"/>
-			<attribute name="test" value="true"/>
-		</attributes>
-	</classpathentry>
-	<classpathentry kind="src" output="bin/test" path="src/test/resources">
-		<attributes>
-			<attribute name="gradle_scope" value="test"/>
-			<attribute name="gradle_used_by_scope" value="test,jmh"/>
-			<attribute name="test" value="true"/>
-		</attributes>
-	</classpathentry>
-	<classpathentry kind="con" path="org.eclipse.jdt.launching.JRE_CONTAINER/org.eclipse.jdt.internal.debug.ui.launcher.StandardVMType/JavaSE-21/"/>
-	<classpathentry kind="con" path="org.eclipse.buildship.core.gradleclasspathcontainer"/>
-	<classpathentry kind="output" path="bin/default"/>
-</classpath>

.github/workflows/benchmarks.yml

@@ -19,6 +19,10 @@ on:
       - 'gradlew.bat'
       - '.github/workflows/benchmarks.yml'
 
+concurrency:
+  group: benchmarks-${{ github.ref }}
+  cancel-in-progress: true
+
 jobs:
   jmh:
     runs-on: ubuntu-latest
@@ -31,15 +35,17 @@ jobs:
       - name: Check out sources
         uses: actions/checkout@v4
 
+      - name: Validate Gradle wrapper
+        uses: gradle/actions/wrapper-validation@v4
+
       - name: Set up JDK 21
         uses: actions/setup-java@v4
         with:
           distribution: temurin
           java-version: '21'
-          cache: gradle
 
-      - name: Make Gradle executable
-        run: chmod +x ./gradlew
+      - name: Set up Gradle caching and instrumentation
+        uses: gradle/actions/setup-gradle@v4
 
       - name: Verify reproducibility inputs
         shell: bash
@@ -50,7 +56,7 @@ jobs:
           test -f gradle/verification-metadata.xml
 
       - name: Run JMH benchmarks
-        run: ./gradlew clean jmh --no-daemon
+        run: ./gradlew clean jmh -Pjmh.includes='.*EnglishStemmerComparisonBenchmark.*' --no-daemon
 
       - name: Upload JMH reports
         uses: actions/upload-artifact@v4

.github/workflows/build.yml

@@ -51,7 +51,7 @@ jobs:
           test -f gradle/verification-metadata.xml
 
       - name: Execute build, tests, PMD, coverage, Javadoc, distribution packaging, and SBOM generation
-        run: ./gradlew --no-daemon clean build pmdMain javadoc jacocoTestReport distZip cyclonedxBom
+        run: ./gradlew --no-daemon clean ciRelease distZip pmdMain javadoc jacocoCiReleaseReport cyclonedxBom
 
       - name: Upload SBOM
         if: always()
@@ -70,8 +70,8 @@ jobs:
         with:
           name: test-reports
           path: |
-            build/reports/tests/test
-            build/test-results/test
+            build/reports/tests
+            build/test-results
           if-no-files-found: warn
           retention-days: 14
 
@@ -90,8 +90,8 @@ jobs:
         with:
           name: coverage-reports
           path: |
-            build/reports/jacoco/test/html
-            build/reports/jacoco/test/jacocoTestReport.xml
+            build/reports/jacoco/jacocoCiReleaseReport/html
+            build/reports/jacoco/jacocoCiReleaseReport/jacocoCiReleaseReport.xml
           if-no-files-found: warn
           retention-days: 14
 
@@ -156,11 +156,22 @@ jobs:
           test -f gradle.properties
           test -f gradle/verification-metadata.xml
 
-      - name: Build release distribution, signed Maven bundle, and SBOM
+      - name: Build release inputs, signed Maven bundle, and SBOM
         env:
           SIGNING_KEY: ${{ secrets.SIGNING_KEY }}
           SIGNING_PASSWORD: ${{ secrets.SIGNING_PASSWORD }}
-        run: ./gradlew --no-daemon clean build pmdMain javadoc jacocoTestReport distZip cyclonedxBom centralBundle
+        run: ./gradlew --no-daemon clean ciRelease distZip pmdMain javadoc jacocoCiReleaseReport cyclonedxBom centralBundle
+
+      - name: Generate release changelog
+        shell: bash
+        run: |
+          set -euo pipefail
+          chmod +x ./tools/generate-release-notes.sh
+          mkdir -p build/generated/release-notes
+          ./tools/generate-release-notes.sh "${GITHUB_REF_NAME}" > build/generated/release-notes/CHANGELOG.md
+
+      - name: Package release distribution
+        run: ./gradlew --no-daemon distZip
 
       - name: Publish bundle to Maven Central
         shell: bash
@@ -188,7 +199,7 @@ jobs:
       - name: Publish GitHub release assets
         uses: softprops/action-gh-release@v2
         with:
-          generate_release_notes: true
+          body_path: build/generated/release-notes/CHANGELOG.md
           files: |
             build/distributions/*.zip
             build/reports/sbom/radixor-sbom.json

.github/workflows/pages.yml

@@ -5,6 +5,8 @@ on:
     branches:
       - main
     paths:
+      - 'docs/**'
+      - 'mkdocs.yml'
       - 'src/main/**'
       - 'src/test/**'
       - 'src/jmh/**'
@@ -17,6 +19,8 @@ on:
       - 'gradlew'
       - 'gradlew.bat'
       - '.github/workflows/pages.yml'
+      - '.github/workflows/benchmarks.yml'
+      - 'tools/generate-pages-badges.py'
   workflow_dispatch:
 
 permissions:
@@ -49,6 +53,14 @@ jobs:
       - name: Set up Gradle caching and instrumentation
         uses: gradle/actions/setup-gradle@v4
 
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.x'
+
+      - name: Install MkDocs Material
+        run: python -m pip install --upgrade pip mkdocs-material
+
       - name: Verify reproducibility inputs
         shell: bash
         run: |
@@ -58,7 +70,7 @@ jobs:
           test -f gradle/verification-metadata.xml
 
       - name: Build reports for publication
-        run: ./gradlew --no-daemon clean build pmdMain javadoc jacocoTestReport pitest jmh cyclonedxBom
+        run: ./gradlew --no-daemon clean ciRelease pmdMain javadoc jacocoCiReleaseReport pitest jmh -Pjmh.includes='.*EnglishStemmerComparisonBenchmark.*' cyclonedxBom
 
       - name: Prepare gh-pages worktree
         shell: bash
@@ -81,25 +93,30 @@ jobs:
         run: |
           set -euo pipefail
 
+          TEST_REPORT_DIR="build/reports/tests/ciRelease"
+          JACOCO_REPORT_DIR="build/reports/jacoco/jacocoCiReleaseReport"
+
           SITE_DIR=".gh-pages"
           RUN_DIR="${SITE_DIR}/builds/${GITHUB_RUN_NUMBER}"
+          RUN_METRICS_DIR="${RUN_DIR}/metrics"
           LATEST_DIR="${SITE_DIR}/builds/latest"
+          LATEST_METRICS_DIR="${LATEST_DIR}/metrics"
 
           mkdir -p "${RUN_DIR}"
           rm -rf "${LATEST_DIR}"
-          mkdir -p "${LATEST_DIR}"
+          mkdir -p "${LATEST_DIR}" "${RUN_METRICS_DIR}" "${LATEST_METRICS_DIR}"
 
           cp -R build/docs/javadoc "${RUN_DIR}/javadoc"
           cp -R build/docs/javadoc "${LATEST_DIR}/javadoc"
 
-          cp -R build/reports/tests/test "${RUN_DIR}/test"
-          cp -R build/reports/tests/test "${LATEST_DIR}/test"
+          cp -R "${TEST_REPORT_DIR}" "${RUN_DIR}/test"
+          cp -R "${TEST_REPORT_DIR}" "${LATEST_DIR}/test"
 
           cp -R build/reports/pmd "${RUN_DIR}/pmd"
           cp -R build/reports/pmd "${LATEST_DIR}/pmd"
 
-          cp -R build/reports/jacoco/test/html "${RUN_DIR}/coverage"
-          cp -R build/reports/jacoco/test/html "${LATEST_DIR}/coverage"
+          cp -R "${JACOCO_REPORT_DIR}/html" "${RUN_DIR}/coverage"
+          cp -R "${JACOCO_REPORT_DIR}/html" "${LATEST_DIR}/coverage"
 
           cp -R build/reports/pitest "${RUN_DIR}/pitest"
           cp -R build/reports/pitest "${LATEST_DIR}/pitest"
@@ -108,12 +125,17 @@ jobs:
           JMH_CSV_LINK=''
           JMH_TXT_LATEST_LINK=''
           JMH_CSV_LATEST_LINK=''
+          JMH_TXT_REPORT_MD='- Benchmark results (TXT): not currently available'
+          JMH_CSV_REPORT_MD='- Benchmark results (CSV): not currently available'
           DEPENDENCY_CHECK_LINK=''
           DEPENDENCY_CHECK_LATEST_LINK=''
+          DEPENDENCY_CHECK_REPORT_MD='- Dependency vulnerability report: not currently available'
           SBOM_JSON_LINK=''
           SBOM_XML_LINK=''
           SBOM_JSON_LATEST_LINK=''
           SBOM_XML_LATEST_LINK=''
+          SBOM_JSON_REPORT_MD='- SBOM (JSON): not currently available'
+          SBOM_XML_REPORT_MD='- SBOM (XML): not currently available'
 
           if [ -d "build/reports/jmh" ]; then
             cp -R build/reports/jmh "${RUN_DIR}/jmh"
@@ -122,10 +144,12 @@ jobs:
             if [ -f "${RUN_DIR}/jmh/jmh-results.txt" ]; then
               JMH_TXT_LINK='<li><a href="./jmh/jmh-results.txt">Benchmark Results (TXT)</a></li>'
               JMH_TXT_LATEST_LINK='<li><a href="./builds/latest/jmh/jmh-results.txt">Benchmark Results (TXT)</a></li>'
+              JMH_TXT_REPORT_MD='- [JMH benchmark results (TXT)](https://leogalambos.github.io/Radixor/builds/latest/jmh/jmh-results.txt)'
             fi
             if [ -f "${RUN_DIR}/jmh/jmh-results.csv" ]; then
               JMH_CSV_LINK='<li><a href="./jmh/jmh-results.csv">Benchmark Results (CSV)</a></li>'
               JMH_CSV_LATEST_LINK='<li><a href="./builds/latest/jmh/jmh-results.csv">Benchmark Results (CSV)</a></li>'
+              JMH_CSV_REPORT_MD='- [JMH benchmark results (CSV)](https://leogalambos.github.io/Radixor/builds/latest/jmh/jmh-results.csv)'
             fi
 
             HAS_JMH="true"
@@ -140,6 +164,7 @@ jobs:
             if [ -f "${RUN_DIR}/dependency-check/dependency-check-report.html" ]; then
               DEPENDENCY_CHECK_LINK='<li><a href="./dependency-check/dependency-check-report.html">Dependency Vulnerability Report</a></li>'
               DEPENDENCY_CHECK_LATEST_LINK='<li><a href="./builds/latest/dependency-check/dependency-check-report.html">Dependency Vulnerability Report</a></li>'
+              DEPENDENCY_CHECK_REPORT_MD='- [Dependency vulnerability report](https://leogalambos.github.io/Radixor/builds/latest/dependency-check/dependency-check-report.html)'
             fi
           fi
 
@@ -150,6 +175,33 @@ jobs:
             SBOM_XML_LINK='<li><a href="./sbom/radixor-sbom.xml">SBOM (XML)</a></li>'
             SBOM_JSON_LATEST_LINK='<li><a href="./builds/latest/sbom/radixor-sbom.json">SBOM (JSON)</a></li>'
             SBOM_XML_LATEST_LINK='<li><a href="./builds/latest/sbom/radixor-sbom.xml">SBOM (XML)</a></li>'
+            SBOM_JSON_REPORT_MD='- [SBOM (JSON)](https://leogalambos.github.io/Radixor/builds/latest/sbom/radixor-sbom.json)'
+            SBOM_XML_REPORT_MD='- [SBOM (XML)](https://leogalambos.github.io/Radixor/builds/latest/sbom/radixor-sbom.xml)'
+          fi
+
+          python3 \
+            ./tools/generate-pages-badges.py \
+            --jacoco-xml "${JACOCO_REPORT_DIR}/jacocoCiReleaseReport.xml" \
+            --pit-xml build/reports/pitest/mutations.xml \
+            --jmh-csv build/reports/jmh/jmh-results.csv \
+            --run-metrics-dir "${RUN_METRICS_DIR}" \
+            --latest-metrics-dir "${LATEST_METRICS_DIR}"
+
+          COVERAGE_BADGE_LINK='<li><a href="./metrics/coverage-badge.json">Coverage Badge Metadata</a></li>'
+          COVERAGE_BADGE_LATEST_LINK='<li><a href="./builds/latest/metrics/coverage-badge.json">Coverage Badge Metadata</a></li>'
+          MUTATION_BADGE_LINK='<li><a href="./metrics/pitest-badge.json">Mutation Badge Metadata</a></li>'
+          MUTATION_BADGE_LATEST_LINK='<li><a href="./builds/latest/metrics/pitest-badge.json">Mutation Badge Metadata</a></li>'
+          COVERAGE_BADGE_REPORT_MD='- [Coverage badge metadata](https://leogalambos.github.io/Radixor/builds/latest/metrics/coverage-badge.json)'
+          MUTATION_BADGE_REPORT_MD='- [Mutation badge metadata](https://leogalambos.github.io/Radixor/builds/latest/metrics/pitest-badge.json)'
+
+          if [ ! -f "${RUN_METRICS_DIR}/coverage-badge.json" ]; then
+            COVERAGE_BADGE_LINK='<li>Coverage Badge Metadata: not available</li>'
+            COVERAGE_BADGE_LATEST_LINK='<li>Coverage Badge Metadata: not available</li>'
+            COVERAGE_BADGE_REPORT_MD='- Coverage badge metadata: not currently available'
+          fi
+
+          if [ ! -f "${RUN_METRICS_DIR}/pitest-badge.json" ]; then
+            MUTATION_BADGE_REPORT_MD='- Mutation badge metadata: not currently available'
           fi
 
           cat > "${RUN_DIR}/index.html" <<EOF
@@ -172,12 +224,14 @@ jobs:
             <p class="meta">Build ${GITHUB_RUN_NUMBER} from commit ${GITHUB_SHA}</p>
             <ul>
               <li><a href="./javadoc/">Javadoc</a></li>
-              <li><a href="./test/">Test Report</a></li>
+              <li><a href="./test/">Release Verification Test Report (ciRelease)</a></li>
               <li><a href="./pmd/main.html">PMD Report</a></li>
               <li><a href="./coverage/">Coverage Report</a></li>
               ${DEPENDENCY_CHECK_LINK:-<li>Dependency Vulnerability Report: not available</li>}
               ${SBOM_JSON_LINK:-<li>SBOM (JSON): not available</li>}
               ${SBOM_XML_LINK:-<li>SBOM (XML): not available</li>}
+              ${COVERAGE_BADGE_LINK}
+              ${MUTATION_BADGE_LINK}
               <li><a href="./pitest/">Mutation Testing Report</a></li>
               $(
                 [ "${HAS_JMH}" = "true" ] && { echo "${JMH_TXT_LINK:-<li>Benchmark Results (TXT): not available</li>}"; echo "${JMH_CSV_LINK:-<li>Benchmark Results (CSV): not available</li>}"; } \
@@ -192,65 +246,87 @@ jobs:
 
           cp "${RUN_DIR}/index.html" "${LATEST_DIR}/index.html"
 
-          cat > "${SITE_DIR}/.nojekyll" <<EOF
+          cat > docs/reports.md <<EOF
+          # CI Reports
+
+          Radixor publishes durable CI artifacts to GitHub Pages on every qualifying run of \`.github/workflows/pages.yml\`.
+
+          ## Primary report entry points
+
+          - [Latest build summary](https://leogalambos.github.io/Radixor/builds/latest/)
+          - [Javadoc](https://leogalambos.github.io/Radixor/builds/latest/javadoc/)
+          - [Release verification test report (ciRelease)](https://leogalambos.github.io/Radixor/builds/latest/test/)
+          - [PMD report](https://leogalambos.github.io/Radixor/builds/latest/pmd/main.html)
+          - [JaCoCo coverage report](https://leogalambos.github.io/Radixor/builds/latest/coverage/)
+          - [PIT mutation testing report](https://leogalambos.github.io/Radixor/builds/latest/pitest/)
+          ${DEPENDENCY_CHECK_REPORT_MD}
+          ${SBOM_JSON_REPORT_MD}
+          ${SBOM_XML_REPORT_MD}
+
+          ## Benchmark reports and badge metadata
+
+          ${JMH_TXT_REPORT_MD}
+          ${JMH_CSV_REPORT_MD}
+          ${COVERAGE_BADGE_REPORT_MD}
+          ${MUTATION_BADGE_REPORT_MD}
+
+          ## Historical runs
+
+          - [Browse historical build reports](https://leogalambos.github.io/Radixor/builds/)
           EOF
 
-          BUILD_LIST=$(find "${SITE_DIR}/builds" -mindepth 1 -maxdepth 1 -type d -printf '%f\n' | grep -E '^[0-9]+$' | sort -nr | head -20)
-
-          {
-            cat <<EOF
-          <!doctype html>
-          <html lang="en">
-          <head>
-            <meta charset="utf-8">
-            <meta name="viewport" content="width=device-width, initial-scale=1">
-            <title>Radixor Reports</title>
-            <style>
-              body { font-family: Arial, sans-serif; max-width: 1000px; margin: 2rem auto; padding: 0 1rem; line-height: 1.5; }
-              h1, h2 { margin-bottom: 0.5rem; }
-              ul { padding-left: 1.25rem; }
-              code { background: #f4f4f4; padding: 0.1rem 0.3rem; }
-              .meta { color: #555; }
-            </style>
-          </head>
-          <body>
-            <h1>Radixor Published Reports</h1>
-            <p class="meta">Durable CI reports published from GitHub Actions to the <code>gh-pages</code> branch.</p>
-
-            <h2>Latest</h2>
-            <ul>
-              <li><a href="./builds/latest/">Latest build summary</a></li>
-              <li><a href="./builds/latest/javadoc/">Javadoc</a></li>
-              <li><a href="./builds/latest/test/">Test Report</a></li>
-              <li><a href="./builds/latest/pmd/main.html">PMD Report</a></li>
-              <li><a href="./builds/latest/coverage/">Coverage Report</a></li>
-              ${DEPENDENCY_CHECK_LATEST_LINK:-<li>Dependency Vulnerability Report: not currently available</li>}
-              ${SBOM_JSON_LATEST_LINK:-<li>SBOM (JSON): not available</li>}
-              ${SBOM_XML_LATEST_LINK:-<li>SBOM (XML): not available</li>}
-              <li><a href="./builds/latest/pitest/">Mutation Testing Report</a></li>
-              $(
-                [ "${HAS_JMH}" = "true" ] && { echo "${JMH_TXT_LATEST_LINK:-<li>Benchmark Results (TXT): not available</li>}"; echo "${JMH_CSV_LATEST_LINK:-<li>Benchmark Results (CSV): not available</li>}"; } \
-                || echo '<li>Benchmark results: not currently available</li>'
+          # Retain only the 10 most recent numbered builds to stay within
+          # GitHub Pages capacity limits. The "latest" alias is kept separately.
+          mapfile -t EXPIRED_BUILDS < <(
+            find "${SITE_DIR}/builds" -mindepth 1 -maxdepth 1 -type d -printf '%P\n' \
+              | grep -E '^[0-9]+$' \
+              | sort -r -n \
+              | tail -n +11
           )
-          EOF
 
-            cat <<EOF
-            </ul>
-
-            <h2>Recent historical builds</h2>
-            <ul>
-          EOF
-
-            for build in ${BUILD_LIST}; do
-              echo "      <li><a href=\"./builds/${build}/\">Build ${build}</a></li>"
+          for build in "${EXPIRED_BUILDS[@]}"; do
+            rm -rf "${SITE_DIR}/builds/${build}"
           done
 
-            cat <<EOF
-            </ul>
-          </body>
-          </html>
+          {
+            echo "# Historical Build Reports"
+            echo
+            echo "The following build report sets are currently published on GitHub Pages."
+            echo
+            echo "To stay within GitHub Pages capacity limits, only the 10 most recent build report sets are retained."
+            echo
+            echo "| Build | Published | Link |"
+            echo "|---:|---|---|"
+
+            find "${SITE_DIR}/builds" -mindepth 1 -maxdepth 1 -type d ! -name latest -printf '%P\n' \
+              | grep -E '^[0-9]+$' \
+              | while read -r build; do
+                  ts="$(git -C "${SITE_DIR}" log --diff-filter=A --format='%ct' --reverse -- "builds/${build}/index.html" | head -n 1)"
+                  if [ -n "${ts}" ]; then
+                    published="$(date -u -d "@${ts}" '+%Y-%m-%d %H:%M')"
+                  else
+                    published="unknown"
+                    ts="0"
+                  fi
+                  printf '%s\t%s\t%s\n' "${ts}" "${build}" "${published}"
+                done \
+              | sort -r -n -k1,1 \
+              | while IFS=$'\t' read -r _ts build published; do
+                  echo "| ${build} | ${published} | [Open](../builds/${build}/) |"
+                done
+          } > docs/builds.md
+
+      - name: Build documentation site (MkDocs Material)
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdocs build --strict --site-dir .mkdocs-site
+          rsync -a --delete --exclude '.git' --exclude '.git/' --exclude 'builds/' .mkdocs-site/ .gh-pages/
+          mkdir -p .gh-pages/builds
+          cp .mkdocs-site/builds/index.html .gh-pages/builds/index.html
+          cat > .gh-pages/.nojekyll <<EOF
           EOF
-          } > "${SITE_DIR}/index.html"
+          rm -rf .mkdocs-site
 
       - name: Commit and push gh-pages
         shell: bash

.gitignore

@@ -37,6 +37,7 @@ local.properties
 .settings/
 .loadpath
 .recommenders
+.classpath
 
 # External tool builders
 .externalToolBuilders/
@@ -90,6 +91,9 @@ local.properties
 # PMD plugin conf
 .pmd
 
+# jqwik local db
+.jqwik-database
+
 ##---------------------------------------------------------------------------------------- Gradle
 .gradle
 **/build/

.project

@@ -1,23 +1,22 @@
 <?xml version="1.0" encoding="UTF-8"?>
 <projectDescription>
 	<name>Radixor</name>
-	<comment>Project Radixor created by Buildship.</comment>
-	<projects>
-	</projects>
-	<buildSpec>
-		<buildCommand>
-			<name>org.eclipse.jdt.core.javabuilder</name>
-			<arguments>
-			</arguments>
-		</buildCommand>
-		<buildCommand>
-			<name>org.eclipse.buildship.core.gradleprojectbuilder</name>
-			<arguments>
-			</arguments>
-		</buildCommand>
-	</buildSpec>
+	<comment></comment>
+	<projects/>
 	<natures>
 		<nature>org.eclipse.jdt.core.javanature</nature>
 		<nature>org.eclipse.buildship.core.gradleprojectnature</nature>
 	</natures>
+	<buildSpec>
+		<buildCommand>
+			<name>org.eclipse.jdt.core.javabuilder</name>
+			<arguments/>
+		</buildCommand>
+		<buildCommand>
+			<name>org.eclipse.buildship.core.gradleprojectbuilder</name>
+			<arguments/>
+		</buildCommand>
+	</buildSpec>
+	<linkedResources/>
+	<filteredResources/>
 </projectDescription>

.ruleset

@@ -162,12 +162,12 @@
    <rule ref="category/java/design.xml/CollapsibleIfStatements"/>
    <rule ref="category/java/design.xml/CouplingBetweenObjects">
       <properties>
-         <property name="threshold" value="50" />
+         <property name="threshold" value="70" />
       </properties>
    </rule>
    <rule ref="category/java/design.xml/CyclomaticComplexity">
       <properties>
-         <property name="methodReportLevel" value="18" />
+         <property name="methodReportLevel" value="19" />
       </properties>
    </rule>
    <rule ref="category/java/design.xml/DataClass"/>

LICENSE-stemmer-data

@@ -0,0 +1,29 @@
+Stemmer data licensing
+
+The software source code in this repository is licensed separately under
+the BSD 3-Clause License.
+
+Stemmer dictionary and morphology data files are not covered by
+the BSD 3-Clause License unless explicitly stated otherwise.
+
+This repository contains adapted data derived from the UniMorph project:
+https://unimorph.github.io/
+
+Only stemmer data derived from sources that permit commercial use are included
+in the main distribution of this repository.
+
+Accepted upstream licenses for distributed stemmer data in this repository:
+- CC BY-SA 3.0
+- CC BY-SA 4.0
+- CC BY 4.0
+
+Sources under non-commercial licenses, including CC BY-NC-SA 4.0, are excluded
+from the main distribution.
+
+Modifications in this repository may include cleaning, normalization,
+deduplication, filtering, conversion, and reformatting.
+
+Copyright (c) 2026 Leo Galambos for the modifications, to the extent permitted
+by the applicable upstream license terms.
+
+Per-file licensing is stated in the header of each generated stemmer data file.

README.md

@@ -1,54 +1,83 @@
-<img src="Radixor.png" width="30%" align="right" alt="Radixor logo" />
+<img src="docs/assets/images/banner.jpg" width="100%" alt="Radixor banner" />
 
-# Radixor
+[![License](https://img.shields.io/github/license/leogalambos/Radixor)](LICENSE)
+[![Java](https://img.shields.io/badge/Java-21%2B-brightgreen)](#)
+[![Maven Central](https://img.shields.io/maven-central/v/org.egothor/radixor)](https://central.sonatype.com/artifact/org.egothor/radixor)
+[![Published reports](https://img.shields.io/badge/reports-GitHub%20Pages-blue)](https://leogalambos.github.io/Radixor/builds/latest/)
+[![Quality gates](https://github.com/leogalambos/Radixor/actions/workflows/build.yml/badge.svg?branch=main)](https://github.com/leogalambos/Radixor/actions/workflows/build.yml)
+[![Coverage](https://img.shields.io/endpoint?url=https://leogalambos.github.io/Radixor/builds/latest/metrics/coverage-badge.json)](https://leogalambos.github.io/Radixor/builds/latest/coverage/)
+[![Mutation score](https://img.shields.io/endpoint?url=https://leogalambos.github.io/Radixor/builds/latest/metrics/pitest-badge.json)](https://leogalambos.github.io/Radixor/builds/latest/pitest/)
 
-*Fast algorithmic stemming with compact patch-command tries — measured at about 4× to 6× the throughput of the Snowball Porter stemmer family on the current English benchmark workload.*
+*Deterministic, multi-language stemming for Java, built around compact dictionary-derived patch-command tries with an explicit quality/speed trade-off.*
 
-**Radixor** is a fast, algorithmic stemming toolkit for Java, built around compact **patch-command tries** in the tradition of the original **Egothor** stemmer.
+**Radixor** is a modern multi-language stemming toolkit for Java in the tradition of the original **Egothor** approach. It learns compact word-to-stem transformations from dictionary data, stores them in compiled patch-command tries, and exposes a runtime model designed for speed, determinism, and operational simplicity. Unlike a closed-form dictionary lookup stemmer, Radixor can also generalize beyond explicitly listed word forms.
 
-On the current JMH English comparison benchmark, Radixor with bundled `US_UK_PROFI`
-reaches approximately **31 to 32 million tokens per second**, compared with about
-**8 million tokens per second** for Snowball original Porter and about
-**5 to 5.5 million tokens per second** for Snowball English (Porter2).
+It is particularly well suited to systems that need stemming which is:
 
-That means the current Radixor implementation is approximately:
+- fast at runtime,
+- compact in memory and on disk,
+- deterministic in behavior,
+- adaptable through dictionary data rather than hardcoded language rules,
+- practical to compile, persist, version, extend, and deploy.
 
-- **4× faster** than Snowball original Porter
-- **6× faster** than Snowball English (Porter2)
-
-It is designed for production search and text-processing systems that need stemming which is:
-
-- fast at runtime
-- compact in memory and on disk
-- deterministic in behavior
-- driven by dictionary data rather than hardcoded language rules
-- practical to maintain, extend, and test
-
-Radixor keeps the valuable core of the original Egothor idea, modernizes the implementation, and adds capabilities that make it more useful in real software systems today.
+It also retains the operational advantages of a compiled artifact model: predictable runtime behavior, direct binary loading, and clear separation between preparation-time compilation and live request processing.
 
 ## Table of Contents
 
 - [Why Radixor](#why-radixor)
+- [Performance](#performance)
 - [Heritage](#heritage)
 - [What Radixor adds](#what-radixor-adds)
 - [Key features](#key-features)
-- [Performance](#performance)
 - [Documentation](#documentation)
 - [Project philosophy](#project-philosophy)
 - [Historical note](#historical-note)
 
 ## Why Radixor
 
-The central idea behind Radixor is simple: learn how to transform a word form into its stem, encode that transformation as a compact patch command, store it in a trie, and make runtime lookup extremely fast.
+The central idea behind Radixor is simple: learn how to transform a word form into its stem, encode that transformation as a compact patch command, store it in a trie, and make the runtime path as small and direct as possible.
 
-This gives you a stemmer that is:
+That produces a stemmer that is:
 
-- data-driven rather than rule-hardcoded
-- reusable across languages
-- compact enough for deployment-friendly binary artifacts
-- suitable for both offline compilation and runtime loading
+- data-driven rather than rule-hardcoded,
+- applicable across languages through compiled transformation models learned from dictionary data,
+- compact enough for deployment-friendly binary artifacts,
+- suitable for both offline compilation and direct runtime loading,
+- capable of exposing either a preferred result or multiple candidate results when ambiguity matters.
 
-Radixor is especially attractive when you want something more adaptable than simple suffix stripping, but much smaller and easier to operate than a full morphological analyzer. In the current English benchmark comparison against the Snowball Porter stemmer family, it also delivers a substantial throughput advantage.
+Radixor is especially attractive when you want something more adaptable than simple suffix stripping, but much smaller and easier to operate than a full morphological analyzer.
+
+## Performance
+
+Radixor performance is best read together with stemming quality. The English dictionary coverage benchmark builds contracted compiled patch tries from deterministic slices of the `US_UK` dictionary and then measures both exact-root agreement and changed-token runtime.
+
+| Used rows | Actual row ratio | All exact | Changed exact | Root preserved | Speed ms/op | Error ms | ns/token |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 100.000% | 97.478% | 97.197% | 97.552% | 23.113 | 7.065 | 109.8 |
+| 90% | 90.000% | 97.047% | 94.913% | 97.613% | 21.270 | 9.914 | 101.0 |
+| 80% | 80.000% | 96.635% | 92.768% | 97.661% | 19.170 | 6.609 | 91.1 |
+| 70% | 70.000% | 96.209% | 90.565% | 97.705% | 20.857 | 6.734 | 99.1 |
+| 60% | 60.000% | 95.750% | 88.384% | 97.703% | 14.975 | 1.215 | 71.1 |
+| 50% | 50.000% | 95.262% | 86.107% | 97.690% | 15.249 | 1.078 | 72.4 |
+| 40% | 40.000% | 94.753% | 83.855% | 97.643% | 15.323 | 2.340 | 72.8 |
+| 30% | 30.000% | 94.208% | 81.651% | 97.537% | 16.778 | 2.643 | 79.7 |
+| 20% | 20.000% | 93.633% | 79.366% | 97.416% | 18.929 | 3.241 | 89.9 |
+| 10% | 10.000% | 92.868% | 76.516% | 97.204% | 19.124 | 1.883 | 90.9 |
+
+Column meanings:
+
+- `Used rows` is the requested deterministic percentage of English dictionary rows used to build the stemmer.
+- `Actual row ratio` is the selected row count divided by the full parsed dictionary row count.
+- `All exact` is exact agreement over every word/root pair in the full dictionary.
+- `Changed exact` is exact agreement only where the word differs from its root.
+- `Root preserved` is the share of already-root forms that remain unchanged.
+- `Speed ms/op` is JMH average time for one changed-token benchmark operation.
+- `Error ms` is the JMH score error converted to milliseconds.
+- `ns/token` is average nanoseconds per changed token in that operation.
+
+The contracted trie result is materially stronger than the older uncontracted profile: full English coverage reaches 97.478% all-token exactness and 97.197% changed-token exactness at 109.8 ns/token, while even a 10% deterministic dictionary slice remains at 92.868% all-token exactness and 76.516% changed-token exactness at 90.9 ns/token. This is why Radixor benchmark results are documented with both speed and quality instead of a single Porter speed badge.
+
+For benchmark scope, workload design, environment, commands, report locations, and interpretation guidance, see [Benchmarking](docs/benchmarking.md).
 
 ## Heritage
 
@@ -60,44 +89,50 @@ Useful historical references:
 
 - [Egothor project](http://www.egothor.org/)
 - [Stempel overview](https://www.getopt.org/stempel/)
+- [Leo Galambos, *Lemmatizer for Document Information Retrieval Systems in JAVA* (SOFSEM 2001)](https://www.researchgate.net/publication/221512865_Lemmatizer_for_Document_Information_Retrieval_Systems_in_JAVA)
 - [Lucene Stempel overview](https://lucene.apache.org/core/5_3_0/analyzers-stempel/index.html)
 - [Elasticsearch Stempel plugin](https://www.elastic.co/docs/reference/elasticsearch/plugins/analysis-stempel)
 
-Radixor is not just a repackaging of legacy code. It is a practical modernization of the approach for current Java development and long-term maintainability.
+The Galambos paper is a useful historical reference for the semi-automatic, transformation-based stemming idea that later informed the Egothor lineage and, in turn, the conceptual background of Radixor. It should be read as research and heritage context rather than as a description of Radixor's present-day implementation.
+
+Radixor is not a repackaging of legacy code. It is a modern implementation that preserves the valuable core idea while reworking the engineering around maintainability, testing, persistence, and long-term operational use.
 
 ## What Radixor adds
 
-Radixor keeps the patch-command trie model, but improves the engineering around it.
+Radixor keeps the patch-command trie model, but improves the engineering around it in ways that matter in real software systems.
 
 Compared with the historical baseline, Radixor emphasizes:
 
-- **simplification to the most practical core**  
-  The implementation focuses on the parts of the original approach that are most useful in production.
+- **a focused practical core**  
+  The implementation concentrates on the parts of the original approach that are most useful in production.
 
 - **immutable compiled tries**  
   Runtime lookup uses compact read-only structures optimized for efficient access.
 
 - **support for more than one stemming result**  
-  Radixor can expose both a preferred result and multiple candidate results where the data is ambiguous.
+  Radixor can expose both a preferred result and multiple candidate results when the underlying data is ambiguous.
 
 - **frequency-aware deterministic ordering**  
   Candidate results are ordered consistently and reproducibly.
 
-- **practical subtree reduction modes**  
-  Reduction can be tuned toward stronger compression or more conservative behavioral preservation.
+- **contracted compiled patch tries**  
+  Uniform patch-command subtrees are collapsed into accepting leaves, reducing hot lookup depth while preserving preferred stemming results.
 
-- **reconstruction of writable builders from compiled tables**  
+- **practical subtree reduction modes**  
+  Reduction can be tuned toward stronger compression or more conservative semantic preservation.
+
+- **reconstruction of writable builders from compiled artifacts**  
   Existing compiled stemmer tables can be reopened, modified, and compiled again.
 
-- **better tests and implementation stability**  
-  Stronger coverage improves confidence during refactoring and further development.
+- **strong validation discipline**  
+  Coverage, mutation testing, benchmark visibility, and published reports are treated as part of the engineering standard rather than optional project decoration.
 
 ## Key features
 
 - Fast algorithmic stemming
 - Compact compiled binary artifacts
 - Patch-command based transformation model
-- Dictionary-driven language adaptation
+- Multi-language stemming through compiled transformation models
 - Single-result and multi-result lookup
 - Deterministic result ordering
 - Compressed binary persistence
@@ -105,57 +140,84 @@ Compared with the historical baseline, Radixor emphasizes:
 - CLI compilation tool
 - Bundled language resources
 - Support for extending compiled stemmer tables
-
-## Performance
-
-Radixor includes a JMH benchmark suite for both its own algorithmic core and a
-side-by-side comparison against the Snowball Porter stemmer family.
-
-On the current English comparison workload, Radixor with bundled `US_UK_PROFI`
-reaches approximately **31 to 32 million tokens per second**. Snowball original
-Porter reaches approximately **8 million tokens per second**, and Snowball
-English (Porter2) approximately **5 to 5.5 million tokens per second**.
-
-That places Radixor at approximately **4× the throughput of Snowball original Porter**
-and approximately **6× the throughput of Snowball English (Porter2)**
-on the current benchmark workload.
-
-This is a throughput comparison on the same deterministic token stream. It is
-not a claim that the compared stemmers are linguistically equivalent or
-interchangeable.
-
-For benchmark scope, workload design, environment, commands, report locations,
-and interpretation guidance, see [Benchmarking](docs/benchmarking.md).
+- Reproducible and auditable engineering posture
 
 ## Documentation
 
 The repository keeps the front page concise and places detailed documentation under `docs/`.
 
-Start here:
+### Getting Started
+
+- [Fast Track](docs/fast-track.md)  
+  The shortest path from adding the dependency to getting a first stem from a bundled dictionary.
 
 - [Quick Start](docs/quick-start.md)  
-  A practical first guide to loading, compiling, and using Radixor.
+  A broader developer walkthrough covering loading options, querying, extension, persistence, and metadata.
 
-- [Dictionary Format](docs/dictionary-format.md)  
-  How to write stemming dictionaries.
-
-- [Compilation (CLI tool)](docs/cli-compilation.md)  
-  How to compile dictionaries with the `Compile` CLI.
-
-- [Programmatic Usage](docs/programmatic-usage.md)  
-  How to build, load, modify, and query Radixor from Java code.
+- [Integration Deep Dive](docs/integration-deep-dive.md)  
+  Dependency setup, bundled dictionary selection, production lifecycle, search-pipeline guidance, and operational checklist.
 
 - [Built-in Languages](docs/built-in-languages.md)  
-  How to use integrated language resources such as `US_UK_PROFI`.
+  Overview of bundled language resources such as `US_UK`.
 
-- [Architecture and Reduction](docs/architecture-and-reduction.md)  
-  Internal model, compiled trie design, and reduction strategies.
+- [Dictionary Format](docs/dictionary-format.md)  
+  How to write and normalize stemming dictionaries.
+
+- [Compilation (CLI tool)](docs/cli-compilation.md)  
+  How to compile dictionaries into deployable binary artifacts.
+
+### Programmatic Usage
+
+- [Programmatic Usage Overview](docs/programmatic-usage.md)  
+  Entry point to the Java API and the overall usage model.
+
+- [Loading and Building Stemmers](docs/programmatic-loading-and-building.md)  
+  Loading bundled resources, textual dictionaries, binary artifacts, and direct builder usage.
+
+- [Querying and Ambiguity Handling](docs/programmatic-querying-and-ambiguity.md)  
+  `get()`, `getAll()`, `getEntries()`, patch application, and ambiguity behavior.
+
+- [Extending and Persisting Compiled Tries](docs/programmatic-extending-and-persistence.md)  
+  Reopening compiled tries, rebuilding them, and writing binary artifacts.
+
+- [Migration and Backward Compatibility](docs/migration-and-backward-compatibility.md)  
+  Migration from serialized String patch-command application to `CompiledPatchCommand`.
+
+### Concepts and Internals
+
+- [Architecture and Reduction Overview](docs/architecture-and-reduction.md)  
+  High-level explanation of the build pipeline and compiled trie model.
+
+- [Architecture](docs/architecture.md)  
+  Structural model, data flow, and runtime lookup behavior.
+
+- [Lookup Edge Optimization](docs/lookup-edge-optimization.md)  
+  Speed/memory trade-off of dense child edge lookup in compiled tries.
+
+- [Reduction Semantics](docs/reduction-semantics.md)  
+  Ranked, unordered, and dominant reduction behavior.
+
+- [Compatibility and Guarantees](docs/compatibility-and-guarantees.md)  
+  Supported public API, internal API boundaries, and compatibility expectations.
+
+### Dictionaries and Language Resources
+
+- [Contributing Dictionaries](docs/contributing-dictionaries.md)  
+  Guidance for high-quality lexical resource contributions.
+
+### Quality and Operations
 
 - [Quality and Operations](docs/quality-and-operations.md)  
-  Testing, persistence, deployment, and operational guidance.
+  Engineering standards, validation posture, auditability, and operational model.
 
 - [Benchmarking](docs/benchmarking.md)  
-  JMH benchmark design, Snowball comparison, execution, and interpretation.
+  JMH benchmark methodology, dictionary coverage trade-offs, speed, quality, and result interpretation.
+
+- [Benchmark Results](docs/benchmarks/index.md)  
+  Structured reference for methodology, corpora, environment, English coverage, and per-language result pages.
+
+- [Published Reports](docs/reports.md)  
+  Entry points to CI-published reports and GitHub Pages artifacts.
 
 ## Project philosophy
 
@@ -163,19 +225,20 @@ Radixor does not preserve historical complexity for its own sake.
 
 It preserves the valuable idea:
 
-- compact learned transformations
-- trie-based lookup
-- language-data driven stemming
-- practical runtime speed
+- compact learned transformations,
+- trie-based lookup,
+- language-data driven stemming,
+- practical runtime speed.
 
 Then it improves the parts modern users care about:
 
-- maintainability
-- testability
-- modification workflows
-- persistence
-- determinism
-- clearer APIs
+- maintainability,
+- testability,
+- modification workflows,
+- persistence,
+- determinism,
+- clearer APIs,
+- explicit quality evidence.
 
 The goal is to keep the Egothor/Stempel lineage useful as a serious contemporary software component.
 

build.gradle

@@ -18,6 +18,8 @@ version = gitVersion(prefix:'release@')
 
 def benchmarkReportsDirectory = layout.buildDirectory.dir('reports/jmh')
 def sbomReportsDirectory = layout.buildDirectory.dir('reports/sbom')
+def jmhIncludesProperty = providers.gradleProperty('jmh.includes')
+        .orElse(providers.systemProperty('jmh.includes'))
 
 def nvdApiKey = providers.gradleProperty('nvdApiKey')
         .orElse(providers.environmentVariable('NVD_API_KEY'))
@@ -28,11 +30,19 @@ apply from: 'gradle/maven-pom.gradle'
 
 configurations {
     mockitoAgent
+    stemmingQualityJmhRuntime {
+        canBeConsumed = false
+        canBeResolved = true
+        extendsFrom(jmhImplementation, jmhRuntimeOnly)
+    }
 }
 
 java {
     withSourcesJar()
     withJavadocJar()
+
+    sourceCompatibility = JavaVersion.VERSION_21
+    targetCompatibility = JavaVersion.VERSION_21
 }
 
 tasks.withType(AbstractArchiveTask).configureEach {
@@ -51,10 +61,6 @@ pmd {
     ruleSetFiles = files(rootProject.file(".ruleset"))
 }
 
-tasks.withType(JavaCompile).configureEach {
-    options.release = 21
-}
-
 dependencyLocking {
     lockAllConfigurations()
 
@@ -70,12 +76,23 @@ dependencies {
 
     testImplementation libs.mockito.core
     testImplementation libs.mockito.junit.jupiter
+    testImplementation libs.jqwik
 
     mockitoAgent(libs.mockito.core) {
         transitive = false
     }
 }
 
+sourceSets.jmh.compileClasspath = sourceSets.jmh.compileClasspath - sourceSets.test.output
+sourceSets.jmh.runtimeClasspath = sourceSets.jmh.runtimeClasspath - sourceSets.test.output
+sourceSets.test.compileClasspath += sourceSets.jmh.output + configurations.jmhCompileClasspath
+sourceSets.test.runtimeClasspath += sourceSets.jmh.output + configurations.jmhCompileClasspath
+
+tasks.named('compileJmhJava', JavaCompile) {
+    classpath = classpath - sourceSets.test.output
+    setDependsOn([tasks.named('classes')])
+}
+
 dependencyCheck {
     failBuildOnCVSS = 7.0
     failOnError = true
@@ -108,14 +125,30 @@ dependencyCheck {
     }
 }
 
-tasks.withType(Test).configureEach {
-    useJUnitPlatform()
+def cliIncludeTags = project.findProperty('includeTags')?.toString() ?: System.getProperty('includeTags')
+def cliExcludeTags = project.findProperty('excludeTags')?.toString() ?: System.getProperty('excludeTags')
 
+def splitTagExpression = { String tagsExpr ->
+    if (tagsExpr == null || tagsExpr.isBlank()) {
+        return []
+    }
+    return tagsExpr.split(',')
+            .collect { it.trim() }
+            .findAll { it != null && !it.isBlank() }
+}
+
+tasks.withType(Test).configureEach {
     doFirst {
         jvmArgs "-javaagent:${configurations.mockitoAgent.singleFile}"
     }
 
-    finalizedBy(tasks.named('jacocoTestReport'))
+    /*
+     * Bundled dictionary integration tests compile and reload large real-world
+     * stemming dictionaries, including large language resources such as es_es.
+     * The default Gradle test executor heap is too small for this workload.
+     */
+    minHeapSize = '1g'
+    maxHeapSize = '4g'
 
     reports {
         junitXml.required = true
@@ -123,6 +156,124 @@ tasks.withType(Test).configureEach {
     }
 }
 
+def configureJUnitPlatformTags = { Test task, String includeTagsExpr, String excludeTagsExpr ->
+    task.useJUnitPlatform {
+        final def includes = splitTagExpression(includeTagsExpr)
+        final def excludes = splitTagExpression(excludeTagsExpr)
+
+        if (!includes.isEmpty()) {
+            includeTags(*includes.toArray(new String[0]))
+        }
+        if (!excludes.isEmpty()) {
+            excludeTags(*excludes.toArray(new String[0]))
+        }
+    }
+}
+
+tasks.named('test', Test) {
+    final def requestedIncludes = splitTagExpression(cliIncludeTags)
+    final boolean slowExplicitlyIncluded = requestedIncludes.contains('slow')
+    final String defaultExcludeTags = cliExcludeTags ?: (slowExplicitlyIncluded ? null : 'slow')
+    configureJUnitPlatformTags(it, cliIncludeTags, defaultExcludeTags)
+    finalizedBy(tasks.named('jacocoTestReport'))
+}
+
+def configureTaggedTestProfile = { String taskName, String includeTagsExpr, String excludeTagsExpr = null,
+                                   String taskDescription = null, String testNameExcludePatterns = null ->
+    tasks.register(taskName, Test) {
+        group = 'verification'
+        description = taskDescription
+
+        configureJUnitPlatformTags(delegate as Test, includeTagsExpr, excludeTagsExpr)
+        testClassesDirs = sourceSets.test.output.classesDirs
+        classpath = sourceSets.test.runtimeClasspath
+        dependsOn(tasks.named('compileTestJava'))
+
+        doFirst {
+            jvmArgs "-javaagent:${configurations.mockitoAgent.singleFile}"
+        }
+
+        if (testNameExcludePatterns != null && !testNameExcludePatterns.isBlank()) {
+            filter {
+                testNameExcludePatterns.split(',').each { String pattern ->
+                    final def trimmedPattern = pattern.trim()
+                    if (!trimmedPattern.isEmpty()) {
+                        excludeTestsMatching(trimmedPattern)
+                    }
+                }
+            }
+        }
+
+        minHeapSize = '1g'
+        maxHeapSize = '4g'
+
+        reports {
+            junitXml.required = true
+            html.required = true
+        }
+    }
+}
+
+configureTaggedTestProfile(
+        'ciSmoke',
+        'unit',
+        'slow',
+        'Fast feedback profile for unit tests with slow tests explicitly excluded.',
+        'org.egothor.stemmer.CompileIntegrationTest*'
+)
+
+configureTaggedTestProfile(
+        'ciCore',
+        'unit,trie,frequency-trie,property',
+        null,
+        'Focused profile for core trie behavior and trie-specific property checks.'
+)
+
+configureTaggedTestProfile(
+        'ciIntegration',
+        'integration',
+        'slow',
+        'Integration pipeline profile (loader/parser/CLI/IO end-to-end flows) excluding slow integration paths.'
+)
+
+configureTaggedTestProfile(
+        'ciSlow',
+        'slow',
+        null,
+        'Targeted profile for all slow tests (large dictionaries, long-running corpus validation, and heavy integration checks).'
+)
+
+configureTaggedTestProfile(
+        'ciCompat',
+        'compat,regression',
+        null,
+        'Compatibility profile guarding persisted artifact and compatibility regressions.'
+)
+
+configureTaggedTestProfile(
+        'ciRelease',
+        null,
+        'slow',
+        'Release-profile validation of all non-slow tests.',
+        'org.egothor.stemmer.CompileIntegrationTest*,org.egothor.stemmer.StemmerPatchTrieLoaderTest$BundledDictionaryTests*'
+)
+
+configureTaggedTestProfile(
+        'ciNightly',
+        'fuzz',
+        null,
+        'Nightly robustness profile with fuzz testing emphasis.'
+)
+
+tasks.register('ci') {
+    group = 'verification'
+    description = 'Runs the full enterprise CI profile set in sequence.'
+    dependsOn(tasks.named('ciSmoke'))
+    dependsOn(tasks.named('ciCore'))
+    dependsOn(tasks.named('ciIntegration'))
+    dependsOn(tasks.named('ciCompat'))
+}
+
 tasks.withType(Pmd).configureEach {
     reports {
         xml.required = true
@@ -133,6 +284,13 @@ tasks.withType(Pmd).configureEach {
 tasks.named('jacocoTestReport', JacocoReport) {
     dependsOn(tasks.named('test'))
 
+    classDirectories.setFrom(
+            files(sourceSets.main.output).asFileTree.matching {
+                exclude 'org/egothor/stemmer/StemmerKnowledgeExperiment*'
+                exclude 'org/egothor/stemmer/DiacriticStripper*'
+            }
+    )
+
     reports {
         xml.required = true
         csv.required = false
@@ -140,6 +298,36 @@ tasks.named('jacocoTestReport', JacocoReport) {
     }
 }
 
+def registerJacocoProfileReport = { String reportTaskName, String sourceTaskName ->
+    tasks.register(reportTaskName, JacocoReport) {
+        group = 'verification'
+        description = "Generates Jacoco report for ${sourceTaskName} execution."
+
+        dependsOn(tasks.named(sourceTaskName))
+
+        classDirectories.setFrom(
+                files(sourceSets.main.output).asFileTree.matching {
+                    exclude 'org/egothor/stemmer/StemmerKnowledgeExperiment*'
+                    exclude 'org/egothor/stemmer/DiacriticStripper*'
+                }
+        )
+
+        executionData.setFrom(
+                fileTree(layout.buildDirectory.dir('jacoco')) {
+                    include "${sourceTaskName}.exec"
+                }
+        )
+
+        reports {
+            xml.required = true
+            csv.required = false
+            html.required = true
+        }
+    }
+}
+
+registerJacocoProfileReport('jacocoCiReleaseReport', 'ciRelease')
+
 tasks.named('check') {
     dependsOn(tasks.named('jacocoTestReport'))
     // no-default, only on-demand: dependsOn(tasks.named('dependencyCheckAnalyze'))
@@ -177,7 +365,17 @@ pitest {
         'org.egothor.stemmer.trie.*Test'
     ]
 
-    excludedClasses = ['org.egothor.stemmer.Compile']
+    excludedClasses = [
+        'org.egothor.stemmer.Compile*',
+        'org.egothor.stemmer.StemmerPatchTrieLoader*',
+        'org.egothor.stemmer.StemmerKnowledgeExperiment*',
+        'org.egothor.stemmer.StemmerKnowledgeExperimentCli*'
+    ]
+    excludedTestClasses = [
+        'org.egothor.stemmer.CompileIntegrationTest',
+        'org.egothor.stemmer.StemmerPatchTrieLoaderTest',
+        'org.egothor.stemmer.StemmerKnowledgeExperimentTest'
+    ]
     outputFormats = ['XML', 'HTML']
     timestampedReports = false
     exportLineCoverage = true
@@ -187,10 +385,63 @@ pitest {
 
 application {
     mainClass = 'org.egothor.stemmer.Compile'
+    applicationName = 'radixor'
+    executableDir = 'bin'
+}
+
+tasks.register('stemmerKnowledgeExperiment', JavaExec) {
+    group = 'application'
+    description = 'Runs the stemmer knowledge evaluation experiment.'
+    classpath = sourceSets.main.runtimeClasspath
+    mainClass = 'org.egothor.stemmer.StemmerKnowledgeExperimentCli'
+}
+
+distributions {
+    main {
+        distributionBaseName = 'radixor'
+
+        contents {
+            from('README.md') {
+                into ''
+            }
+
+            from('LICENSE') {
+                into ''
+            }
+
+            from('LICENSE-stemmer-data') {
+                into ''
+            }
+
+            from('docs') {
+                into 'docs'
+                include '**/*.md'
+            }
+
+            from(layout.buildDirectory.dir('generated/release-notes')) {
+                into ''
+                include 'CHANGELOG.md'
+            }
+        }
+    }
+}
+
+tasks.named('startScripts') {
+    applicationName = 'radixor'
+}
+
+tasks.named('distZip', Zip) {
+    archiveBaseName = 'radixor'
+    archiveClassifier = 'bin'
+}
+
+tasks.named('distTar') {
+    enabled = false
 }
 
 jmh {
     jmhVersion = '1.37'
+    includeTests = false
     warmupIterations = 3
     iterations = 5
     fork = 1
@@ -200,13 +451,18 @@ jmh {
     resultsFile = benchmarkReportsDirectory.map { it.file('jmh-results.csv').asFile }.get()
     humanOutputFile = benchmarkReportsDirectory.map { it.file('jmh-results.txt').asFile }.get()
     duplicateClassesStrategy = DuplicatesStrategy.EXCLUDE
+    if (jmhIncludesProperty.isPresent()) {
+        includes = [jmhIncludesProperty.get()]
+    }
 }
 
 tasks.named('jmh') {
     group = 'verification'
-    description = 'Runs JMH benchmarks for the Radixor algorithmic core and Snowball comparison suite.'
+    description = 'Runs JMH benchmarks for the Radixor algorithmic core and external stemmer comparison suites.'
 }
 
+apply from: 'gradle/lucene-benchmarks.gradle'
+
 tasks.register('regressionArtifactGenerator', JavaExec) {
     group = 'verification'
     description = 'Generates deterministic compiled trie regression artifacts.'
@@ -228,6 +484,85 @@ tasks.register('regressionArtifactGenerator', JavaExec) {
     }
 }
 
+tasks.register('stemmingQuality', JavaExec) {
+    group = 'verification'
+    description = 'Evaluates pairwise over-stemming and under-stemming against bundled dictionary groups.'
+    dependsOn(tasks.named('testClasses'))
+    dependsOn(tasks.named('jmhClasses'))
+    classpath = files(sourceSets.test.runtimeClasspath, configurations.stemmingQualityJmhRuntime)
+    mainClass = 'org.egothor.stemmer.benchmark.quality.StemmingQualityApplication'
+    args layout.buildDirectory.dir('reports/stemming-quality').get().asFile.absolutePath,
+            layout.projectDirectory.dir('src/main/resources').asFile.absolutePath,
+            providers.gradleProperty('stemmingQualityLanguage').getOrElse(''),
+            providers.gradleProperty('stemmingQualityStemmer').getOrElse(''),
+            providers.gradleProperty('stemmingQualityMode').getOrElse(''),
+            providers.gradleProperty('stemmingQualityOutputPolicy').getOrElse(''),
+            providers.gradleProperty('stemmingQualityRankMetric').getOrElse('PAIRWISE_F05'),
+            providers.gradleProperty('stemmingQualityAudit').getOrElse('false'),
+            providers.gradleProperty('stemmingQualityAuditLimit').getOrElse('25')
+    maxHeapSize = '6g'
+}
+
+tasks.register('publishStemmingQualityDocumentation', JavaExec) {
+    group = 'documentation'
+    description = 'Publishes validated complete stemming-quality results on the language benchmark pages.'
+    dependsOn(tasks.named('testClasses'))
+    classpath = sourceSets.test.runtimeClasspath
+    mainClass = 'org.egothor.stemmer.benchmark.quality.StemmingQualityDocumentationPublisher'
+    args layout.buildDirectory.file('reports/stemming-quality/stemming-quality.csv').get().asFile.absolutePath,
+            layout.projectDirectory.dir('docs').asFile.absolutePath,
+            'update'
+    doFirst {
+        if (!file("$buildDir/reports/stemming-quality/stemming-quality.csv").isFile()) {
+            throw new GradleException('A complete stemming-quality CSV is required. Run stemmingQuality only when no validated complete report is available.')
+        }
+    }
+}
+
+tasks.register('verifyStemmingQualityDocumentation', JavaExec) {
+    group = 'verification'
+    description = 'Verifies published language-page quality tables against the checked-in authoritative CSV.'
+    dependsOn(tasks.named('testClasses'))
+    classpath = sourceSets.test.runtimeClasspath
+    mainClass = 'org.egothor.stemmer.benchmark.quality.StemmingQualityDocumentationPublisher'
+    args layout.projectDirectory.file('docs/benchmarks/data/stemming-quality.csv').asFile.absolutePath,
+            layout.projectDirectory.dir('docs').asFile.absolutePath,
+            'verify'
+}
+
+tasks.named('check') {
+    dependsOn(tasks.named('verifyStemmingQualityDocumentation'))
+}
+
+tasks.register('verifyStemmingQualitySourceSets') {
+    group = 'verification'
+    description = 'Verifies the production, JMH, and standard-test ownership of stemming-quality infrastructure.'
+    doLast {
+        if (sourceSets.findByName('stemmingQualityTest') != null || file('src/stemmingQualityTest').exists()) {
+            throw new GradleException('The obsolete stemmingQualityTest source set or directory still exists.')
+        }
+        if (!file('src/jmh/java/org/egothor/stemmer/benchmark/QualityStemmerMatrix.java').isFile()) {
+            throw new GradleException('The authoritative JMH stemmer matrix is not in src/jmh.')
+        }
+        if (!file('src/test/java/org/egothor/stemmer/benchmark/quality/StemmingQualityApplication.java').isFile()) {
+            throw new GradleException('The stemming-quality evaluator is not in the standard test source set.')
+        }
+    }
+}
+
+tasks.register('verifyProductionJarExcludesStemmingQuality') {
+    group = 'verification'
+    description = 'Verifies that analytical stemming-quality classes are absent from the production JAR.'
+    dependsOn(tasks.named('jar'))
+    doLast {
+        final File archive = tasks.named('jar').get().archiveFile.get().asFile
+        final def forbidden = zipTree(archive).matching { include '**/benchmark/**' }.files
+        if (!forbidden.isEmpty()) {
+            throw new GradleException("Production JAR contains analytical stemming-quality classes: ${forbidden}")
+        }
+    }
+}
+
 tasks.register('printDependencyCheckNvdConfig') {
     doLast {
         System.out.println("NVD API key present: " + (nvdApiKey != null && !nvdApiKey.isBlank()))
@@ -260,11 +595,26 @@ javadoc {
     options.version = true
     options.windowTitle = 'Radixor - Egothor Stemmer'
     options.docTitle = 'Radixor - Egothor Stemmer API'
+    options.overview = file('src/main/javadoc/overview.html')
+    options.bottom = """
+        <div class="legal-copy">
+            &copy; 2026 Egothor
+            <br/>
+            Licensed under <a href="https://github.com/leogalambos/Radixor/blob/main/LICENSE">BSD-3-Clause</a>
+        </div>
+    """
+    options.links('https://docs.oracle.com/en/java/javase/21/docs/api/')
+    options.group('Core Stemming API', 'org.egothor.stemmer')
+    options.group('Trie Infrastructure', 'org.egothor.stemmer.trie')
 
     source = sourceSets.main.allJava
 }
 
 apply from: 'gradle/snowball-benchmarks.gradle'
+apply from: 'gradle/paicehusk-benchmarks.gradle'
+apply from: 'gradle/opennlp-benchmarks.gradle'
+apply from: 'gradle/hunspell-benchmarks.gradle'
+apply from: 'gradle/cistem-benchmarks.gradle'
 
 gradle.taskGraph.whenReady { taskGraph ->
     def banner = """

docs/architecture-and-reduction.md

@@ -1,470 +1,52 @@
 # Architecture and Reduction
 
-> ← Back to [README.md](../README.md)
+This section explains how **Radixor** turns textual dictionary input into a compact compiled stemmer and how reduction affects the semantics preserved in the final runtime artifact.
 
-This document describes the internal architecture of **Radixor** and the principles behind its **trie compilation and reduction model**.
+Radixor is easiest to understand when separated into two related concerns:
 
-It explains:
+- **architecture**: what structures exist, how data moves through them, and what runtime lookup actually does,
+- **reduction semantics**: what it means for two subtrees to be considered equivalent and how that choice affects `get()` and `getAll()` behavior.
 
-- how data flows from dictionary input to compiled trie
-- how patch-command tries are structured
-- how subtree reduction works
-- how reduction modes affect behavior and size
+## The short version
 
+Radixor does not keep a large flat table of final stems. Instead, it converts dictionary entries into **patch commands**, stores them in a trie, reduces equivalent subtrees, and freezes the result into an immutable compiled structure.
 
+The build-time flow is:
 
-## Overview
-
-Radixor transforms dictionary data into an optimized runtime structure through three stages:
-
-1. **Mutable construction**
-2. **Reduction (canonicalization)**
-3. **Compilation (freezing)**
-
-```
-Dictionary → Mutable trie → Reduced trie → Compiled trie
+```text
+Dictionary -> Mutable trie -> Reduced trie -> Compiled trie
 ```
 
-Each stage has a distinct purpose:
+At runtime, the compiled trie does not directly return the final stem string. It returns one or more stored patch commands for the addressed key, and those commands are then applied to the original input word.
 
-| Stage       | Purpose                         | Structure               |
-|------------|----------------------------------|-------------------------|
-| Build       | Collect mappings                 | `MutableNode`           |
-| Reduction   | Merge equivalent subtrees        | `ReducedNode`           |
-| Compilation | Optimize for runtime lookup      | `CompiledNode`          |
+## Why this matters
 
+This design gives Radixor several practical properties at once:
 
+- compact deployable artifacts,
+- deterministic runtime behavior,
+- support for both preferred and multiple candidate results,
+- separation of preparation-time complexity from runtime lookup.
 
-## Core data model
+It also explains why a large source dictionary can be transformed into a much smaller compiled artifact without discarding the operational behavior that matters to the caller.
 
-### Patch-command trie
+## Reading guide
 
-Radixor stores **patch commands** instead of stems directly.
+Use the following pages depending on what you need to understand:
 
-- keys: word forms
-- values: transformation commands
-- structure: trie (prefix tree)
+- [Architecture](architecture.md) explains the data flow, core structures, patch-command lookup model, and why the compiled trie is efficient at runtime.
+- [Reduction Semantics](reduction-semantics.md) explains how subtree equivalence is defined, what ranked, unordered, and dominant reduction preserve, and how those choices affect observable lookup behavior.
 
-At runtime:
+## Recommended reading order
 
-1. the word is traversed through the trie
-2. a patch command is retrieved
-3. the patch is applied to reconstruct the stem
+For most readers, the best order is:
 
+1. [Architecture](architecture.md)
+2. [Reduction Semantics](reduction-semantics.md)
 
+## Related documentation
 
-## Stage 1: Mutable construction
-
-The builder (`FrequencyTrie.Builder`) constructs a trie using:
-
-- `MutableNode`
-- maps of children (`char → node`)
-- maps of value counts (`value → frequency`)
-
-Characteristics:
-
-- insertion-order preserving
-- mutable
-- optimized for building, not querying
-
-Example structure:
-
-```
-g
- └─ n
-     └─ i
-         └─ n
-             └─ n
-                 └─ u
-                     └─ r
-                         └─ (values: {
-                               "<patch-command-1>": 3,
-                               "<patch-command-2>": 1
-                           })
-```
-
-This example represents the word "running", stored in reversed form.
-
-- each edge corresponds to one character of the word
-- the path is traversed from the end of the word toward the beginning
-- the terminal node stores one or more patch commands together with their local frequencies
-
-The values represent transformations from the word form to candidate stems, and the counts indicate how often each mapping was observed during construction.
-
-Note: Radixor stores word forms in reversed order so that suffix-based transformations can be matched efficiently in a trie.
-
-
-## Local value summary
-
-Before reduction, each node is summarized using `LocalValueSummary`.
-
-It computes:
-
-- ordered values (by frequency)
-- aligned counts
-- total frequency
-- dominant value (if any)
-- second-best value
-
-This summary is critical for:
-
-- deterministic ordering
-- reduction decisions
-- dominance evaluation
-
-
-
-## Stage 2: Reduction (canonicalization)
-
-Reduction is the process of merging **semantically equivalent subtrees**.
-
-### Why reduction exists
-
-Without reduction:
-
-- trie size grows linearly with input data
-- repeated patterns are duplicated
-
-With reduction:
-
-- identical subtrees are shared
-- memory footprint is reduced
-- binary output becomes smaller
-
-
-
-## Reduction signature
-
-Each subtree is represented by a **ReductionSignature**.
-
-A signature consists of:
-
-1. **local descriptor** (node semantics)
-2. **child descriptors** (structure)
-
-```
-Signature = (LocalDescriptor, SortedChildDescriptors)
-```
-
-Two subtrees are merged if their signatures are equal.
-
-
-
-## Local descriptors
-
-The local descriptor encodes how values at a node are interpreted.
-
-Radixor supports three descriptor types:
-
-### 1. Ranked descriptor
-
-Preserves:
-
-- full ordering of values (`getAll()`)
-
-Uses:
-
-- ordered value list
-
-Best for:
-
-- correctness
-- deterministic multi-result behavior
-
-
-
-### 2. Unordered descriptor
-
-Preserves:
-
-- only membership (set of values)
-
-Ignores:
-
-- ordering differences
-
-Best for:
-
-- higher compression
-- use cases where ordering is irrelevant
-
-
-
-### 3. Dominant descriptor
-
-Preserves:
-
-- only the dominant value (`get()`)
-
-Condition:
-
-- dominant value must satisfy thresholds:
-  - minimum percentage
-  - ratio over second-best
-
-Fallback:
-
-- if dominance is not strong enough → ranked descriptor is used
-
-Best for:
-
-- maximum compression
-- single-result workflows
-
-
-
-## Child descriptors
-
-Each child is represented as:
-
-```
-(edge character, child signature)
-```
-
-Children are sorted by edge character to ensure:
-
-- deterministic signatures
-- stable equality comparisons
-
-
-
-## Reduction context
-
-`ReductionContext` maintains:
-
-- mapping: `ReductionSignature → ReducedNode`
-- canonical instances of subtrees
-
-Workflow:
-
-1. compute signature
-2. check if already exists
-3. reuse existing node or create new one
-
-This ensures:
-
-- structural sharing
-- no duplicate equivalent subtrees
-
-
-
-## Reduced nodes
-
-`ReducedNode` represents:
-
-- canonical subtree
-- aggregated value counts
-- canonical children
-
-It supports:
-
-- merging local counts
-- verifying structural consistency
-
-At this stage:
-
-- structure is canonical
-- still mutable (internally)
-
-
-
-## Stage 3: Compilation (freezing)
-
-The reduced trie is converted into a **CompiledNode** structure.
-
-### CompiledNode characteristics
-
-- immutable
-- array-based storage
-- optimized for fast lookup
-
-Fields:
-
-- `char[] edgeLabels`
-- `CompiledNode[] children`
-- `V[] orderedValues`
-- `int[] orderedCounts`
-
-
-
-## Lookup algorithm
-
-Runtime lookup:
-
-1. traverse trie using `edgeLabels` (matching characters from the end of the word toward the beginning)
-2. binary search per node
-3. retrieve values
-4. apply patch command
-
-Properties:
-
-- O(length of word)
-- low memory overhead
-- minimal memory allocation during lookup; patch application produces the resulting string
-
-
-## Deterministic ordering
-
-Value ordering is deterministic and stable:
-
-1. higher frequency first
-2. shorter string first
-3. lexicographically smaller
-4. insertion order
-
-This guarantees:
-
-- reproducible builds
-- stable query results
-- predictable ranking
-
-
-
-## Reduction modes
-
-Reduction modes control how local descriptors are chosen.
-
-### Ranked mode
-
-```
-MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
-```
-
-- preserves full semantics
-- safest option
-- recommended default
-
-
-
-### Unordered mode
-
-```
-MERGE_SUBTREES_WITH_EQUIVALENT_UNORDERED_GET_ALL_RESULTS
-```
-
-- ignores ordering
-- higher compression
-- slightly weaker semantics
-
-
-
-### Dominant mode
-
-```
-MERGE_SUBTREES_WITH_EQUIVALENT_DOMINANT_GET_RESULTS
-```
-
-- keeps only dominant result
-- highest compression
-- may lose alternative candidates
-
-
-
-## Trade-offs
-
-| Aspect        | Ranked | Unordered | Dominant |
-|---------------|--------|----------|----------|
-| Compression   | Medium | High     | Highest  |
-| Accuracy      | High   | Medium   | Lower    |
-| getAll()      | Full   | Partial  | Limited  |
-| get()         | Exact  | Exact    | Heuristic|
-
-
-
-## Deserialization model
-
-Binary loading uses:
-
-- `NodeData` as intermediate representation
-- reconstruction of `CompiledNode`
-
-This separates:
-
-- I/O format
-- in-memory structure
-
-
-
-## Why this architecture works
-
-Radixor achieves:
-
-### Compactness
-
-- subtree sharing
-- efficient encoding
-- compressed binary output
-
-### Performance
-
-- array-based lookup
-- no runtime reduction
-- minimal branching
-
-### Flexibility
-
-- configurable reduction strategies
-- multiple result support
-- dictionary-driven behavior
-
-### Determinism
-
-- stable ordering
-- canonical signatures
-- reproducible builds
-
-
-
-## Design philosophy
-
-The architecture reflects a few key principles:
-
-- separate build-time complexity from runtime simplicity
-- encode semantics explicitly (not implicitly in code)
-- favor deterministic behavior over heuristic shortcuts
-- allow controlled trade-offs between size and fidelity
-
-
-
-## When to tune reduction
-
-You should consider changing reduction mode when:
-
-- binary size is too large
-- memory footprint must be minimized
-- only single-result stemming is needed
-
-Otherwise:
-
-**use ranked mode by default**
-
-
-
-## Next steps
-
+- [Quick start](quick-start.md)
 - [Programmatic usage](programmatic-usage.md)
 - [CLI compilation](cli-compilation.md)
 - [Dictionary format](dictionary-format.md)
-
-
-
-## Summary
-
-Radixor’s architecture is built around:
-
-- patch-command tries
-- canonical subtree reduction
-- immutable compiled structures
-
-This design allows the system to remain:
-
-- fast
-- compact
-- deterministic
-- adaptable
-
-while still supporting advanced use cases such as:
-
-- ambiguity-aware stemming
-- dictionary evolution
-- controlled trade-offs between size and behavior

docs/architecture.md

@@ -0,0 +1,220 @@
+# Architecture
+
+This document explains the structural architecture of **Radixor**: what data is stored, how it flows through the build pipeline, and how runtime lookup works once a compiled trie has been produced.
+
+## The central idea
+
+Radixor does not store final stems directly as a large flat lookup table. Instead, it stores **patch commands** that describe how a word form should be transformed into a canonical stem.
+
+For example, if a dictionary states that `running` should reduce to `run`, the final runtime artifact does not need to store a full redundant `running -> run` output string entry in the simplest possible form. It can store a compact transformation command that expresses how to turn the source form into the target form.
+
+That matters because many words share similar transformation patterns. Once those mappings are organized in a trie and compiled into a canonical structure, the result is much smaller and more reusable than a naive direct-output table.
+
+## End-to-end build flow
+
+The full build-time flow is:
+
+```text
+Dictionary -> Mutable trie -> Reduced trie -> Compiled trie
+```
+
+Each stage has a different purpose.
+
+### Dictionary input
+
+The textual dictionary groups known word forms under a canonical stem:
+
+```text
+run	running	runs	ran
+connect	connected	connecting	connection
+```
+
+The first column is the canonical stem. The following tab-separated columns are known variants.
+
+### Patch-command generation
+
+Each variant is converted into a patch command that transforms the variant into the stem.
+
+Conceptually:
+
+```text
+running -> <patch> -> run
+runs    -> <patch> -> run
+ran     -> <patch> -> run
+```
+
+If `storeOriginal` is enabled, the stem itself is also inserted using a canonical no-op patch.
+
+### Mutable trie construction
+
+Those patch-command values are inserted into a mutable trie keyed by the source surface form.
+
+### Reduction
+
+Equivalent subtrees are merged into canonical reduced nodes.
+
+Before a selected semantic reduction mode is applied, Radixor also performs uniform-subtree
+contraction. If every reachable entry below a subtree resolves to the same preferred patch
+command, that subtree can be represented as an accepting leaf for that command. Runtime lookup can
+then stop at that leaf even when the input word still has remaining characters.
+
+This is a structural optimization of preferred-result lookup. It reduces trie depth in regions
+where the remaining suffix cannot change the selected command, while preserving the `get()` result
+used by the standard stemmer path. The benchmark tables in `docs/benchmarks/` are based on this
+contracted compiled representation.
+
+### Compilation
+
+The reduced structure is frozen into an immutable compiled trie optimized for runtime lookup.
+
+## Why a trie is used
+
+A trie is useful because many word forms share structural fragments. Instead of storing each word independently, the trie reuses paths and organizes lookup by character traversal.
+
+A trie node can contain:
+
+- outgoing edges,
+- one or more ordered values,
+- counts aligned with those values.
+
+This is why the structure can represent both:
+
+- a single preferred result,
+- multiple competing results for the same key.
+
+## Stage 1: Mutable construction
+
+The mutable build-time structure is created by `FrequencyTrie.Builder`.
+
+This stage is optimized for insertion rather than runtime lookup. As dictionary data is added, the builder accumulates:
+
+- child edges,
+- local values,
+- local frequencies of those values.
+
+Those frequencies are not incidental metadata. They later influence both result ordering and, depending on reduction mode, the semantic identity of subtrees during reduction.
+
+### Why the build-time form is mutable
+
+The builder must be easy to extend and easy to aggregate into. That is the opposite of what a runtime lookup structure needs.
+
+Build-time priorities are:
+
+- flexibility,
+- accumulation of counts,
+- structural growth.
+
+Runtime priorities are:
+
+- compactness,
+- immutability,
+- fast lookup.
+
+Radixor therefore keeps construction and runtime representation strictly separate.
+
+## What a compiled node contains
+
+After reduction and freezing, the runtime structure uses immutable compiled nodes.
+
+A compiled node stores:
+
+- `char[] edgeLabels`
+- child-node references aligned with those labels
+- ordered value arrays
+- aligned count arrays
+
+This array-based form is compact and efficient for lookup.
+
+## Runtime lookup model
+
+At runtime, lookup is conceptually simple:
+
+1. traverse the compiled trie by the input key,
+2. reach the node addressed by that key,
+3. retrieve one or more stored patch commands,
+4. apply the chosen patch command to the original word.
+
+The trie itself does not create the final stem string. It selects the stored transformation command. Runtime code should use `CompiledPatchCommand.apply(...)` so the serialized command is compiled once and reused.
+
+That separation is architecturally important:
+
+- the trie is responsible for **selection**,
+- patch application is responsible for **transformation**.
+
+## `get()` and `getAll()`
+
+The runtime API exposes two complementary views of the addressed node.
+
+### `get()`
+
+`get()` returns the locally preferred value stored at that node.
+
+Preference is deterministic:
+
+1. higher local frequency wins,
+2. shorter textual representation wins,
+3. lexicographically lower textual representation wins,
+4. stable first-seen order acts as the final tie-breaker.
+
+### `getAll()`
+
+`getAll()` returns all locally stored values in deterministic ranked order.
+
+This is what allows Radixor to preserve ambiguity explicitly instead of forcing every key into a single answer.
+
+## Why multiple results can exist
+
+Some stemming systems discard ambiguity early because they insist on returning exactly one answer.
+
+Radixor does not require that simplification. If multiple plausible patch commands exist for a key, the compiled trie can preserve them and the runtime API can expose them.
+
+That is useful when downstream logic wants to:
+
+- inspect ambiguity,
+- preserve alternatives for retrieval,
+- apply later ranking or domain-specific selection.
+
+## Why compiled artifacts are compact
+
+The final compiled trie can be much smaller than the original dictionary for several reasons working together:
+
+- patch commands are compact,
+- trie paths reuse shared structure,
+- uniform preferred-command subtrees can be contracted into accepting leaves,
+- reduction merges equivalent subtrees,
+- binary persistence stores the already reduced form,
+- GZip compression is applied on top of the binary format.
+
+This is why a very large dictionary can still produce a manageable deployable runtime artifact.
+
+## Why preparation can still use more memory
+
+The compactness of the final artifact should not be confused with the memory usage of preparation.
+
+Before reduction has completed, the mutable build-time structure must exist in memory. For large dictionaries, that temporary preparation cost can be noticeably higher than the size of the final persisted artifact or the loaded compiled trie.
+
+That is why the preferred operational model is usually:
+
+- compile offline,
+- persist the compiled artifact,
+- load the finished artifact in runtime services.
+
+## Determinism as a design principle
+
+Radixor favors deterministic behavior throughout the pipeline.
+
+This appears in:
+
+- lowercased dictionary parsing,
+- stable value ordering,
+- sorted child descriptors,
+- canonical reduction signatures,
+- reproducible compiled lookup behavior.
+
+Determinism matters not only for tests, but also for operational trust. It makes stemming behavior explainable and reproducible across builds and environments.
+
+## Continue with
+
+- [Reduction Semantics](reduction-semantics.md)
+- [Programmatic usage](programmatic-usage.md)
+- [CLI compilation](cli-compilation.md)

docs/assets/stylesheets/extra.css

@@ -0,0 +1,207 @@
+/* Compact technical typography for Radixor */
+
+:root {
+  --md-text-font: "Inter", "Segoe UI", "Roboto", "Helvetica Neue", Arial, sans-serif;
+}
+
+/* Hide page title only on the landing page */
+.visually-hidden {
+  display: none;
+}
+
+/* Main article text */
+.md-typeset {
+  font-size: 0.78rem;
+  line-height: 1.3;
+}
+
+/* Paragraph spacing */
+.md-typeset p,
+.md-typeset ul,
+.md-typeset ol,
+.md-typeset dl,
+.md-typeset blockquote {
+  margin-top: 0.45em;
+  margin-bottom: 0.45em;
+}
+
+/* Headings */
+.md-typeset h1 {
+  margin: 0 0 0.7rem;
+  font-size: 1.8rem;
+  line-height: 1.15;
+}
+
+.md-typeset h2 {
+  margin: 1.2rem 0 0.55rem;
+  font-size: 1.3rem;
+  line-height: 1.2;
+}
+
+.md-typeset h3 {
+  margin: 1rem 0 0.45rem;
+  font-size: 1.05rem;
+  line-height: 1.25;
+}
+
+.md-typeset h4,
+.md-typeset h5,
+.md-typeset h6 {
+  margin: 0.85rem 0 0.35rem;
+  line-height: 1.25;
+}
+
+/* Lists */
+.md-typeset li {
+  margin-bottom: 0.15em;
+}
+
+.md-typeset ul,
+.md-typeset ol {
+  padding-left: 1.1rem;
+}
+
+/* Tables */
+.md-typeset table:not([class]) td,
+.md-typeset table:not([class]) th {
+  padding: 0.45rem 0.7rem;
+}
+
+/* Publication-quality benchmark tables retain identity columns while scrolling. */
+.quality-table {
+  max-width: 100%;
+  overflow-x: auto;
+  margin: 0.65rem 0 1rem;
+  border: 1px solid var(--md-default-fg-color--lightest);
+  border-radius: 0.2rem;
+  scrollbar-gutter: stable;
+}
+
+.quality-table:focus {
+  outline: 0.15rem solid var(--md-accent-fg-color);
+  outline-offset: 0.1rem;
+}
+
+.quality-table::before {
+  content: "Scrollable table: Rank, Stemmer, and Output policy remain visible.";
+  display: block;
+  padding: 0.35rem 0.55rem;
+  color: var(--md-default-fg-color--light);
+  font-size: 0.68rem;
+}
+
+.quality-table .md-typeset__table,
+.quality-table table {
+  margin: 0;
+}
+
+.quality-table table th:nth-child(1),
+.quality-table table td:nth-child(1),
+.quality-table table th:nth-child(2),
+.quality-table table td:nth-child(2),
+.quality-table table th:nth-child(3),
+.quality-table table td:nth-child(3) {
+  position: sticky;
+  z-index: 2;
+  background: var(--md-default-bg-color);
+  background-clip: padding-box;
+}
+
+.quality-table table th:nth-child(1),
+.quality-table table td:nth-child(1) {
+  left: 0;
+  min-width: 2.8rem;
+}
+
+.quality-table table th:nth-child(2),
+.quality-table table td:nth-child(2) {
+  left: 2.8rem;
+  min-width: 13rem;
+  white-space: normal;
+}
+
+.quality-table table th:nth-child(3),
+.quality-table table td:nth-child(3) {
+  left: 15.8rem;
+  min-width: 8.5rem;
+  box-shadow: 0.2rem 0 0.25rem rgb(0 0 0 / 8%);
+}
+
+.quality-details > summary {
+  font-weight: 600;
+}
+
+@media screen and (max-width: 44.99em) {
+  .quality-table table th:nth-child(2),
+  .quality-table table td:nth-child(2) {
+    min-width: 10rem;
+  }
+
+  .quality-table table th:nth-child(3),
+  .quality-table table td:nth-child(3) {
+    position: static;
+    min-width: 7.5rem;
+    box-shadow: none;
+  }
+}
+
+@media print {
+  .quality-table {
+    overflow: visible;
+    border: 0;
+  }
+
+  .quality-table::before {
+    display: none;
+  }
+
+  .quality-table table th,
+  .quality-table table td {
+    position: static !important;
+  }
+
+  .quality-details:not([open]) > *:not(summary) {
+    display: block;
+  }
+}
+
+/* Code blocks */
+.md-typeset pre > code {
+  font-size: 0.72rem;
+  line-height: 1.4;
+}
+
+/* Inline code */
+.md-typeset code {
+  font-size: 0.72rem;
+}
+
+/* Navigation density */
+.md-nav__item .md-nav__link {
+  margin-top: 0.12rem;
+  margin-bottom: 0.12rem;
+}
+
+.md-sidebar__scrollwrap {
+  padding-top: 0.3rem;
+  padding-bottom: 0.3rem;
+}
+
+/* Slightly narrower content rhythm */
+.md-content__inner {
+  margin-top: 0.6rem;
+  padding-bottom: 1.2rem;
+}
+
+/* Admonitions more compact */
+.md-typeset .admonition,
+.md-typeset details {
+  margin: 0.8rem 0;
+}
+
+/* Optional: use a bit wider content area on large screens */
+@media screen and (min-width: 76.25em) {
+  .md-grid {
+    max-width: 68rem;
+  }
+}

docs/benchmarking.md

@@ -1,134 +1,42 @@
 # Benchmarking
 
-> ← Back to [README.md](../README.md)
+Radixor contains internal trie microbenchmarks, a separate stemmer comparison suite, and a dictionary coverage benchmark for Radixor itself. Published stemmer comparison results must come only from benchmark classes matching `.*StemmerComparisonBenchmark.*`; internal `FrequencyTrie*` microbenchmarks are not part of those results.
 
-Radixor includes a JMH benchmark suite for both the internal algorithmic core and a side-by-side English comparison against the Snowball Porter stemmer family.
+This page is the entry point for benchmark interpretation. Detailed tables and long reference material are split into focused subpages so that important points do not get buried.
 
-This document explains what is benchmarked, how to run it, and how to interpret the results responsibly.
+## Key Takeaways
 
-## Scope
+- Speed and accuracy must be read together. A faster row is not necessarily a better stemmer.
+- Radixor is the quality-oriented baseline in same-language comparisons. Its exact-root accuracy is often close to 100%, while many faster competitors are light, minimal, possessive, or aggressive rule-based stemmers with much lower root agreement.
+- The measured Radixor cost buys dictionary-trained stemming precision. That precision improves search quality by mapping inflected forms to intended dictionary roots instead of approximate or over-reduced stems.
+- Speed benchmarks process changed dictionary tokens where the surface form differs from the expected root. Accuracy benchmarks process the complete dictionary.
+- Accuracy tables use deterministic auxiliary counters from the current JMH reports. Repeated measurement samples duplicate the same exact-root accounting and are not interpreted as timing results.
+- The historical Porter performance badge is retired. Benchmark reporting now uses speed and quality tables rather than a single Porter ratio.
 
-The benchmark suite currently covers two categories:
+## Benchmark Documentation Map
 
-- Radixor core operations
-- English stemmer comparison on the same token workload
+| Page | Purpose |
+| --- | --- |
+| [Benchmark methodology](benchmarks/reference/methodology.md) | Workload design, speed pass, quality pass, normalization policy, and exact-root metrics. |
+| [Linguistic quality methodology](benchmarks/reference/linguistic-quality.md) | Pairwise gold standard, over/under-stemming, candidate policies, metrics, and ranking rules. |
+| [Tested stemmers](benchmarks/reference/tested-stemmers.md) | Upstream attribution, tested versions, language coverage, adapter behaviour, and limitations. |
+| [Reproducibility and raw data](benchmarks/reference/reproducibility.md) | Versioned quality snapshot, checksum, commands, reports, and provenance limitations. |
+| [Benchmark corpora](benchmarks/reference/corpora.md) | Dictionary row counts, complete quality tokens, already-root tokens, changed speed tokens, and timing token counts. |
+| [Benchmark environment and reports](benchmarks/reference/environment.md) | Hardware, OS, JVM, JMH settings, report files, and current badge/report policy. |
+| [English dictionary coverage benchmark](benchmarks/reference/english-coverage.md) | The quality/speed operating curve for contracted Radixor tries built from 100% down to 10% of English dictionary rows. |
+| [Candidate evaluation](benchmarks/reference/candidates.md) | Included benchmark families and evaluated candidates that were skipped. |
+| [Language benchmark pages](benchmarks/languages/index.md) | Per-language accuracy tables, speed tables, and implementation notes. |
 
-The comparison benchmark processes the same deterministic English token stream through:
+## How To Read Results
 
-- Radixor with bundled `US_UK_PROFI`
-- Snowball original Porter
-- Snowball English, commonly referred to as Porter2
+Start with the [language benchmark pages](benchmarks/languages/index.md). Each language page lists accuracy first and speed second because throughput without root agreement is not enough to interpret stemmer quality.
 
-The purpose of the comparison is throughput measurement on identical input. It is not intended to prove linguistic equivalence between the compared stemmers.
+When Radixor is slower than a narrow competitor, check the accuracy table before drawing a conclusion. Many Lucene light/minimal filters and possessive filters intentionally do less work. They can be fast precisely because they are not trying to match the dictionary root with the same precision.
 
-## Current snapshot
+The [English dictionary coverage benchmark](benchmarks/reference/english-coverage.md) shows the central operating curve explicitly: contracted tries preserve high quality even at reduced dictionary coverage, while changed-form exactness still reflects how much language knowledge was available during training. This is why Radixor performance should be discussed as a configurable quality/speed point, not as a single fixed ratio against Porter.
 
-A recent JMH run on JDK 21.0.10 with JMH 1.37, one thread, three warmup iterations, and five measurement iterations produced the following approximate throughput ranges:
+## Current Result Locations
 
-| Workload | Radixor `US_UK_PROFI` | Snowball Porter | Snowball English |
-| --- | ---: | ---: | ---: |
-| About 12,000 generated tokens | 30.99 M tokens/s | 8.21 M tokens/s | 5.46 M tokens/s |
-| About 60,000 generated tokens | 32.25 M tokens/s | 8.02 M tokens/s | 5.11 M tokens/s |
+The current measured language results are published in [Language Benchmark Pages](benchmarks/languages/index.md). Generated local report files for this benchmark update are listed in [Benchmark environment and reports](benchmarks/reference/environment.md).
 
-On that workload, Radixor is approximately:
-
-- 4 times faster than Snowball original Porter
-- 6 times faster than Snowball English
-
-These values are workload- and environment-dependent. Treat them as measured results for the documented benchmark setup, not as universal constants.
-
-## Benchmark classes
-
-The main benchmark classes are under `src/jmh/java/org/egothor/stemmer/benchmark`.
-
-Relevant classes include:
-
-- `FrequencyTrieLookupBenchmark`
-- `FrequencyTrieCompilationBenchmark`
-- `EnglishStemmerComparisonBenchmark`
-
-The English comparison benchmark uses the bundled Radixor English resource and the official Snowball Java distribution integrated into the JMH source set.
-
-## Workload design
-
-The English comparison benchmark uses a deterministic generated corpus rather than an uncontrolled ad hoc text sample.
-
-The workload intentionally mixes:
-
-- simple inflections
-- common derivational forms
-- US and UK spelling families
-- lexical forms appropriate for `US_UK_PROFI`
-
-This design keeps runs reproducible across environments and avoids accidental drift caused by changing external corpora.
-
-## Running benchmarks
-
-Run the full benchmark suite:
-
-```bash
-./gradlew jmh
-```
-
-Run only the English comparison benchmark:
-
-```bash
-./gradlew jmh -Pjmh.includes=EnglishStemmerComparisonBenchmark
-```
-
-## Generated reports
-
-JMH reports are written to:
-
-- `build/reports/jmh/jmh-results.txt`
-- `build/reports/jmh/jmh-results.csv`
-
-The text report is convenient for human review. The CSV report is more useful for CI archiving, historical tracking, and external processing.
-
-## Interpreting results
-
-Benchmark numbers should be read with care.
-
-Important factors include:
-
-- CPU model and frequency behavior
-- thermal throttling
-- JVM vendor and version
-- system background load
-- operating-system scheduling noise
-- benchmark parameter changes
-
-For meaningful comparison, keep these stable:
-
-- hardware or VM class
-- JDK version
-- benchmark parameters
-- thread count
-- benchmark source revision
-
-If a regression is suspected, repeat the run and compare against the previous CSV output rather than relying on a single measurement.
-
-## Regression tracking
-
-The recommended regression workflow is:
-
-1. archive `jmh-results.csv`
-2. compare the same benchmark names across runs
-3. compare only like-for-like environments
-4. investigate sustained regressions rather than one-off noise
-
-For public reporting, the README should keep only the condensed benchmark summary, while detailed benchmark methodology and interpretation should remain in this document.
-
-## Notes on comparison fairness
-
-Radixor, Snowball Porter, and Snowball English are not the same kind of stemmer.
-
-Radixor uses a compiled patch-command trie driven by dictionary data. Snowball Porter and Snowball English are rule-based English stemmers.
-
-Because of that, the comparison should be understood as:
-
-- equal input workload
-- different stemming strategies
-- measured throughput, not semantic identity
-
-That distinction matters whenever performance claims are discussed in documentation or release notes.
+JMH TXT and CSV reports are still published as benchmark artifacts. They are no longer converted into a Shields endpoint benchmark badge.

docs/benchmarks/data/stemming-quality.csv

@@ -0,0 +1,309 @@
+Stemmer,Language,Dictionary mode,Output policy,Applied dictionary rows,Processed word forms,Singleton dictionary rows,Forms with one candidate,Forms with multiple candidates,Maximum candidates for one form,Total candidate assignments,Distinct output stems,True-positive pairs,False-positive pairs,False-negative pairs,True-negative pairs,Over-stemming error pairs,Over-stemming possible pairs,Over-stemming percentage,Under-stemming error pairs,Under-stemming possible pairs,Under-stemming percentage,Pairwise precision,Pairwise recall,Pairwise specificity,Pairwise accuracy,Balanced accuracy,Pairwise F0.5,Pairwise F1,Pairwise F2,Jaccard index,Fowlkes-Mallows index,Matthews correlation coefficient,Pairwise error rate,Adjusted Rand Index,Homogeneity,Completeness,V-measure,Normalized mutual information
+"CZECH_LUCENE_CZECH_STEM_FILTER","CS_CZ","ALL_WORDS","PRIMARY_OUTPUT","5113","51676","2","51676","0","1","51676","9647","177249","14480","124586","1334862335","14480","1334876815","0.001085","124586","301835","41.276194","0.924476735392","0.587238060530","0.999989152557","0.999895844650","0.793613606543","0.829234311828","0.718241200736","0.633453389361","0.560355974266","0.736809286788","0.736765291417","0.000104155350","0.718191706079","0.993800637348","0.944976928457","0.968774025802","0.968774025802"
+"CZECH_LUCENE_CZECH_STEM_FILTER","CS_CZ","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","5038","50968","2","50968","0","1","50968","9558","174387","13950","124426","1298530265","13950","1298544215","0.001074","124426","298813","41.640089","0.925930645598","0.583599107134","0.999989257201","0.999893462107","0.791794182167","0.828708724235","0.715947860002","0.630197985095","0.557569149804","0.735100195918","0.735055396892","0.000106537893","0.715897321649","0.993897397445","0.944297456221","0.968462775946","0.968462775946"
+"CZECH_RADIXOR","CS_CZ","ALL_WORDS","PRIMARY_OUTPUT","5113","51676","2","51676","0","1","51676","5162","299762","3867","2073","1334872948","3867","1334876815","0.000290","2073","301835","0.686799","0.987264062392","0.993132009210","0.999997103103","0.999995551157","0.996564556157","0.988432097845","0.990189342389","0.991952846154","0.980569312599","0.990193689085","0.990191466141","0.000004448843","0.990187117482","0.998733220675","0.998685552738","0.998709386137","0.998709386137"
+"CZECH_RADIXOR","CS_CZ","ALL_WORDS","ANY_CANDIDATE","5113","51676","2","51080","596","4","52319","5166","301835","0","0","1334876815","0","1334876815","0.000000","0","301835","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"CZECH_RADIXOR","CS_CZ","ALL_WORDS","ALL_CANDIDATES","5113","51676","2","51080","596","4","52319","5166","301835","5850","0","1334870965","5850","1334876815","0.000438","0","301835","0.000000","0.980987048442","1.000000000000","0.999995617573","0.999995618564","0.999997808787","0.984731579205","0.990402283764","0.996138677580","0.980987048442","0.990447902942","0.990445732657","0.000004381436","","","","",""
+"CZECH_RADIXOR","CS_CZ","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","5038","50968","2","50968","0","1","50968","5037","297104","3863","1709","1298540352","3863","1298544215","0.000297","1709","298813","0.571930","0.987164705765","0.994280703985","0.999997025130","0.999995710028","0.997138864558","0.988579745136","0.990709926973","0.992849308824","0.981590876052","0.990716315904","0.990714173387","0.000004289972","0.990707781520","0.998726091764","0.999029907266","0.998877976413","0.998877976413"
+"CZECH_RADIXOR","CS_CZ","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","5038","50968","2","50428","540","4","51543","5040","298813","0","0","1298544215","0","1298544215","0.000000","0","298813","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"CZECH_RADIXOR","CS_CZ","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","5038","50968","2","50428","540","4","51543","5040","298813","5782","0","1298538433","5782","1298544215","0.000445","0","298813","0.000000","0.981017416570","1.000000000000","0.999995547321","0.999995548346","0.999997773661","0.984756059381","0.990417760454","0.996144940117","0.981017416570","0.990463233325","0.990461028216","0.000004451654","","","","",""
+"DA_DK_RADIXOR","DA_DK","ALL_WORDS","PRIMARY_OUTPUT","4179","28079","32","28079","0","1","28079","4184","89188","1165","707","394110021","1165","394111186","0.000296","707","89895","0.786473","0.987106128186","0.992135268925","0.999997043981","0.999995251155","0.996066156453","0.988107873355","0.989614309174","0.991125345329","0.979442126071","0.989617503860","0.989615130363","0.000004748845","0.989611934224","0.998465862775","0.998718664384","0.998592247580","0.998592247580"
+"DA_DK_RADIXOR","DA_DK","ALL_WORDS","ANY_CANDIDATE","4179","28079","32","27756","323","3","28405","4187","89895","0","0","394111186","0","394111186","0.000000","0","89895","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"DA_DK_RADIXOR","DA_DK","ALL_WORDS","ALL_CANDIDATES","4179","28079","32","27756","323","3","28405","4187","89895","1849","0","394109337","1849","394111186","0.000469","0","89895","0.000000","0.979846093478","1.000000000000","0.999995308431","0.999995309500","0.999997654215","0.983811622975","0.989820468071","0.995903164910","0.979846093478","0.989871756076","0.989869434047","0.000004690500","","","","",""
+"DA_DK_RADIXOR","DA_DK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4173","28033","32","28033","0","1","28033","4170","89077","1165","663","392819623","1165","392820788","0.000297","663","89740","0.738801","0.987090268389","0.992611990194","0.999997034271","0.999995347541","0.996304512232","0.988189692661","0.989843428787","0.991502709249","0.979891095099","0.989847279032","0.989844954043","0.000004652459","0.989841102043","0.998463063294","0.998811876590","0.998637439483","0.998637439483"
+"DA_DK_RADIXOR","DA_DK","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","4173","28033","32","27718","315","3","28351","4173","89740","0","0","392820788","0","392820788","0.000000","0","89740","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"DA_DK_RADIXOR","DA_DK","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","4173","28033","32","27718","315","3","28351","4173","89740","1849","0","392818939","1849","392820788","0.000471","0","89740","0.000000","0.979811986156","1.000000000000","0.999995293019","0.999995294094","0.999997646509","0.983784115625","0.989803065147","0.995896117847","0.979811986156","0.989854527774","0.989852198158","0.000004705906","","","","",""
+"ENGLISH_LUCENE_KSTEM_FILTER","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","371125","237565","1368501","76305","184489083270","1368501","184490451771","0.000742","76305","313870","24.311020","0.147917333410","0.756889795138","0.999992582267","0.999992168681","0.878441188702","0.176283968232","0.247471790726","0.415099040868","0.141208449266","0.334599940499","0.334597833111","0.000007831319","0.247469648794","0.980686838187","0.992107972963","0.986364345289","0.986364345289"
+"ENGLISH_LUCENE_KSTEM_FILTER","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","347624","237551","1367069","74340","170473473135","1367069","170474840204","0.000802","74340","311891","23.835250","0.148041904002","0.761647498645","0.999991980817","0.999991544756","0.880819739731","0.176476898525","0.247899438093","0.416437018089","0.141486991947","0.335791223646","0.335788961354","0.000008455244","0.247897133948","0.979822223835","0.991990504792","0.985868818390","0.985868818390"
+"ENGLISH_LUCENE_MINIMAL_FILTER","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","453328","137225","1122264","176645","184489329507","1122264","184490451771","0.000608","176645","313870","56.279670","0.108952916619","0.437203300730","0.999993916953","0.999992959490","0.718598608842","0.128203906480","0.174435713655","0.272816484020","0.095551668577","0.218253464509","0.218250987161","0.000007040510","0.174433464995","0.995202198233","0.981173943304","0.988138284715","0.988138284715"
+"ENGLISH_LUCENE_MINIMAL_FILTER","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","430129","136932","1120871","174959","170473719333","1120871","170474840204","0.000657","174959","311891","56.096200","0.108866014789","0.439037997249","0.999993425006","0.999992398716","0.719515711128","0.128139023335","0.174469673707","0.273277328232","0.095572048952","0.218623688336","0.218621020778","0.000007601284","0.174467253215","0.994993790771","0.980519680106","0.987703711280","0.987703711280"
+"ENGLISH_LUCENE_PORTER_COPIED","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","319968","285390","1557406","28480","184488894365","1557406","184490451771","0.000844","28480","313870","9.073820","0.154867928951","0.909261796285","0.999991558338","0.999991403982","0.954626677312","0.185678591198","0.264658505304","0.460562583837","0.152510906996","0.375253902399","0.375251973425","0.000008596018","0.264656367392","0.969648379409","0.997199419831","0.983230936080","0.983230936080"
+"ENGLISH_LUCENE_PORTER_COPIED","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","298779","283761","1552702","28130","170473287502","1552702","170474840204","0.000911","28130","311891","9.019177","0.154514956196","0.909808234287","0.999990891899","0.999990726907","0.954899563093","0.185277176317","0.264165961476","0.460049474275","0.152183881415","0.374938634269","0.374936557303","0.000009273093","0.264163659878","0.968644600847","0.997108656044","0.982670549062","0.982670549062"
+"ENGLISH_LUCENE_PORTER_FILTER","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","319968","285390","1557406","28480","184488894365","1557406","184490451771","0.000844","28480","313870","9.073820","0.154867928951","0.909261796285","0.999991558338","0.999991403982","0.954626677312","0.185678591198","0.264658505304","0.460562583837","0.152510906996","0.375253902399","0.375251973425","0.000008596018","0.264656367392","0.969648379409","0.997199419831","0.983230936080","0.983230936080"
+"ENGLISH_LUCENE_PORTER_FILTER","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","298779","283761","1552702","28130","170473287502","1552702","170474840204","0.000911","28130","311891","9.019177","0.154514956196","0.909808234287","0.999990891899","0.999990726907","0.954899563093","0.185277176317","0.264165961476","0.460049474275","0.152183881415","0.374938634269","0.374936557303","0.000009273093","0.264163659878","0.968644600847","0.997108656044","0.982670549062","0.982670549062"
+"ENGLISH_LUCENE_POSSESSIVE_FILTER","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","591899","7","1115154","313863","184489336617","1115154","184490451771","0.000604","313863","313870","99.997770","0.000006277121","0.000022302227","0.999993955492","0.999992254263","0.500008128860","0.000007330589","0.000009796848","0.000014763939","0.000004898448","0.000011831896","0.000008625150","0.000007745737","0.000007141644","0.995789196698","0.958018540631","0.976538780935","0.976538780935"
+"ENGLISH_LUCENE_POSSESSIVE_FILTER","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","568400","5","1113773","311886","170473726431","1113773","170474840204","0.000653","311886","311891","99.998397","0.000004489225","0.000016031242","0.999993466643","0.999991637145","0.500004748942","0.000005244385","0.000007014251","0.000010587200","0.000003507138","0.000008483387","0.000005026087","0.000008362855","0.000004155674","0.995605378040","0.956423154691","0.975621022465","0.975621022465"
+"ENGLISH_OPENNLP_PORTER","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","319968","285390","1557406","28480","184488894365","1557406","184490451771","0.000844","28480","313870","9.073820","0.154867928951","0.909261796285","0.999991558338","0.999991403982","0.954626677312","0.185678591198","0.264658505304","0.460562583837","0.152510906996","0.375253902399","0.375251973425","0.000008596018","0.264656367392","0.969648379409","0.997199419831","0.983230936080","0.983230936080"
+"ENGLISH_OPENNLP_PORTER","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","298779","283761","1552702","28130","170473287502","1552702","170474840204","0.000911","28130","311891","9.019177","0.154514956196","0.909808234287","0.999990891899","0.999990726907","0.954899563093","0.185277176317","0.264165961476","0.460049474275","0.152183881415","0.374938634269","0.374936557303","0.000009273093","0.264163659878","0.968644600847","0.997108656044","0.982670549062","0.982670549062"
+"ENGLISH_PAICE_HUSK_LANCASTER","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","268169","283991","3062661","29879","184487389110","3062661","184490451771","0.001660","29879","313870","9.519546","0.084858240415","0.904804536910","0.999983399352","0.999983237427","0.952393968131","0.103642734217","0.155164208820","0.308542866654","0.084107327905","0.277092260667","0.277089454298","0.000016762573","0.155161580693","0.937768073854","0.996599815184","0.966289292109","0.966289292109"
+"ENGLISH_PAICE_HUSK_LANCASTER","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","249411","282398","3045870","29493","170471794334","3045870","170474840204","0.001787","29493","311891","9.456188","0.084848335531","0.905438117804","0.999982133023","0.999981960051","0.952710125414","0.103632575002","0.155156958803","0.308575577075","0.084103067491","0.277173081705","0.277170064389","0.000018039949","0.155154132315","0.936076835754","0.996486960554","0.965337716341","0.965337716341"
+"ENGLISH_RADIXOR","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","390361","292001","1149886","21869","184489301885","1149886","184490451771","0.000623","21869","313870","6.967534","0.202513095686","0.930324656705","0.999993767233","0.999993648707","0.965159211969","0.240076409811","0.332621199859","0.541270431499","0.199487482886","0.434054059102","0.434052478080","0.000006351293","0.332619335001","0.994214506865","0.997769723414","0.995988942533","0.995988942533"
+"ENGLISH_RADIXOR","US_UK","ALL_WORDS","ANY_CANDIDATE","396939","607439","250964","578231","29208","1355","2838145","397392","313855","12","15","184490451759","12","184490451771","0.000000","15","313870","0.004779","0.999961767245","0.999952209513","0.999999999935","0.999999999854","0.999976104724","0.999959855684","0.999956988357","0.999954121045","0.999913980413","0.999956988368","0.999956988295","0.000000000146","","","","",""
+"ENGLISH_RADIXOR","US_UK","ALL_WORDS","ALL_CANDIDATES","396939","607439","250964","578231","29208","1355","2838145","397392","313855","11482166","15","184478969605","11482166","184490451771","0.006224","15","313870","0.004779","0.026606853277","0.999952209513","0.999937762817","0.999937762842","0.999944986165","0.033038791524","0.051834488023","0.120237128281","0.026606819443","0.163112175274","0.163107098882","0.000062237158","","","","",""
+"ENGLISH_RADIXOR","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","367590","290572","1148489","21319","170473691715","1148489","170474840204","0.000674","21319","311891","6.835401","0.201917778329","0.931645991709","0.999993263000","0.999993137956","0.965819627354","0.239424468968","0.331901731173","0.540775136091","0.198970131062","0.433723286019","0.433721583515","0.000006862044","0.331899721995","0.993959181482","0.997731171071","0.995841604460","0.995841604460"
+"ENGLISH_RADIXOR","US_UK","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","374384","583910","228735","555084","28826","1355","2812871","374506","311891","0","0","170474840204","0","170474840204","0.000000","0","311891","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"ENGLISH_RADIXOR","US_UK","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","374384","583910","228735","555084","28826","1355","2812871","374506","311891","11470018","0","170463370186","11470018","170474840204","0.006728","0","311891","0.000000","0.026472025883","1.000000000000","0.999932717239","0.999932717362","0.999966358619","0.032872482055","0.051578660140","0.119686728696","0.026472025883","0.162702261457","0.162696787836","0.000067282638","","","","",""
+"ENGLISH_SNOWBALL_ORIGINAL_PORTER","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","321092","285304","1555293","28566","184488896478","1555293","184490451771","0.000843","28566","313870","9.101220","0.155006228957","0.908987797496","0.999991569791","0.999991414969","0.954489683644","0.185835337999","0.264848800190","0.460750814660","0.152637303435","0.375364850057","0.375362921954","0.000008585031","0.264846663203","0.969891477221","0.997192899073","0.983352728141","0.983352728141"
+"ENGLISH_SNOWBALL_ORIGINAL_PORTER","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","299877","283675","1550615","28216","170473289589","1550615","170474840204","0.000910","28216","311891","9.046750","0.154651118416","0.909532496930","0.999990904142","0.999990738644","0.954761700536","0.185431499934","0.264353286139","0.460234326480","0.152308234175","0.375046954242","0.375044878196","0.000009261356","0.264350985524","0.968893806180","0.997101844661","0.982795461434","0.982795461434"
+"ENGLISH_SNOWBALL_PORTER2","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","318385","285334","1566711","28536","184488885060","1566711","184490451771","0.000849","28536","313870","9.091662","0.154064291094","0.909083378469","0.999991507902","0.999991353242","0.954537443185","0.184752753479","0.263476636895","0.459101696688","0.151726514306","0.374242282819","0.374240346981","0.000008646758","0.263474493989","0.969037354042","0.997181597682","0.982908049045","0.982908049045"
+"ENGLISH_SNOWBALL_PORTER2","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","297220","283730","1561891","28161","170473278313","1561891","170474840204","0.000916","28161","311891","9.029116","0.153731454074","0.909708840589","0.999990837997","0.999990672823","0.954849839293","0.184374949232","0.263015918336","0.458637294569","0.151421029768","0.373966392672","0.373964308569","0.000009327177","0.263013611479","0.968019617024","0.997095706553","0.982342555079","0.982342555079"
+"FINNISH_LUCENE_FINNISH_LIGHT_STEM_FILTER","FI_FI","ALL_WORDS","PRIMARY_OUTPUT","57027","1811717","292","1811717","0","1","1811717","439975","12355389","2223150","19168306","1641124591341","2223150","1641126814491","0.000135","19168306","31523695","60.806025","0.847505295284","0.391939745642","0.999998645351","0.999986965635","0.695969195497","0.687649407375","0.535999578676","0.439151826652","0.366119825424","0.576342788507","0.576337821084","0.000013034365","0.535993941880","0.988126027331","0.886473473160","0.934543630400","0.934543630400"
+"FINNISH_LUCENE_FINNISH_LIGHT_STEM_FILTER","FI_FI","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","54762","1757055","274","1757055","0","1","1757055","431848","11988389","1806392","18825444","1543587637760","1806392","1543589444152","0.000117","18825444","30813833","61.094133","0.869052506162","0.389058673746","0.999998829746","0.999986634125","0.694528751746","0.697056446146","0.537492108587","0.437372459518","0.367513988637","0.581474346349","0.581469391800","0.000013365875","0.537486398327","0.989268269625","0.885293761089","0.934397488899","0.934397488899"
+"FINNISH_RADIXOR","FI_FI","ALL_WORDS","PRIMARY_OUTPUT","57027","1811717","292","1811717","0","1","1811717","69091","30552427","731279","971268","1641126083212","731279","1641126814491","0.000045","971268","31523695","3.081073","0.976624284859","0.969189271753","0.999999554404","0.999998962594","0.984594413078","0.975128170336","0.972892573600","0.970667204156","0.947215985975","0.972899675927","0.972899157490","0.000001037406","0.972892054895","0.996084757586","0.993746341306","0.994914175412","0.994914175412"
+"FINNISH_RADIXOR","FI_FI","ALL_WORDS","ANY_CANDIDATE","57027","1811717","292","1754389","57328","6","1876272","69769","31523695","0","0","1641126814491","0","1641126814491","0.000000","0","31523695","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"FINNISH_RADIXOR","FI_FI","ALL_WORDS","ALL_CANDIDATES","57027","1811717","292","1754389","57328","6","1876272","69769","31523695","1683575","0","1641125130916","1683575","1641126814491","0.000103","0","31523695","0.000000","0.949301011495","1.000000000000","0.999998974135","0.999998974154","0.999999487067","0.959025334376","0.973991195713","0.989431554710","0.949301011495","0.974320794962","0.974320295201","0.000001025846","","","","",""
+"FINNISH_RADIXOR","FI_FI","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","54762","1757055","274","1757055","0","1","1757055","54633","30078528","730145","735305","1543588714007","730145","1543589444152","0.000047","735305","30813833","2.386282","0.976300667023","0.976137178390","0.999999526982","0.999999050641","0.988068352686","0.976267964916","0.976218915862","0.976169871736","0.953542638154","0.976218919284","0.976218444595","0.000000949359","0.976218441173","0.996000407428","0.996068984852","0.996034694959","0.996034694959"
+"FINNISH_RADIXOR","FI_FI","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","54762","1757055","274","1712724","44331","6","1805864","54984","30813833","0","0","1543589444152","0","1543589444152","0.000000","0","30813833","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"FINNISH_RADIXOR","FI_FI","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","54762","1757055","274","1712724","44331","6","1805864","54984","30813833","1653320","0","1543587790832","1653320","1543589444152","0.000107","0","30813833","0.000000","0.949077148834","1.000000000000","0.999998928912","0.999998928933","0.999999464456","0.958842548108","0.973873352732","0.989382907677","0.949077148834","0.974205906795","0.974205385065","0.000001071067","","","","",""
+"FRENCH_LUCENE_FRENCH_LIGHT_STEM_FILTER","FR_FR","ALL_WORDS","PRIMARY_OUTPUT","59240","425210","2301","425210","0","1","425210","245918","202782","276403","5251833","90395828427","276403","90396104830","0.000306","5251833","5454615","96.282377","0.423181026117","0.037176226003","0.999996942313","0.999938848002","0.518586584158","0.137547303040","0.068348107452","0.045471618191","0.035383242558","0.125428359900","0.125414592230","0.000061151998","0.068339028277","0.974109647704","0.812375827422","0.885921707253","0.885921707253"
+"FRENCH_LUCENE_FRENCH_LIGHT_STEM_FILTER","FR_FR","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","57698","421231","2133","421231","0","1","421231","245182","200690","262689","5239869","88711863817","262689","88712126506","0.000296","5239869","5440559","96.311225","0.433101197939","0.036887753630","0.999997038860","0.999937976681","0.518442396245","0.137570562409","0.067985131280","0.045148356975","0.035188720533","0.126396717862","0.126383026150","0.000062023319","0.067976159358","0.975085555241","0.811143708698","0.885591261484","0.885591261484"
+"FRENCH_LUCENE_FRENCH_MINIMAL_STEM_FILTER","FR_FR","ALL_WORDS","PRIMARY_OUTPUT","59240","425210","2301","425210","0","1","425210","269236","183612","160438","5271003","90395944392","160438","90396104830","0.000177","5271003","5454615","96.633823","0.533678244441","0.033661770812","0.999998225167","0.999939918724","0.516829997990","0.134399775137","0.063329059361","0.041424008382","0.032699958487","0.134031916915","0.134021061615","0.000060081276","0.063322352769","0.984019125555","0.810978546011","0.889158144694","0.889158144694"
+"FRENCH_LUCENE_FRENCH_MINIMAL_STEM_FILTER","FR_FR","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","57698","421231","2133","421231","0","1","421231","268411","181686","147476","5258873","88711979030","147476","88712126506","0.000166","5258873","5440559","96.660527","0.551965293685","0.033394730211","0.999998337589","0.999939061122","0.516696533900","0.134438681544","0.062979128454","0.041121435592","0.032513396928","0.135767198057","0.135756528540","0.000060938878","0.062972571968","0.985086367216","0.809773733549","0.888868235590","0.888868235590"
+"FRENCH_RADIXOR","FR_FR","ALL_WORDS","PRIMARY_OUTPUT","59240","425210","2301","425210","0","1","425210","60225","4985455","318767","469160","90395786063","318767","90396104830","0.000353","469160","5454615","8.601157","0.939903156391","0.913988429981","0.999996473664","0.999991284144","0.956992451823","0.934603310507","0.926764667966","0.919056419273","0.863524187383","0.926855226151","0.926850879168","0.000008715856","0.926760310630","0.988772235003","0.985214034569","0.986989927876","0.986989927876"
+"FRENCH_RADIXOR","FR_FR","ALL_WORDS","ANY_CANDIDATE","59240","425210","2301","382170","43040","56","477024","60383","5454383","12","232","90396104818","12","90396104830","0.000000","232","5454615","0.004253","0.999997799939","0.999957467209","0.999999999867","0.999999997301","0.999978733538","0.999989733133","0.999977633167","0.999965533495","0.999955267335","0.999977633371","0.999977632021","0.000000002699","","","","",""
+"FRENCH_RADIXOR","FR_FR","ALL_WORDS","ALL_CANDIDATES","59240","425210","2301","382170","43040","56","477024","60383","5454383","1056255","232","90395048575","1056255","90396104830","0.001168","232","5454615","0.004253","0.837764747479","0.999957467209","0.999988315260","0.999988313399","0.999972891234","0.865852951156","0.911703747510","0.962682080453","0.837734895644","0.915275431226","0.915270082203","0.000011686601","","","","",""
+"FRENCH_RADIXOR","FR_FR","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","57698","421231","2133","421231","0","1","421231","58069","4975123","315266","465436","88711811240","315266","88712126506","0.000355","465436","5440559","8.554930","0.940407784758","0.914450702584","0.999996446190","0.999991200142","0.957223574387","0.935099145312","0.927247620620","0.919526848134","0.864363145162","0.927338427699","0.927334038919","0.000008799858","0.927243221287","0.988915897225","0.985549842615","0.987230000708","0.987230000708"
+"FRENCH_RADIXOR","FR_FR","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","57698","421231","2133","380101","41130","56","468574","58208","5440559","0","0","88712126506","0","88712126506","0.000000","0","5440559","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"FRENCH_RADIXOR","FR_FR","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","57698","421231","2133","380101","41130","56","468574","58208","5440559","938985","0","88711187521","938985","88712126506","0.001058","0","5440559","0.000000","0.852813147774","1.000000000000","0.999989415370","0.999989416019","0.999994707685","0.878679151458","0.920560336911","0.966633773699","0.852813147774","0.923478829088","0.923473941734","0.000010583981","","","","",""
+"GERMAN_CISTEM","DE_DE","ALL_WORDS","PRIMARY_OUTPUT","54092","296974","1474","296974","0","1","296974","59097","1053889","477122","329983","44094768857","477122","44095245979","0.001082","329983","1383872","23.844908","0.688361481400","0.761550923785","0.999989179741","0.999981696901","0.880770051763","0.701851885397","0.723108954973","0.745693871888","0.566304351331","0.724031989665","0.724022910459","0.000018303099","0.723099826442","0.974048119240","0.975147027686","0.974597263694","0.974597263694"
+"GERMAN_CISTEM","DE_DE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","16007","150098","228","150098","0","1","150098","23023","725447","156784","147964","11263599558","156784","11263756342","0.001392","147964","873411","16.940936","0.822286906717","0.830590638313","0.999986080665","0.999972946470","0.915288359489","0.823934343933","0.826417914358","0.828916502414","0.704184159310","0.826428343371","0.826414817348","0.000027053530","0.826404386881","0.985935685912","0.973569821618","0.979713735095","0.979713735095"
+"GERMAN_LUCENE_GERMAN_LIGHT_STEM_FILTER","DE_DE","ALL_WORDS","PRIMARY_OUTPUT","54092","296974","1474","296974","0","1","296974","98357","709263","205740","674609","44095040239","205740","44095245979","0.000467","674609","1383872","48.747933","0.775148278202","0.512520666651","0.999995334191","0.999980035912","0.756258000421","0.703092101246","0.617052253820","0.549774428024","0.446186239158","0.630301128270","0.630292039259","0.000019964088","0.617042686770","0.980753120457","0.936533167951","0.958133203614","0.958133203614"
+"GERMAN_LUCENE_GERMAN_LIGHT_STEM_FILTER","DE_DE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","16007","150098","228","150098","0","1","150098","50335","471565","55477","401846","11263700865","55477","11263756342","0.000493","401846","873411","46.008809","0.894738939212","0.539911908597","0.999995074734","0.999959401861","0.769953491666","0.790797426464","0.673446377708","0.586423560557","0.507666155661","0.695039717114","0.695022690564","0.000040598139","0.673427319227","0.991320177896","0.915069562866","0.951669957566","0.951669957566"
+"GERMAN_LUCENE_GERMAN_MINIMAL_STEM_FILTER","DE_DE","ALL_WORDS","PRIMARY_OUTPUT","54092","296974","1474","296974","0","1","296974","140505","271626","110840","1112246","44095135139","110840","44095245979","0.000251","1112246","1383872","80.372029","0.710196461908","0.196279713731","0.999997486350","0.999972263504","0.598138600041","0.466112921692","0.307558349534","0.229493166050","0.181724639931","0.373359288402","0.373350267608","0.000027736496","0.307548938689","0.983615403456","0.896263607272","0.937910029995","0.937910029995"
+"GERMAN_LUCENE_GERMAN_MINIMAL_STEM_FILTER","DE_DE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","16007","150098","228","150098","0","1","150098","80363","132221","21214","741190","11263735128","21214","11263756342","0.000188","741190","873411","84.861537","0.861739498811","0.151384628772","0.999998116614","0.999932318770","0.575691372693","0.444544636019","0.257528392768","0.181269722976","0.147794886125","0.361184321539","0.361168285320","0.000067681230","0.257511188319","0.992642524078","0.854402700840","0.918349417869","0.918349417869"
+"GERMAN_LUCENE_GERMAN_STEM_FILTER","DE_DE","ALL_WORDS","PRIMARY_OUTPUT","54092","296974","1474","296974","0","1","296974","81085","619354","331871","764518","44094914108","331871","44095245979","0.000753","764518","1383872","55.244849","0.651111987174","0.447551507654","0.999992473769","0.999975136671","0.723771990712","0.596821367368","0.530473894660","0.477402037056","0.360982967729","0.539820480819","0.539808754751","0.000024863329","0.530461889452","0.975549631706","0.942889706548","0.958941664321","0.958941664321"
+"GERMAN_LUCENE_GERMAN_STEM_FILTER","DE_DE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","16007","150098","228","150098","0","1","150098","41574","378734","78723","494677","11263677619","78723","11263756342","0.000699","494677","873411","56.637368","0.827911694432","0.433626322545","0.999993010946","0.999949097306","0.716809666745","0.700518896035","0.569153364571","0.479276535831","0.397773842757","0.599169678345","0.599148958371","0.000050902694","0.569130398173","0.988583531594","0.918717729459","0.952371013508","0.952371013508"
+"GERMAN_RADIXOR","DE_DE","ALL_WORDS","PRIMARY_OUTPUT","54092","296974","1474","296974","0","1","296974","68104","1128969","98192","254903","44095147787","98192","44095245979","0.000223","254903","1383872","18.419550","0.919984419322","0.815804496370","0.999997773184","0.999991992699","0.907901134777","0.897072808397","0.864768082211","0.834709150216","0.761754553110","0.866329859738","0.866325955582","0.000008007301","0.864764092865","0.989946248415","0.975085217969","0.982459538105","0.982459538105"
+"GERMAN_RADIXOR","DE_DE","ALL_WORDS","ANY_CANDIDATE","54092","296974","1474","248400","48574","8","361016","70717","1272705","1375","111167","44095244604","1375","44095245979","0.000003","111167","1383872","8.033041","0.998920789903","0.919669593720","0.999999968818","0.999997447832","0.959834781269","0.981996366774","0.957658377578","0.934497606897","0.918756727140","0.958476435291","0.958475209548","0.000002552168","","","","",""
+"GERMAN_RADIXOR","DE_DE","ALL_WORDS","ALL_CANDIDATES","54092","296974","1474","248400","48574","8","361016","70717","1272705","244817","111167","44095001162","244817","44095245979","0.000555","111167","1383872","8.033041","0.838673179038","0.919669593720","0.999994447996","0.999991927184","0.959832020858","0.853710645080","0.877305874349","0.902242446842","0.781429112618","0.878238135035","0.878234164088","0.000008072816","","","","",""
+"GERMAN_RADIXOR","DE_DE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","16007","150098","228","150098","0","1","150098","17264","814297","47898","59114","11263708444","47898","11263756342","0.000425","59114","873411","6.768177","0.944446441930","0.932318232768","0.999995747600","0.999990500176","0.966156990184","0.941995622128","0.938343149309","0.934718891125","0.883847872972","0.938362743125","0.938357995965","0.000009499824","0.938338399230","0.994062310308","0.990664418294","0.992360455671","0.992360455671"
+"GERMAN_RADIXOR","DE_DE","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","16007","150098","228","135120","14978","8","167157","18366","873411","0","0","11263756342","0","11263756342","0.000000","0","873411","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"GERMAN_RADIXOR","DE_DE","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","16007","150098","228","135120","14978","8","167157","18366","873411","97544","0","11263658798","97544","11263756342","0.000866","0","873411","0.000000","0.899538083639","1.000000000000","0.999991340012","0.999991340683","0.999995670006","0.917982540684","0.947112449481","0.978151677228","0.899538083639","0.948439815507","0.948435708759","0.000008659317","","","","",""
+"HE_IL_RADIXOR","HE_IL","ALL_WORDS","PRIMARY_OUTPUT","2358","58714","0","58714","0","1","58714","2358","688361","25234","19916","1722904030","25234","1722929264","0.001465","19916","708277","2.811894","0.964638205144","0.971881057835","0.999985354013","0.999973805398","0.985933205924","0.966078126522","0.968246086849","0.970423799230","0.938446730860","0.968252859146","0.968239762122","0.000026194602","0.968232984327","0.993166390361","0.993627509527","0.993396896433","0.993396896433"
+"HE_IL_RADIXOR","HE_IL","ALL_WORDS","ANY_CANDIDATE","2358","58714","0","56674","2040","40","62376","2358","708277","0","0","1722929264","0","1722929264","0.000000","0","708277","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"HE_IL_RADIXOR","HE_IL","ALL_WORDS","ALL_CANDIDATES","2358","58714","0","56674","2040","40","62376","2358","708277","86628","0","1722842636","86628","1722929264","0.005028","0","708277","0.000000","0.891020939609","1.000000000000","0.999949720513","0.999949741174","0.999974860256","0.910874182109","0.942370251906","0.976122467036","0.891020939609","0.943939055029","0.943915324345","0.000050258826","","","","",""
+"HE_IL_RADIXOR","HE_IL","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","2358","58714","0","58714","0","1","58714","2358","688361","25234","19916","1722904030","25234","1722929264","0.001465","19916","708277","2.811894","0.964638205144","0.971881057835","0.999985354013","0.999973805398","0.985933205924","0.966078126522","0.968246086849","0.970423799230","0.938446730860","0.968252859146","0.968239762122","0.000026194602","0.968232984327","0.993166390361","0.993627509527","0.993396896433","0.993396896433"
+"HE_IL_RADIXOR","HE_IL","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","2358","58714","0","56674","2040","40","62376","2358","708277","0","0","1722929264","0","1722929264","0.000000","0","708277","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"HE_IL_RADIXOR","HE_IL","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","2358","58714","0","56674","2040","40","62376","2358","708277","86628","0","1722842636","86628","1722929264","0.005028","0","708277","0.000000","0.891020939609","1.000000000000","0.999949720513","0.999949741174","0.999974860256","0.910874182109","0.942370251906","0.976122467036","0.891020939609","0.943939055029","0.943915324345","0.000050258826","","","","",""
+"HUNGARIAN_LUCENE_HUNGARIAN_LIGHT_STEM_FILTER","HU_HU","ALL_WORDS","PRIMARY_OUTPUT","19406","916344","1","916344","0","1","916344","94328","14036270","4132555","8125833","419816410338","4132555","419820542893","0.000984","8125833","22162103","36.665442","0.772546931351","0.633345580968","0.999990156377","0.999970802427","0.816667868673","0.740017627855","0.696054898613","0.657022705053","0.533806904809","0.699492090778","0.699477892426","0.000029197573","0.696040442258","0.982615378770","0.926771756762","0.953876941828","0.953876941828"
+"HUNGARIAN_LUCENE_HUNGARIAN_LIGHT_STEM_FILTER","HU_HU","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","18360","878513","1","878513","0","1","878513","91516","13492703","3639046","7918708","385867055871","3639046","385870694917","0.000943","7918708","21411411","36.983588","0.787584676848","0.630164121365","0.999990569261","0.999970049260","0.815077345313","0.750107959995","0.700134758022","0.656404225003","0.538621031944","0.704491026122","0.704476576404","0.000029950740","0.700119966552","0.983686881315","0.925487153321","0.953699929950","0.953699929950"
+"HUNGARIAN_RADIXOR","HU_HU","ALL_WORDS","PRIMARY_OUTPUT","19406","916344","1","916344","0","1","916344","20535","21962266","272900","199837","419820269993","272900","419820542893","0.000065","199837","22162103","0.901706","0.987726648859","0.990982940563","0.999999349960","0.999998874014","0.995491145262","0.988376194087","0.989352115329","0.990329965723","0.978928596533","0.989353455019","0.989352892139","0.000001125986","0.989351552308","0.998036093538","0.997808712909","0.997922390271","0.997922390271"
+"HUNGARIAN_RADIXOR","HU_HU","ALL_WORDS","ANY_CANDIDATE","19406","916344","1","904024","12320","5","929326","20567","22162103","0","0","419820542893","0","419820542893","0.000000","0","22162103","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"HUNGARIAN_RADIXOR","HU_HU","ALL_WORDS","ALL_CANDIDATES","19406","916344","1","904024","12320","5","929326","20567","22162103","460158","0","419820082735","460158","419820542893","0.000110","0","22162103","0.000000","0.979659062372","1.000000000000","0.999998903917","0.999998903975","0.999999451959","0.983660778882","0.989725029923","0.995864516790","0.979659062372","0.989777279176","0.989776736737","0.000001096025","","","","",""
+"HUNGARIAN_RADIXOR","HU_HU","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","18360","878513","1","878513","0","1","878513","18363","21247134","272775","164277","385870422142","272775","385870694917","0.000071","164277","21411411","0.767240","0.987324528185","0.992327595785","0.999999293092","0.999998867424","0.996163444439","0.988321101757","0.989819739994","0.991322930046","0.979844666523","0.989822900984","0.989822335019","0.000001132576","0.989819173678","0.997945135090","0.998273386381","0.998109233747","0.998109233747"
+"HUNGARIAN_RADIXOR","HU_HU","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","18360","878513","1","867360","11153","5","890245","18375","21411411","0","0","385870694917","0","385870694917","0.000000","0","21411411","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"HUNGARIAN_RADIXOR","HU_HU","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","18360","878513","1","867360","11153","5","890245","18375","21411411","458462","0","385870236455","458462","385870694917","0.000119","0","21411411","0.000000","0.979036823854","1.000000000000","0.999998811877","0.999998811943","0.999999405938","0.983158850285","0.989407384494","0.995735852714","0.979036823854","0.989462896653","0.989462308851","0.000001188057","","","","",""
+"HUNSPELL_CZECH_LUCENE_FILTER","CS_CZ","ALL_WORDS","PRIMARY_OUTPUT","5113","51676","2","51676","0","1","51676","10920","213552","11408","88283","1334865407","11408","1334876815","0.000855","88283","301835","29.248762","0.949288762447","0.707512382593","0.999991453893","0.999925335085","0.853751918243","0.888559718726","0.810759403563","0.745486280807","0.681745481942","0.819532521678","0.819499025505","0.000074664915","0.810722859062","0.995776551361","0.952852006662","0.973841506371","0.973841506371"
+"HUNSPELL_CZECH_LUCENE_FILTER","CS_CZ","ALL_WORDS","ANY_CANDIDATE","5113","51676","2","48359","3317","5","55596","11359","224312","10102","77523","1334866713","10102","1334876815","0.000757","77523","301835","25.683900","0.956905304291","0.743160998559","0.999992432261","0.999934372078","0.871576715410","0.904855299474","0.836596431881","0.777913569166","0.719093919606","0.843288029954","0.843258147533","0.000065627922","","","","",""
+"HUNSPELL_CZECH_LUCENE_FILTER","CS_CZ","ALL_WORDS","ALL_CANDIDATES","5113","51676","2","48359","3317","5","55596","11359","224312","13917","77523","1334862898","13917","1334876815","0.001043","77523","301835","25.683900","0.941581419558","0.743160998559","0.999989574319","0.999931514783","0.871575286439","0.893850652440","0.830686733424","0.775860578084","0.710405634802","0.836508570179","0.836476906392","0.000068485217","","","","",""
+"HUNSPELL_CZECH_LUCENE_FILTER","CS_CZ","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","5038","50968","2","50968","0","1","50968","10816","210827","11239","87986","1298532976","11239","1298544215","0.000866","87986","298813","29.445171","0.949388920411","0.705548286052","0.999991344923","0.999923605087","0.852769815487","0.888008949714","0.809504702628","0.743753342581","0.679973036781","0.818437368155","0.818403143837","0.000076394913","0.809467327086","0.995812473772","0.952393750423","0.973619286126","0.973619286126"
+"HUNSPELL_CZECH_LUCENE_FILTER","CS_CZ","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","5038","50968","2","47731","3237","5","54804","11240","221382","10028","77431","1298534187","10028","1298544215","0.000772","77431","298813","25.912862","0.956665658355","0.740871381098","0.999992277506","0.999932663918","0.870431829302","0.904003665310","0.835052421340","0.775874033233","0.716815448726","0.841882537861","0.841851913927","0.000067336082","","","","",""
+"HUNSPELL_CZECH_LUCENE_FILTER","CS_CZ","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","5038","50968","2","47731","3237","5","54804","11240","221382","13601","77431","1298530614","13601","1298544215","0.001047","77431","298813","25.912862","0.942119217135","0.740871381098","0.999989525963","0.999929913009","0.870430453530","0.893573737936","0.829462940899","0.773935751817","0.708617411512","0.835457458856","0.835425115185","0.000070086991","","","","",""
+"HUNSPELL_DUTCH_LUCENE_FILTER","NL_NL","ALL_WORDS","PRIMARY_OUTPUT","4992","26477","85","26477","0","1","26477","15909","18482","1333","46084","350436627","1333","350437960","0.000380","46084","64566","71.375027","0.932727731517","0.286249728960","0.999996196188","0.999864717095","0.643122962574","0.642512480358","0.438060700869","0.332315636923","0.280459491039","0.516713712165","0.516673857221","0.000135282905","0.438012080403","0.996931617211","0.889026094124","0.939891935467","0.939891935467"
+"HUNSPELL_DUTCH_LUCENE_FILTER","NL_NL","ALL_WORDS","ANY_CANDIDATE","4992","26477","85","25223","1254","3","27763","16027","21374","1164","43192","350436796","1164","350437960","0.000332","43192","64566","66.895889","0.948353891206","0.331041105226","0.999996678442","0.999873450270","0.665518891834","0.690740573172","0.490769654666","0.380588457347","0.325178761600","0.560307166017","0.560267948638","0.000126549730","","","","",""
+"HUNSPELL_DUTCH_LUCENE_FILTER","NL_NL","ALL_WORDS","ALL_CANDIDATES","4992","26477","85","25223","1254","3","27763","16027","21374","1738","43192","350436222","1738","350437960","0.000496","43192","64566","66.895889","0.924800969193","0.331041105226","0.999995040492","0.999871812621","0.665518072859","0.680639942935","0.487556741714","0.379812066416","0.322363658301","0.553305643343","0.553264631551","0.000128187379","","","","",""
+"HUNSPELL_DUTCH_LUCENE_FILTER","NL_NL","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4796","25678","84","25678","0","1","25678","15258","18333","1310","44814","329602546","1310","329603856","0.000397","44814","63147","70.967742","0.933309575930","0.290322580645","0.999996025532","0.999860089122","0.645159303089","0.646808120294","0.442879574828","0.336717714000","0.284422172921","0.520538994337","0.520497519470","0.000139910878","0.442828931093","0.996884174988","0.889060613994","0.939890140871","0.939890140871"
+"HUNSPELL_DUTCH_LUCENE_FILTER","NL_NL","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","4796","25678","84","24492","1186","3","26896","15323","21212","1141","41935","329602715","1141","329603856","0.000346","41935","63147","66.408539","0.948955397486","0.335914611937","0.999996538269","0.999869334815","0.667955575103","0.695206444720","0.496187134503","0.385755489360","0.329952712792","0.564595416287","0.564554662442","0.000130665185","","","","",""
+"HUNSPELL_DUTCH_LUCENE_FILTER","NL_NL","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","4796","25678","84","24492","1186","3","26896","15323","21212","1712","41935","329602144","1712","329603856","0.000519","41935","63147","66.408539","0.925318443553","0.335914611937","0.999994805886","0.999867602764","0.667954708912","0.684951854459","0.492895400309","0.384956009176","0.327047903915","0.557519493726","0.557476858392","0.000132397236","","","","",""
+"HUNSPELL_ENGLISH_LUCENE_FILTER","US_UK","ALL_WORDS","PRIMARY_OUTPUT","396939","607439","250964","607439","0","1","607439","557518","46002","1981986","267868","184488469785","1981986","184490451771","0.001074","267868","313870","85.343614","0.022683566175","0.146563864020","0.999989256972","0.999987805059","0.573276560496","0.027298226808","0.039286754363","0.070050933952","0.020036970960","0.057659267324","0.057655308782","0.000012194941","0.039283923590","0.993096189204","0.963677172906","0.978165531652","0.978165531652"
+"HUNSPELL_ENGLISH_LUCENE_FILTER","US_UK","ALL_WORDS","ANY_CANDIDATE","396939","607439","250964","600602","6837","4","614296","557638","51229","1978852","262641","184488472919","1978852","184490451771","0.001073","262641","313870","83.678274","0.025234953679","0.163217255552","0.999989273960","0.999987850378","0.581603264756","0.030369825498","0.043711664621","0.077960810954","0.022344183028","0.064177721084","0.064173802046","0.000012149622","","","","",""
+"HUNSPELL_ENGLISH_LUCENE_FILTER","US_UK","ALL_WORDS","ALL_CANDIDATES","396939","607439","250964","600602","6837","4","614296","557638","51229","2008917","262641","184488442854","2008917","184490451771","0.001089","262641","313870","83.678274","0.024866684206","0.163217255552","0.999989110997","0.999987687416","0.581603183275","0.029942881217","0.043158091605","0.077253888104","0.022054971033","0.063707707153","0.063703758400","0.000012312584","","","","",""
+"HUNSPELL_ENGLISH_LUCENE_FILTER","US_UK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","374384","583910","228735","583910","0","1","583910","535362","45926","1978041","265965","170472862163","1978041","170474840204","0.001160","265965","311891","85.274984","0.022691081426","0.147250161114","0.999988396874","0.999986836756","0.573619278994","0.027311677226","0.039322595808","0.070190378755","0.020055617372","0.057803679777","0.057799415161","0.000013163244","0.039319549964","0.993066314983","0.962316521867","0.977449637185","0.977449637185"
+"HUNSPELL_ENGLISH_LUCENE_FILTER","US_UK","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","374384","583910","228735","577124","6786","4","590716","535485","51150","1974950","260741","170472865254","1974950","170474840204","0.001158","260741","311891","83.600040","0.025245545630","0.163999602425","0.999988415006","0.999986885532","0.581994008716","0.030387494919","0.043755514884","0.078123472659","0.022367099418","0.064344847861","0.064340626006","0.000013114468","","","","",""
+"HUNSPELL_ENGLISH_LUCENE_FILTER","US_UK","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","374384","583910","228735","577124","6786","4","590716","535485","51150","2004598","260741","170472835606","2004598","170474840204","0.001176","260741","311891","83.600040","0.024881454342","0.163999602425","0.999988241092","0.999986711618","0.581993921758","0.029965261387","0.043207600483","0.077422296168","0.022080830084","0.063879172034","0.063874918547","0.000013288382","","","","",""
+"HUNSPELL_FRENCH_LUCENE_FILTER","FR_FR","ALL_WORDS","PRIMARY_OUTPUT","59240","425210","2301","425210","0","1","425210","154336","3422734","776728","2031881","90395328102","776728","90396104830","0.000859","2031881","5454615","37.250677","0.815041069547","0.627493232795","0.999991407506","0.999968931852","0.813742320150","0.769068574566","0.709075347131","0.657764674673","0.549277098051","0.715145268872","0.715130511338","0.000031068148","0.709060074832","0.978337247291","0.913705954219","0.944917713687","0.944917713687"
+"HUNSPELL_FRENCH_LUCENE_FILTER","FR_FR","ALL_WORDS","ANY_CANDIDATE","59240","425210","2301","411699","13511","4","439015","154718","3610612","745831","1844003","90395358999","745831","90396104830","0.000825","1844003","5454615","33.806291","0.828798173189","0.661937093635","0.999991749302","0.999971351888","0.830964421468","0.789018996925","0.736029080656","0.689708764155","0.582314885091","0.740683639600","0.740669908387","0.000028648112","","","","",""
+"HUNSPELL_FRENCH_LUCENE_FILTER","FR_FR","ALL_WORDS","ALL_CANDIDATES","59240","425210","2301","411699","13511","4","439015","154718","3610612","1043199","1844003","90395061631","1043199","90396104830","0.001154","1844003","5454615","33.806291","0.775839843947","0.661937093635","0.999988459691","0.999968062476","0.830962776663","0.750027659074","0.714376699201","0.681961135862","0.555665643861","0.716629033342","0.716613365354","0.000031937524","","","","",""
+"HUNSPELL_FRENCH_LUCENE_FILTER","FR_FR","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","57698","421231","2133","421231","0","1","421231","153822","3412548","763305","2028011","88711363201","763305","88712126506","0.000860","2028011","5440559","37.275784","0.817209801207","0.627242163903","0.999991395708","0.999968537054","0.813616779806","0.770536594362","0.709734150326","0.657825640123","0.550068151075","0.715952822518","0.715937898033","0.000031462946","0.709718690125","0.979328164393","0.913161860024","0.945088340370","0.945088340370"
+"HUNSPELL_FRENCH_LUCENE_FILTER","FR_FR","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","57698","421231","2133","407794","13437","4","434961","154205","3600083","733584","1840476","88711392922","733584","88712126506","0.000827","1840476","5440559","33.828803","0.830724418835","0.661711967465","0.999991730736","0.999970985904","0.830851849101","0.790350629656","0.736648201095","0.689779349655","0.583090317150","0.741417756470","0.741403865778","0.000029014096","","","","",""
+"HUNSPELL_FRENCH_LUCENE_FILTER","FR_FR","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","57698","421231","2133","407794","13437","4","434961","154205","3600083","1027635","1840476","88711098871","1027635","88712126506","0.001158","1840476","5440559","33.828803","0.777939148410","0.661711967465","0.999988416071","0.999967671442","0.830850191768","0.751538185756","0.715133880405","0.682093458746","0.556582409247","0.717475884237","0.717460037184","0.000032328558","","","","",""
+"HUNSPELL_GERMAN_LUCENE_FILTER","DE_DE","ALL_WORDS","PRIMARY_OUTPUT","54092","296974","1474","296974","0","1","296974","182774","391862","203883","992010","44095042096","203883","44095245979","0.000462","992010","1383872","71.683653","0.657768004767","0.283163471766","0.999995376304","0.999972880172","0.641579424035","0.520145203475","0.395896782054","0.319562150060","0.246802560848","0.431573715426","0.431562811676","0.000027119828","0.395885371175","0.980462638581","0.886872825924","0.931322397630","0.931322397630"
+"HUNSPELL_GERMAN_LUCENE_FILTER","DE_DE","ALL_WORDS","ANY_CANDIDATE","54092","296974","1474","289083","7891","3","305052","183111","408175","158403","975697","44095087576","158403","44095245979","0.000359","975697","1383872","70.504859","0.720421548313","0.294951411691","0.999996407708","0.999974281481","0.647473909700","0.559115650060","0.418544438463","0.334456395588","0.264657729653","0.460965674088","0.460955788036","0.000025718519","","","","",""
+"HUNSPELL_GERMAN_LUCENE_FILTER","DE_DE","ALL_WORDS","ALL_CANDIDATES","54092","296974","1474","289083","7891","3","305052","183111","408175","242551","975697","44095003428","242551","44095245979","0.000550","975697","1383872","70.504859","0.627260936247","0.294951411691","0.999994499384","0.999972373218","0.647472955538","0.511911128190","0.401234052132","0.329906951166","0.250964847398","0.430129630047","0.430118032816","0.000027626782","","","","",""
+"HUNSPELL_GERMAN_LUCENE_FILTER","DE_DE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","16007","150098","228","150098","0","1","150098","86983","278093","84679","595318","11263671663","84679","11263756342","0.000752","595318","873411","68.160122","0.766577905682","0.318398783620","0.999992482170","0.999939634323","0.659195632895","0.598178360154","0.449922058465","0.360558871242","0.290257700216","0.494041974653","0.494019111671","0.000060365677","0.449897024669","0.988041339480","0.865580709092","0.922765807515","0.922765807515"
+"HUNSPELL_GERMAN_LUCENE_FILTER","DE_DE","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","16007","150098","228","145109","4989","3","155207","87393","288864","60996","584547","11263695346","60996","11263756342","0.000542","584547","873411","66.926911","0.825655976676","0.330730893016","0.999994584755","0.999942692923","0.665362738886","0.635466205220","0.472281285177","0.375782098835","0.309141519702","0.522560942370","0.522540242219","0.000057307077","","","","",""
+"HUNSPELL_GERMAN_LUCENE_FILTER","DE_DE","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","16007","150098","228","145109","4989","3","155207","87393","288864","96545","584547","11263659797","96545","11263756342","0.000857","584547","873411","66.926911","0.749499881944","0.330730893016","0.999991428703","0.999939537116","0.665361160860","0.598050472724","0.458944090497","0.372338300095","0.297811447117","0.497878263505","0.497854575726","0.000060462884","","","","",""
+"HUNSPELL_POLISH_LUCENE_FILTER","PL_PL","ALL_WORDS","PRIMARY_OUTPUT","9990","122341","1","122341","0","1","122341","18419","971262","52652","149705","7482425351","52652","7482478003","0.000704","149705","1120967","13.354987","0.948577712581","0.866450127435","0.999992963294","0.999972959935","0.933221545364","0.930929837176","0.905655838249","0.881717903868","0.827578626454","0.906584403102","0.906571161039","0.000027040065","0.905642343969","0.994545991966","0.970520439141","0.982386343372","0.982386343372"
+"HUNSPELL_POLISH_LUCENE_FILTER","PL_PL","ALL_WORDS","ANY_CANDIDATE","9990","122341","1","110894","11447","6","135231","19068","1040224","42213","80743","7482435790","42213","7482478003","0.000564","80743","1120967","7.202977","0.961001887408","0.927970225707","0.999994358420","0.999983569937","0.963982292063","0.954208759768","0.944197251162","0.934393641743","0.894293230626","0.944341642819","0.944333470354","0.000016430063","","","","",""
+"HUNSPELL_POLISH_LUCENE_FILTER","PL_PL","ALL_WORDS","ALL_CANDIDATES","9990","122341","1","110894","11447","6","135231","19068","1040224","82745","80743","7482395258","82745","7482478003","0.001106","80743","1120967","7.202977","0.926315864463","0.927970225707","0.999988941498","0.999978153827","0.963979583602","0.926646264647","0.927142307089","0.927638880888","0.864180136112","0.927142676087","0.927131751477","0.000021846173","","","","",""
+"HUNSPELL_POLISH_LUCENE_FILTER","PL_PL","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","9846","120925","1","120925","0","1","120925","18149","965984","51950","148667","7310200749","51950","7310252699","0.000711","148667","1114651","13.337538","0.948965257080","0.866624620621","0.999992893543","0.999972560946","0.933308757082","0.931268723294","0.905927782480","0.881929423296","0.828032892137","0.906860880124","0.906847444801","0.000027439054","0.905914089179","0.994583905165","0.970514203019","0.982401644006","0.982401644006"
+"HUNSPELL_POLISH_LUCENE_FILTER","PL_PL","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","9846","120925","1","109660","11265","6","133595","18789","1034283","41671","80368","7310211028","41671","7310252699","0.000570","80368","1114651","7.210149","0.961270649117","0.927898508143","0.999994299650","0.999983308321","0.963946403896","0.954405554191","0.944289819479","0.934386268967","0.894459328803","0.944437187555","0.944428885928","0.000016691679","","","","",""
+"HUNSPELL_POLISH_LUCENE_FILTER","PL_PL","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","9846","120925","1","109660","11265","6","133595","18789","1034283","81865","80368","7310170834","81865","7310252699","0.001120","80368","1114651","7.210149","0.926653992123","0.927898508143","0.999988801345","0.999977810854","0.963943654744","0.926902628188","0.927275832560","0.927649337585","0.864412176686","0.927276041347","0.927264945147","0.000022189146","","","","",""
+"HUNSPELL_SPANISH_LUCENE_FILTER","ES_ES","ALL_WORDS","PRIMARY_OUTPUT","65059","871332","3589","871332","0","1","871332","495840","9662476","536192","32310860","379566781918","536192","379567318110","0.000141","32310860","41973336","76.979490","0.947425291224","0.230205099733","0.999998587360","0.999913471422","0.615101843546","0.583708381625","0.370408466579","0.271277635967","0.227301418167","0.467014061518","0.466991649518","0.000086528578","0.370381248953","0.993314263125","0.790558492734","0.880413722434","0.880413722434"
+"HUNSPELL_SPANISH_LUCENE_FILTER","ES_ES","ALL_WORDS","ANY_CANDIDATE","65059","871332","3589","853455","17877","5","890999","496361","10079118","416345","31894218","379566901765","416345","379567318110","0.000110","31894218","41973336","75.986855","0.960330954432","0.240131449166","0.999998903106","0.999914884689","0.620065176136","0.600267728541","0.384194728757","0.282504215637","0.237772914592","0.480214185303","0.480192080762","0.000085115311","","","","",""
+"HUNSPELL_SPANISH_LUCENE_FILTER","ES_ES","ALL_WORDS","ALL_CANDIDATES","65059","871332","3589","853455","17877","5","890999","496361","10079118","888077","31894218","379566430033","888077","379567318110","0.000234","31894218","41973336","75.986855","0.919024235459","0.240131449166","0.999997660291","0.999913642011","0.620064554728","0.587073016700","0.380771322449","0.281759130783","0.235155989841","0.469772946730","0.469749183448","0.000086357989","","","","",""
+"HUNSPELL_SPANISH_LUCENE_FILTER","ES_ES","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","64918","869371","3525","869371","0","1","869371","495045","9628515","531181","32234855","377860138584","531181","377860669765","0.000141","32234855","41863370","77.000144","0.947716841134","0.229998564377","0.999998594241","0.999913295008","0.614998579309","0.583531128169","0.370163304101","0.271052948234","0.227116805648","0.466876335765","0.466853897372","0.000086704992","0.370136051001","0.993362468962","0.790499503024","0.880396073652","0.880396073652"
+"HUNSPELL_SPANISH_LUCENE_FILTER","ES_ES","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","64918","869371","3525","851564","17807","5","888962","495572","10041262","412198","31822108","377860257567","412198","377860669765","0.000109","31822108","41863370","76.014205","0.960568271175","0.239857947413","0.999998909127","0.999914702064","0.619928428270","0.599999808789","0.383863548307","0.282205460900","0.237519268813","0.479999931119","0.479977799265","0.000085297936","","","","",""
+"HUNSPELL_SPANISH_LUCENE_FILTER","ES_ES","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","64918","869371","3525","851564","17807","5","888962","495572","10041262","878949","31822108","377859790816","878949","377860669765","0.000233","31822108","41863370","76.014205","0.919511720057","0.239857947413","0.999997673881","0.999913466955","0.619927810647","0.586904802235","0.380469146267","0.281467012980","0.234925531298","0.469629847641","0.469606065497","0.000086533045","","","","",""
+"HUNSPELL_UKRAINIAN_LUCENE_FILTER","UK_UA","ALL_WORDS","PRIMARY_OUTPUT","1493","14245","4","14245","0","1","14245","3137","50416","794","14924","101386756","794","101387550","0.000783","14924","65340","22.840526","0.984495215778","0.771594735231","0.999992168664","0.999845070949","0.885793451947","0.933007624547","0.865139425139","0.806475349522","0.762331024889","0.871568313648","0.871498740996","0.000154929051","0.865063055969","0.998114340300","0.949803904722","0.973360047526","0.973360047526"
+"HUNSPELL_UKRAINIAN_LUCENE_FILTER","UK_UA","ALL_WORDS","ANY_CANDIDATE","1493","14245","4","12923","1322","6","15740","3311","55875","326","9465","101387224","326","101387550","0.000322","9465","65340","14.485767","0.994199391470","0.855142332415","0.999996784615","0.999903492153","0.927569558515","0.962883947281","0.919442821764","0.879752236578","0.850896963421","0.922053136488","0.922008115880","0.000096507847","","","","",""
+"HUNSPELL_UKRAINIAN_LUCENE_FILTER","UK_UA","ALL_WORDS","ALL_CANDIDATES","1493","14245","4","12923","1322","6","15740","3311","55875","1271","9465","101386279","1271","101387550","0.001254","9465","65340","14.485767","0.977758723270","0.855142332415","0.999987463944","0.999894177485","0.927564898180","0.950500809733","0.912349166435","0.877142031861","0.838825419225","0.914397547654","0.914347195556","0.000105822515","","","","",""
+"HUNSPELL_UKRAINIAN_LUCENE_FILTER","UK_UA","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","1491","14236","4","14236","0","1","14236","3134","50404","794","14920","101258612","794","101259406","0.000784","14920","65324","22.839998","0.984491581702","0.771600024493","0.999992158753","0.999844914465","0.885796091623","0.933006560145","0.865141346698","0.806479484406","0.762334008893","0.871569692311","0.871500048785","0.000155085535","0.865064900251","0.998112856419","0.949788155938","0.973351072054","0.973351072054"
+"HUNSPELL_UKRAINIAN_LUCENE_FILTER","UK_UA","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","1491","14236","4","12915","1321","6","15730","3308","55859","326","9465","101259080","326","101259406","0.000322","9465","65324","14.489315","0.994197739610","0.855106851999","0.999996780546","0.999903370085","0.927551816273","0.962873710629","0.919421606630","0.879721936116","0.850860624524","0.922033242016","0.921988165267","0.000096629915","","","","",""
+"HUNSPELL_UKRAINIAN_LUCENE_FILTER","UK_UA","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","1491","14236","4","12915","1321","6","15730","3308","55859","1271","9465","101258135","1271","101259406","0.001255","9465","65324","14.489315","0.977752494311","0.855106851999","0.999987448080","0.999894043636","0.927547150039","0.950487333415","0.912326261290","0.877111165546","0.838786695698","0.914375665383","0.914325250217","0.000105956364","","","","",""
+"ITALIAN_LUCENE_ITALIAN_LIGHT_STEM_FILTER","IT_IT","ALL_WORDS","PRIMARY_OUTPUT","10009","327551","0","327551","0","1","327551","244870","109684","10589","6034130","53638510622","10589","53638521211","0.000020","6034130","6143814","98.214725","0.911958627456","0.017852754006","0.999999802586","0.999887319289","0.508926278296","0.082781551919","0.035019947839","0.022207258650","0.017822037328","0.127596916262","0.127588341500","0.000112680711","0.035015703871","0.997481424185","0.737537113266","0.848036553212","0.848036553212"
+"ITALIAN_LUCENE_ITALIAN_LIGHT_STEM_FILTER","IT_IT","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","10007","327469","0","327469","0","1","327469","244808","109658","10588","6032516","53611656484","10588","53611667072","0.000020","6032516","6142174","98.214671","0.911947174958","0.017853287777","0.999999802506","0.999887292971","0.508926545141","0.082783771729","0.035020966336","0.022207918023","0.017822564890","0.127598022525","0.127589445488","0.000112707029","0.035016721203","0.997481197125","0.737534145120","0.848034509070","0.848034509070"
+"ITALIAN_RADIXOR","IT_IT","ALL_WORDS","PRIMARY_OUTPUT","10009","327551","0","327551","0","1","327551","10010","6100906","124172","42908","53638397039","124172","53638521211","0.000231","42908","6143814","0.698394","0.980052940702","0.993016064614","0.999997685022","0.999996885431","0.996506874818","0.982618418699","0.986491918597","0.990396078172","0.973343909830","0.986513210398","0.986511657877","0.000003114569","0.986490361200","0.995780270704","0.997112550353","0.996445965204","0.996445965204"
+"ITALIAN_RADIXOR","IT_IT","ALL_WORDS","ANY_CANDIDATE","10009","327551","0","321297","6254","4","334175","10012","6143734","0","80","53638521211","0","53638521211","0.000000","80","6143814","0.001302","1.000000000000","0.999986978772","1.000000000000","0.999999998509","0.999993489386","0.999997395727","0.999993489344","0.999989582991","0.999986978772","0.999993489365","0.999993488619","0.000000001491","","","","",""
+"ITALIAN_RADIXOR","IT_IT","ALL_WORDS","ALL_CANDIDATES","10009","327551","0","321297","6254","4","334175","10012","6143734","170950","80","53638350261","170950","53638521211","0.000319","80","6143814","0.001302","0.972928178195","0.999986978772","0.999996812925","0.999996811799","0.999991895849","0.978222150749","0.986272020913","0.994455476443","0.972915852437","0.986364795335","0.986363222748","0.000003188201","","","","",""
+"ITALIAN_RADIXOR","IT_IT","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","10007","327469","0","327469","0","1","327469","10007","6099346","124171","42828","53611542901","124171","53611667072","0.000232","42828","6142174","0.697278","0.980048098206","0.993027224563","0.999997683881","0.999996885382","0.996512454222","0.982616709845","0.986494972258","0.990403969991","0.973349855458","0.986516316590","0.986514764059","0.000003114618","0.986493414837","0.995779575755","0.997114909855","0.996446795437","0.996446795437"
+"ITALIAN_RADIXOR","IT_IT","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","10007","327469","0","321217","6252","4","334089","10007","6142174","0","0","53611667072","0","53611667072","0.000000","0","6142174","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"ITALIAN_RADIXOR","IT_IT","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","10007","327469","0","321217","6252","4","334089","10007","6142174","170949","0","53611496123","170949","53611667072","0.000319","0","6142174","0.000000","0.972921642743","1.000000000000","0.999996811347","0.999996811712","0.999998405674","0.978219357390","0.986274996092","0.994464412864","0.972921642743","0.986367904356","0.986366331762","0.000003188288","","","","",""
+"NL_NL_RADIXOR","NL_NL","ALL_WORDS","PRIMARY_OUTPUT","4992","26477","85","26477","0","1","26477","5015","63102","1214","1464","350436746","1214","350437960","0.000346","1464","64566","2.267447","0.981124448038","0.977325527367","0.999996535763","0.999992359542","0.988661031565","0.980362303079","0.979221303208","0.978082956166","0.959288537549","0.979223145453","0.979219325206","0.000007640458","0.979217482290","0.997464133435","0.997003025118","0.997233525974","0.997233525974"
+"NL_NL_RADIXOR","NL_NL","ALL_WORDS","ANY_CANDIDATE","4992","26477","85","25905","572","3","27061","5016","64566","0","0","350437960","0","350437960","0.000000","0","64566","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"NL_NL_RADIXOR","NL_NL","ALL_WORDS","ALL_CANDIDATES","4992","26477","85","25905","572","3","27061","5016","64566","2651","0","350435309","2651","350437960","0.000756","0","64566","0.000000","0.960560572474","1.000000000000","0.999992435180","0.999992436574","0.999996217590","0.968197604323","0.979883596519","0.991855131329","0.960560572474","0.980081921308","0.980078214229","0.000007563426","","","","",""
+"NL_NL_RADIXOR","NL_NL","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4796","25678","84","25678","0","1","25678","4797","61763","1214","1384","329602642","1214","329603856","0.000368","1384","63147","2.191711","0.980723121139","0.978082885964","0.999996316791","0.999992119320","0.989039601378","0.980193934392","0.979401224192","0.978609795129","0.959633939808","0.979402113872","0.979398173122","0.000007880680","0.979397283106","0.997373193672","0.997139403762","0.997256285015","0.997256285015"
+"NL_NL_RADIXOR","NL_NL","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","4796","25678","84","25129","549","3","26239","4797","63147","0","0","329603856","0","329603856","0.000000","0","63147","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"NL_NL_RADIXOR","NL_NL","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","4796","25678","84","25129","549","3","26239","4797","63147","2651","0","329601205","2651","329603856","0.000804","0","63147","0.000000","0.959710021581","1.000000000000","0.999991957012","0.999991958552","0.999995978506","0.967506182222","0.979440846873","0.991673628866","0.959710021581","0.979647906945","0.979643967288","0.000008041448","","","","",""
+"NN_NO_RADIXOR","NN_NO","ALL_WORDS","PRIMARY_OUTPUT","4688","18250","23","18250","0","1","18250","4680","26716","6230","3936","166485243","6230","166491473","0.003742","3936","30652","12.840924","0.810902689249","0.871590760799","0.999962580666","0.999938951055","0.935776670732","0.822354650447","0.840152206044","0.858737158800","0.724364188493","0.840699287413","0.840668985911","0.000061048945","0.840121715471","0.983845117159","0.986801676045","0.985321178741","0.985321178741"
+"NN_NO_RADIXOR","NN_NO","ALL_WORDS","ANY_CANDIDATE","4688","18250","23","15846","2404","5","21513","4693","30652","0","0","166491473","0","166491473","0.000000","0","30652","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"NN_NO_RADIXOR","NN_NO","ALL_WORDS","ALL_CANDIDATES","4688","18250","23","15846","2404","5","21513","4693","30652","13214","0","166478259","13214","166491473","0.007937","0","30652","0.000000","0.698764418912","1.000000000000","0.999920632572","0.999920647181","0.999960316286","0.743561877778","0.822673716418","0.920624241623","0.698764418912","0.835921299473","0.835888126353","0.000079352819","","","","",""
+"NN_NO_RADIXOR","NN_NO","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4681","18219","23","18219","0","1","18219","4668","26671","6230","3924","165920046","6230","165926276","0.003755","3924","30595","12.825625","0.810644053372","0.871743748979","0.999962453204","0.999938815429","0.935853101091","0.822169063928","0.840084414766","0.858797921188","0.724263408011","0.840638974932","0.840608609146","0.000061184571","0.840053856992","0.983814668550","0.986841730061","0.985325874420","0.985325874420"
+"NN_NO_RADIXOR","NN_NO","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","4681","18219","23","15820","2399","5","21477","4681","30595","0","0","165926276","0","165926276","0.000000","0","30595","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"NN_NO_RADIXOR","NN_NO","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","4681","18219","23","15820","2399","5","21477","4681","30595","13214","0","165913062","13214","165926276","0.007964","0","30595","0.000000","0.698372480541","1.000000000000","0.999920362222","0.999920376903","0.999960181111","0.743206805583","0.822402021397","0.920488118949","0.698372480541","0.835686831618","0.835653554835","0.000079623097","","","","",""
+"NORWEGIAN_BOKMAL_LUCENE_NORWEGIAN_LIGHT_STEM_FILTER","NB_NO","ALL_WORDS","PRIMARY_OUTPUT","17929","75310","252","75310","0","1","75310","25999","99529","25171","42651","2835593044","25171","2835618215","0.000888","42651","142180","29.997890","0.798147554130","0.700021100014","0.999991123276","0.999976083311","0.850006111645","0.776381478361","0.745870803357","0.717667503101","0.594732030284","0.747475838282","0.747464055956","0.000023916689","0.745858895755","0.987774378220","0.965622291196","0.976572729149","0.976572729149"
+"NORWEGIAN_BOKMAL_LUCENE_NORWEGIAN_LIGHT_STEM_FILTER","NB_NO","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","17914","75251","252","75251","0","1","75251","25985","99450","25118","42641","2831151666","25118","2831176784","0.000887","42641","142091","30.009642","0.798359129150","0.699903582915","0.999991128071","0.999976068044","0.849947355493","0.776512696705","0.745896444523","0.717602881668","0.594764635875","0.747512150366","0.747500361706","0.000023931956","0.745884529658","0.987789184092","0.965602788874","0.976569991255","0.976569991255"
+"NORWEGIAN_BOKMAL_LUCENE_NORWEGIAN_MINIMAL_STEM_FILTER","NB_NO","ALL_WORDS","PRIMARY_OUTPUT","17929","75310","252","75310","0","1","75310","27457","94526","14772","47654","2835603443","14772","2835618215","0.000521","47654","142180","33.516669","0.864846566268","0.664833309889","0.999994790554","0.999977986151","0.832414050221","0.815762584315","0.751763573752","0.697075888841","0.602260563739","0.758273568838","0.758263230800","0.000022013849","0.751752754540","0.992088894987","0.962515968891","0.977078714611","0.977078714611"
+"NORWEGIAN_BOKMAL_LUCENE_NORWEGIAN_MINIMAL_STEM_FILTER","NB_NO","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","17914","75251","252","75251","0","1","75251","27443","94447","14719","47644","2831162065","14719","2831176784","0.000520","47644","142091","33.530625","0.865168642251","0.664693752595","0.999994801102","0.999977973869","0.832344276848","0.815949754214","0.751795969864","0.696994967013","0.602302149098","0.758335144541","0.758324803793","0.000022026131","0.751785145440","0.992107474043","0.962494026490","0.977076419047","0.977076419047"
+"NORWEGIAN_BOKMAL_RADIXOR","NB_NO","ALL_WORDS","PRIMARY_OUTPUT","17929","75310","252","75310","0","1","75310","17886","135010","11482","7170","2835606733","11482","2835618215","0.000405","7170","142180","5.042903","0.921620293258","0.949570966381","0.999995950795","0.999993422575","0.974783458588","0.927078011613","0.935386875069","0.943846020481","0.878616704195","0.935491246621","0.935487968734","0.000006577425","0.935383586923","0.993354053349","0.994615320153","0.993984286646","0.993984286646"
+"NORWEGIAN_BOKMAL_RADIXOR","NB_NO","ALL_WORDS","ANY_CANDIDATE","17929","75310","252","71073","4237","9","79825","17962","142180","0","0","2835618215","0","2835618215","0.000000","0","142180","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"NORWEGIAN_BOKMAL_RADIXOR","NB_NO","ALL_WORDS","ALL_CANDIDATES","17929","75310","252","71073","4237","9","79825","17962","142180","20161","0","2835598054","20161","2835618215","0.000711","0","142180","0.000000","0.875810793330","1.000000000000","0.999992890087","0.999992890443","0.999996445043","0.898118108406","0.933794385280","0.972422273928","0.875810793330","0.935847633608","0.935844306704","0.000007109557","","","","",""
+"NORWEGIAN_BOKMAL_RADIXOR","NB_NO","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","17914","75251","252","75251","0","1","75251","17838","134987","11482","7104","2831165302","11482","2831176784","0.000406","7104","142091","4.999613","0.921607985307","0.950003870759","0.999995944443","0.999993435568","0.974999907601","0.927150543912","0.935590518436","0.944185565020","0.878976122105","0.935698217036","0.935694946007","0.000006564432","0.935587236809","0.993348241255","0.994693619298","0.994020475044","0.994020475044"
+"NORWEGIAN_BOKMAL_RADIXOR","NB_NO","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","17914","75251","252","71047","4204","9","79733","17914","142091","0","0","2831176784","0","2831176784","0.000000","0","142091","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"NORWEGIAN_BOKMAL_RADIXOR","NB_NO","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","17914","75251","252","71047","4204","9","79733","17914","142091","20161","0","2831156623","20161","2831176784","0.000712","0","142091","0.000000","0.875742671893","1.000000000000","0.999992878933","0.999992879290","0.999996439466","0.898060798964","0.933755663840","0.972405477022","0.875742671893","0.935811237319","0.935807905326","0.000007120710","","","","",""
+"PERSIAN_LUCENE_PERSIAN_STEM_FILTER","FA_IR","ALL_WORDS","PRIMARY_OUTPUT","69","3701","0","3701","0","1","3701","3190","430","179","98018","6748223","179","6748402","0.002652","98018","98448","99.563221","0.706075533662","0.004367788071","0.999973475202","0.985658076342","0.502170631636","0.021311605408","0.008681870034","0.005451304637","0.004359860890","0.055533668104","0.054800599292","0.014341923658","0.008506575635","0.985685778881","0.520346904570","0.681125382676","0.681125382676"
+"PERSIAN_LUCENE_PERSIAN_STEM_FILTER","FA_IR","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","69","3701","0","3701","0","1","3701","3190","430","179","98018","6748223","179","6748402","0.002652","98018","98448","99.563221","0.706075533662","0.004367788071","0.999973475202","0.985658076342","0.502170631636","0.021311605408","0.008681870034","0.005451304637","0.004359860890","0.055533668104","0.054800599292","0.014341923658","0.008506575635","0.985685778881","0.520346904570","0.681125382676","0.681125382676"
+"PERSIAN_RADIXOR","FA_IR","ALL_WORDS","PRIMARY_OUTPUT","69","3701","0","3701","0","1","3701","69","93636","8621","4812","6739781","8621","6748402","0.127749","4812","98448","4.887860","0.915692813206","0.951121404193","0.998722512381","0.998038075904","0.974921958287","0.922565796215","0.933070924989","0.943818050233","0.874538848780","0.933239001706","0.932248664283","0.001961924096","0.932075735269","0.980342969277","0.984586447427","0.982460126226","0.982460126226"
+"PERSIAN_RADIXOR","FA_IR","ALL_WORDS","ANY_CANDIDATE","69","3701","0","3387","314","2","4015","69","98448","0","0","6748402","0","6748402","0.000000","0","98448","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"PERSIAN_RADIXOR","FA_IR","ALL_WORDS","ALL_CANDIDATES","69","3701","0","3387","314","2","4015","69","98448","13433","0","6734969","13433","6748402","0.199055","0","98448","0.000000","0.879934930864","1.000000000000","0.998009454683","0.998038075904","0.999004727341","0.901584696651","0.936133391021","0.973435401930","0.879934930864","0.938048469358","0.937114390300","0.001961924096","","","","",""
+"PERSIAN_RADIXOR","FA_IR","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","69","3701","0","3701","0","1","3701","69","93636","8621","4812","6739781","8621","6748402","0.127749","4812","98448","4.887860","0.915692813206","0.951121404193","0.998722512381","0.998038075904","0.974921958287","0.922565796215","0.933070924989","0.943818050233","0.874538848780","0.933239001706","0.932248664283","0.001961924096","0.932075735269","0.980342969277","0.984586447427","0.982460126226","0.982460126226"
+"PERSIAN_RADIXOR","FA_IR","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","69","3701","0","3387","314","2","4015","69","98448","0","0","6748402","0","6748402","0.000000","0","98448","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"PERSIAN_RADIXOR","FA_IR","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","69","3701","0","3387","314","2","4015","69","98448","13433","0","6734969","13433","6748402","0.199055","0","98448","0.000000","0.879934930864","1.000000000000","0.998009454683","0.998038075904","0.999004727341","0.901584696651","0.936133391021","0.973435401930","0.879934930864","0.938048469358","0.937114390300","0.001961924096","","","","",""
+"POLISH_LUCENE_MORFOLOGIK_FILTER","PL_PL","ALL_WORDS","PRIMARY_OUTPUT","9990","122341","1","122341","0","1","122341","15519","1004747","99228","116220","7482378775","99228","7482478003","0.001326","116220","1120967","10.367834","0.910117529835","0.896321657997","0.999986738618","0.999971210643","0.948154198307","0.907324485129","0.903166914014","0.899047271013","0.823431500703","0.903193253581","0.903178865015","0.000028789357","0.903152518035","0.990022261217","0.977053921984","0.983495343428","0.983495343428"
+"POLISH_LUCENE_MORFOLOGIK_FILTER","PL_PL","ALL_WORDS","ANY_CANDIDATE","9990","122341","1","109468","12873","5","136636","16295","1093112","85532","27855","7482392471","85532","7482478003","0.001143","27855","1120967","2.484908","0.927431862377","0.975150918805","0.999988569028","0.999984848600","0.987569743916","0.936598359399","0.950692965028","0.965218263555","0.906019814355","0.950992130738","0.950984648194","0.000015151400","","","","",""
+"POLISH_LUCENE_MORFOLOGIK_FILTER","PL_PL","ALL_WORDS","ALL_CANDIDATES","9990","122341","1","109468","12873","5","136636","16295","1093112","143096","27855","7482334907","143096","7482478003","0.001912","27855","1120967","2.484908","0.884246016852","0.975150918805","0.999980875854","0.999977156579","0.987565897330","0.901045352805","0.927476322293","0.955504786999","0.864760696263","0.928586730350","0.928575670129","0.000022843421","","","","",""
+"POLISH_LUCENE_MORFOLOGIK_FILTER","PL_PL","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","9846","120925","1","120925","0","1","120925","15277","999138","99224","115513","7310153475","99224","7310252699","0.001357","115513","1114651","10.363154","0.909661841906","0.896368459724","0.999986426735","0.999970629707","0.948177443229","0.906971715650","0.902966227492","0.898995962905","0.823097930182","0.902990688822","0.902976009256","0.000029370293","0.902951540918","0.989889334726","0.977011745487","0.983408384376","0.983408384376"
+"POLISH_LUCENE_MORFOLOGIK_FILTER","PL_PL","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","9846","120925","1","108162","12763","5","135105","16044","1087157","85532","27494","7310167167","85532","7310252699","0.001170","27494","1114651","2.466602","0.927063356099","0.975333983462","0.999988299720","0.999984541059","0.987661141591","0.936331423447","0.950586270515","0.965281863330","0.905826028197","0.950892420848","0.950884788442","0.000015458941","","","","",""
+"POLISH_LUCENE_MORFOLOGIK_FILTER","PL_PL","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","9846","120925","1","108162","12763","5","135105","16044","1087157","143085","27494","7310109614","143085","7310252699","0.001957","27494","1114651","2.466602","0.883693614752","0.975333983462","0.999980426805","0.999976669344","0.987657205134","0.900617649988","0.927255102898","0.955516285728","0.864376148890","0.928383764096","0.928372472930","0.000023330656","","","","",""
+"POLISH_LUCENE_STEMPEL_DIRECT","PL_PL","ALL_WORDS","PRIMARY_OUTPUT","9990","122341","1","122341","0","1","122341","31432","797573","66669","323394","7482411334","66669","7482478003","0.000891","323394","1120967","28.849556","0.922858412343","0.711504442147","0.999991089984","0.999947877619","0.855747766065","0.871105640425","0.803515398127","0.745658746735","0.671563509358","0.810319603524","0.810295555502","0.000052122381","0.803489769425","0.991766794523","0.931068514076","0.960459620808","0.960459620808"
+"POLISH_LUCENE_STEMPEL_DIRECT","PL_PL","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","9846","120925","1","120925","0","1","120925","30830","794493","66274","320158","7310186425","66274","7310252699","0.000907","320158","1114651","28.722712","0.923005877316","0.712772876892","0.999990934103","0.999947146412","0.856381905497","0.871590591697","0.804379630033","0.746792242917","0.672771767894","0.811106376847","0.811081975971","0.000052853588","0.804353636298","0.991711086900","0.931317880193","0.960566151650","0.960566151650"
+"POLISH_LUCENE_STEMPEL_FILTER","PL_PL","ALL_WORDS","PRIMARY_OUTPUT","9990","122341","1","122341","0","1","122341","31432","797573","66669","323394","7482411334","66669","7482478003","0.000891","323394","1120967","28.849556","0.922858412343","0.711504442147","0.999991089984","0.999947877619","0.855747766065","0.871105640425","0.803515398127","0.745658746735","0.671563509358","0.810319603524","0.810295555502","0.000052122381","0.803489769425","0.991766794523","0.931068514076","0.960459620808","0.960459620808"
+"POLISH_LUCENE_STEMPEL_FILTER","PL_PL","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","9846","120925","1","120925","0","1","120925","30830","794493","66274","320158","7310186425","66274","7310252699","0.000907","320158","1114651","28.722712","0.923005877316","0.712772876892","0.999990934103","0.999947146412","0.856381905497","0.871590591697","0.804379630033","0.746792242917","0.672771767894","0.811106376847","0.811081975971","0.000052853588","0.804353636298","0.991711086900","0.931317880193","0.960566151650","0.960566151650"
+"POLISH_RADIXOR","PL_PL","ALL_WORDS","PRIMARY_OUTPUT","9990","122341","1","122341","0","1","122341","10074","1099420","13669","21547","7482464334","13669","7482478003","0.000183","21547","1120967","1.922180","0.987719760055","0.980778203105","0.999998173199","0.999995294243","0.990388188152","0.986323599045","0.984236742499","0.982158698021","0.968962733423","0.984242862020","0.984240510632","0.000004705757","0.984234389298","0.996967243455","0.996469409869","0.996718264498","0.996718264498"
+"POLISH_RADIXOR","PL_PL","ALL_WORDS","ANY_CANDIDATE","9990","122341","1","119475","2866","4","125778","10079","1120967","0","0","7482478003","0","7482478003","0.000000","0","1120967","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"POLISH_RADIXOR","PL_PL","ALL_WORDS","ALL_CANDIDATES","9990","122341","1","119475","2866","4","125778","10079","1120967","38073","0","7482439930","38073","7482478003","0.000509","0","1120967","0.000000","0.967151263114","1.000000000000","0.999994911712","0.999994912475","0.999997455856","0.973547222425","0.983301367057","0.993252946885","0.967151263114","0.983438489746","0.983435987734","0.000005087525","","","","",""
+"POLISH_RADIXOR","PL_PL","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","9846","120925","1","120925","0","1","120925","9844","1093651","13669","21000","7310239030","13669","7310252699","0.000187","21000","1114651","1.883998","0.987655781527","0.981160022285","0.999998130160","0.999995258206","0.990579076223","0.986349757961","0.984397186102","0.982452329568","0.969273787578","0.984402543989","0.984400174373","0.000004741794","0.984394814870","0.996926141446","0.996646530259","0.996786316244","0.996786316244"
+"POLISH_RADIXOR","PL_PL","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","9846","120925","1","118145","2780","4","124274","9847","1114651","0","0","7310252699","0","7310252699","0.000000","0","1114651","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"POLISH_RADIXOR","PL_PL","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","9846","120925","1","118145","2780","4","124274","9847","1114651","38073","0","7310214626","38073","7310252699","0.000521","0","1114651","0.000000","0.966971278467","1.000000000000","0.999994791835","0.999994792629","0.999997395918","0.973401318686","0.983208335630","0.993214975136","0.966971278467","0.983346977657","0.983344416937","0.000005207371","","","","",""
+"PORTUGUESE_LUCENE_PORTUGUESE_LIGHT_STEM_FILTER","PT_PT","ALL_WORDS","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","112814","149830","2511","5339230","22358201245","2511","22358203756","0.000011","5339230","5489060","97.270389","0.983517240927","0.027296112631","0.999999887692","0.999761142266","0.513648000162","0.122843213263","0.053118010934","0.033885033146","0.027283631587","0.163848092400","0.163827867245","0.000238857734","0.053105458848","0.999226179883","0.720580123442","0.837329788424","0.837329788424"
+"PORTUGUESE_LUCENE_PORTUGUESE_LIGHT_STEM_FILTER","PT_PT","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","112814","149830","2511","5339230","22358201245","2511","22358203756","0.000011","5339230","5489060","97.270389","0.983517240927","0.027296112631","0.999999887692","0.999761142266","0.513648000162","0.122843213263","0.053118010934","0.033885033146","0.027283631587","0.163848092400","0.163827867245","0.000238857734","0.053105458848","0.999226179883","0.720580123442","0.837329788424","0.837329788424"
+"PORTUGUESE_LUCENE_PORTUGUESE_MINIMAL_STEM_FILTER","PT_PT","ALL_WORDS","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","167745","43406","598","5445654","22358203158","598","22358203756","0.000003","5445654","5489060","99.209227","0.986410326334","0.007907729192","0.999999973254","0.999756469021","0.503953851223","0.038310165654","0.015689679353","0.009864890589","0.007906867787","0.088319113068","0.088308061848","0.000243530979","0.015685836581","0.999664059174","0.692382565086","0.818121623354","0.818121623354"
+"PORTUGUESE_LUCENE_PORTUGUESE_MINIMAL_STEM_FILTER","PT_PT","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","167745","43406","598","5445654","22358203158","598","22358203756","0.000003","5445654","5489060","99.209227","0.986410326334","0.007907729192","0.999999973254","0.999756469021","0.503953851223","0.038310165654","0.015689679353","0.009864890589","0.007906867787","0.088319113068","0.088308061848","0.000243530979","0.015685836581","0.999664059174","0.692382565086","0.818121623354","0.818121623354"
+"PORTUGUESE_LUCENE_PORTUGUESE_STEM_FILTER","PT_PT","ALL_WORDS","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","27586","3803488","99075","1685572","22358104681","99075","22358203756","0.000443","1685572","5489060","30.707844","0.974612837768","0.692921556696","0.999995568741","0.999920198913","0.846458562719","0.901329863268","0.809974591186","0.735433886866","0.680636384053","0.821784792219","0.821750382444","0.000079801087","0.809935821347","0.996728545383","0.918475110538","0.956003146785","0.956003146785"
+"PORTUGUESE_LUCENE_PORTUGUESE_STEM_FILTER","PT_PT","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","27586","3803488","99075","1685572","22358104681","99075","22358203756","0.000443","1685572","5489060","30.707844","0.974612837768","0.692921556696","0.999995568741","0.999920198913","0.846458562719","0.901329863268","0.809974591186","0.735433886866","0.680636384053","0.821784792219","0.821750382444","0.000079801087","0.809935821347","0.996728545383","0.918475110538","0.956003146785","0.956003146785"
+"PORTUGUESE_RADIXOR","PT_PT","ALL_WORDS","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","4001","5472616","20678","16444","22358183078","20678","22358203756","0.000092","16444","5489060","0.299578","0.996235774018","0.997004222945","0.999999075149","0.999998340077","0.998501649047","0.996389369023","0.996619850353","0.996850438335","0.993262474550","0.996619924417","0.996619094289","0.000001659923","0.996619020188","0.999299376330","0.999346803887","0.999323089546","0.999323089546"
+"PORTUGUESE_RADIXOR","PT_PT","ALL_WORDS","ANY_CANDIDATE","4001","211489","0","210699","790","3","212297","4001","5489060","0","0","22358203756","0","22358203756","0.000000","0","5489060","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"PORTUGUESE_RADIXOR","PT_PT","ALL_WORDS","ALL_CANDIDATES","4001","211489","0","210699","790","3","212297","4001","5489060","38310","0","22358165446","38310","22358203756","0.000171","0","5489060","0.000000","0.993069036450","1.000000000000","0.999998286535","0.999998286956","0.999999143267","0.994447532369","0.996522466897","0.998606078314","0.993069036450","0.996528492543","0.996527638784","0.000001713044","","","","",""
+"PORTUGUESE_RADIXOR","PT_PT","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","4001","5472616","20678","16444","22358183078","20678","22358203756","0.000092","16444","5489060","0.299578","0.996235774018","0.997004222945","0.999999075149","0.999998340077","0.998501649047","0.996389369023","0.996619850353","0.996850438335","0.993262474550","0.996619924417","0.996619094289","0.000001659923","0.996619020188","0.999299376330","0.999346803887","0.999323089546","0.999323089546"
+"PORTUGUESE_RADIXOR","PT_PT","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","4001","211489","0","210699","790","3","212297","4001","5489060","0","0","22358203756","0","22358203756","0.000000","0","5489060","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"PORTUGUESE_RADIXOR","PT_PT","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","4001","211489","0","210699","790","3","212297","4001","5489060","38310","0","22358165446","38310","22358203756","0.000171","0","5489060","0.000000","0.993069036450","1.000000000000","0.999998286535","0.999998286956","0.999999143267","0.994447532369","0.996522466897","0.998606078314","0.993069036450","0.996528492543","0.996527638784","0.000001713044","","","","",""
+"RUSSIAN_LUCENE_RUSSIAN_LIGHT_STEM_FILTER","RU_RU","ALL_WORDS","PRIMARY_OUTPUT","37410","768882","10","768882","0","1","768882","232250","3081067","321183","10008438","295575969833","321183","295576291016","0.000109","10008438","13089505","76.461547","0.905596884415","0.235384531348","0.999998913367","0.999965054154","0.617691722357","0.577011147253","0.373649378129","0.276277984307","0.229747124085","0.461696326851","0.461686629842","0.000034945846","0.373637933830","0.994310930069","0.870888421754","0.928516167212","0.928516167212"
+"RUSSIAN_LUCENE_RUSSIAN_LIGHT_STEM_FILTER","RU_RU","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","37297","768133","10","768133","0","1","768133","232143","3078888","318921","10008238","295000362731","318921","295000681652","0.000108","10008238","13087126","76.473918","0.906139220892","0.235260820443","0.999998918914","0.999964994315","0.617629869679","0.577038425373","0.373539598427","0.276151716079","0.229664120975","0.461713175622","0.461703471649","0.000035005685","0.373528142103","0.994349544377","0.870767393403","0.928464208705","0.928464208705"
+"RUSSIAN_RADIXOR","RU_RU","ALL_WORDS","PRIMARY_OUTPUT","37410","768882","10","768882","0","1","768882","37561","12823203","155850","266302","295576135166","155850","295576291016","0.000053","266302","13089505","2.034470","0.987992190185","0.979655304001","0.999999472725","0.999998571830","0.989827388363","0.986313480705","0.983806085477","0.981311406466","0.968128298562","0.983814916245","0.983814202914","0.000001428170","0.983805371373","0.997699288696","0.997273959852","0.997486578934","0.997486578934"
+"RUSSIAN_RADIXOR","RU_RU","ALL_WORDS","ANY_CANDIDATE","37410","768882","10","749720","19162","4","788492","37593","13089492","0","13","295576291016","0","295576291016","0.000000","13","13089505","0.000099","1.000000000000","0.999999006838","1.000000000000","0.999999999956","0.999999503419","0.999999801367","0.999999503419","0.999999205470","0.999999006838","0.999999503419","0.999999503397","0.000000000044","","","","",""
+"RUSSIAN_RADIXOR","RU_RU","ALL_WORDS","ALL_CANDIDATES","37410","768882","10","749720","19162","4","788492","37593","13089492","434710","13","295575856306","434710","295576291016","0.000147","13","13089505","0.000099","0.967856883534","0.999999006838","0.999998529280","0.999998529301","0.999998768059","0.974118939969","0.983665447282","0.993400920813","0.967855953192","0.983796687479","0.983795964011","0.000001470699","","","","",""
+"RUSSIAN_RADIXOR","RU_RU","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","37297","768133","10","768133","0","1","768133","37282","12821513","155850","265613","295000525802","155850","295000681652","0.000053","265613","13087126","2.029575","0.987990626447","0.979704252867","0.999999471696","0.999998571379","0.989851862281","0.986322156837","0.983829991833","0.981350389119","0.968174600634","0.983838715706","0.983838002141","0.000001428621","0.983829277503","0.997696524283","0.997321167437","0.997508810549","0.997508810549"
+"RUSSIAN_RADIXOR","RU_RU","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","37297","768133","10","749142","18991","4","787549","37306","13087126","0","0","295000681652","0","295000681652","0.000000","0","13087126","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"RUSSIAN_RADIXOR","RU_RU","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","37297","768133","10","749142","18991","4","787549","37306","13087126","434710","0","295000246942","434710","295000681652","0.000147","0","13087126","0.000000","0.967851259252","1.000000000000","0.999998526410","0.999998526476","0.999999263205","0.974114570610","0.983663023007","0.993400519870","0.967851259252","0.983794317554","0.983793592699","0.000001473524","","","","",""
+"SNOWBALL_DANISH_DIRECT","DA_DK","ALL_WORDS","PRIMARY_OUTPUT","4179","28079","32","28079","0","1","28079","5553","78732","6341","11163","394104845","6341","394111186","0.001609","11163","89895","12.417821","0.925464013259","0.875821792091","0.999983910632","0.999955596266","0.937902851361","0.915090414169","0.899958849618","0.885319563795","0.818113803566","0.900300811178","0.900278764621","0.000044403734","0.899936659693","0.994195946109","0.978579164615","0.986325742978","0.986325742978"
+"SNOWBALL_DANISH_DIRECT","DA_DK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4173","28033","32","28033","0","1","28033","5539","78627","6341","11113","392814447","6341","392820788","0.001614","11113","89740","12.383552","0.925371904717","0.876164475150","0.999983857779","0.999955577673","0.938074166465","0.915093153823","0.900096160451","0.885582797210","0.818340774971","0.900432112497","0.900410054086","0.000044422327","0.900073960926","0.994185354918","0.978644331817","0.986353630937","0.986353630937"
+"SNOWBALL_DANISH_LUCENE_FILTER","DA_DK","ALL_WORDS","PRIMARY_OUTPUT","4179","28079","32","28079","0","1","28079","5546","78744","6507","11151","394104679","6507","394111186","0.001651","11151","89895","12.404472","0.923672449590","0.875955281161","0.999983489431","0.999955205602","0.937969385296","0.913717599716","0.899181254496","0.885100184115","0.816829526358","0.899497504322","0.899475250557","0.000044794398","0.899158868074","0.994052746860","0.978603476314","0.986267614487","0.986267614487"
+"SNOWBALL_DANISH_LUCENE_FILTER","DA_DK","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4173","28033","32","28033","0","1","28033","5539","78627","6341","11113","392814447","6341","392820788","0.001614","11113","89740","12.383552","0.925371904717","0.876164475150","0.999983857779","0.999955577673","0.938074166465","0.915093153823","0.900096160451","0.885582797210","0.818340774971","0.900432112497","0.900410054086","0.000044422327","0.900073960926","0.994185354918","0.978644331817","0.986353630937","0.986353630937"
+"SNOWBALL_DUTCH_DIRECT","NL_NL","ALL_WORDS","PRIMARY_OUTPUT","4992","26477","85","26477","0","1","26477","12051","29325","4382","35241","350433578","4382","350437960","0.001250","35241","64566","54.581359","0.869997329931","0.454186413902","0.999987495647","0.999886953739","0.727086954774","0.735353119953","0.596806854375","0.502190286022","0.425320531415","0.628602392126","0.628557411604","0.000113046261","0.596755898222","0.992814719235","0.917346281080","0.953589661124","0.953589661124"
+"SNOWBALL_DUTCH_DIRECT","NL_NL","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4796","25678","84","25678","0","1","25678","11466","29111","4382","34036","329599474","4382","329603856","0.001329","34036","63147","53.899631","0.869166691547","0.461003689803","0.999986705253","0.999883464224","0.730495197528","0.738411822300","0.602462748344","0.508789468717","0.431088865524","0.633000040962","0.632953313739","0.000116535776","0.602409959779","0.992557044900","0.918058993802","0.953855618789","0.953855618789"
+"SNOWBALL_DUTCH_LUCENE_FILTER","NL_NL","ALL_WORDS","PRIMARY_OUTPUT","4992","26477","85","26477","0","1","26477","14573","15302","1588","49264","350436372","1588","350437960","0.000453","49264","64566","76.300220","0.905979869745","0.236997800700","0.999995468527","0.999854916880","0.618496634614","0.579068464950","0.375712040856","0.278062466837","0.231308764398","0.463373754768","0.463333378452","0.000145083120","0.375664346452","0.995827666179","0.888410302915","0.939057139355","0.939057139355"
+"SNOWBALL_DUTCH_LUCENE_FILTER","NL_NL","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4796","25678","84","25678","0","1","25678","14116","14972","1544","48175","329602312","1544","329603856","0.000468","48175","63147","76.290243","0.906514894648","0.237097565997","0.999995315589","0.999849184178","0.618546440793","0.579362438183","0.375883408860","0.278182412747","0.231438685443","0.463608105042","0.463566154833","0.000150815822","0.375833834664","0.995816517119","0.887491664267","0.938538755173","0.938538755173"
+"SNOWBALL_FINNISH_DIRECT","FI_FI","ALL_WORDS","PRIMARY_OUTPUT","57027","1811717","292","1811717","0","1","1811717","381483","15114332","1544812","16409363","1641125269679","1544812","1641126814491","0.000094","16409363","31523695","52.054060","0.907269425128","0.479459403474","0.999999058688","0.999989060059","0.739729231081","0.769880311353","0.627374073993","0.529384116965","0.457061215373","0.659544431681","0.659540149918","0.000010939941","0.627369124557","0.991871857177","0.904138579582","0.945975396220","0.945975396220"
+"SNOWBALL_FINNISH_DIRECT","FI_FI","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","54762","1757055","274","1757055","0","1","1757055","372232","14692070","1513705","16121763","1543587930447","1513705","1543589444152","0.000098","16121763","30813833","52.319888","0.906594717007","0.476801117213","0.999999019360","0.999988575255","0.738400068286","0.768116957494","0.624933751043","0.526744348874","0.454475376380","0.657468914800","0.657464451566","0.000011424745","0.624928589970","0.991731678095","0.902933406396","0.945251664438","0.945251664438"
+"SNOWBALL_FINNISH_LUCENE_FILTER","FI_FI","ALL_WORDS","PRIMARY_OUTPUT","57027","1811717","292","1811717","0","1","1811717","377778","15153638","1922153","16370057","1641124892338","1922153","1641126814491","0.000117","16370057","31523695","51.929372","0.887434028678","0.480706275073","0.999998828760","0.999988854086","0.740352551917","0.758996033322","0.623613097472","0.529216231176","0.453079796332","0.653142485450","0.653138019077","0.000011145914","0.623608016975","0.990717710840","0.904385055188","0.945584912497","0.945584912497"
+"SNOWBALL_FINNISH_LUCENE_FILTER","FI_FI","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","54762","1757055","274","1757055","0","1","1757055","372232","14692070","1513705","16121763","1543587930447","1513705","1543589444152","0.000098","16121763","30813833","52.319888","0.906594717007","0.476801117213","0.999999019360","0.999988575255","0.738400068286","0.768116957494","0.624933751043","0.526744348874","0.454475376380","0.657468914800","0.657464451566","0.000011424745","0.624928589970","0.991731678095","0.902933406396","0.945251664438","0.945251664438"
+"SNOWBALL_FRENCH_DIRECT","FR_FR","ALL_WORDS","PRIMARY_OUTPUT","59240","425210","2301","425210","0","1","425210","85627","3766640","1654723","1687975","90394450107","1654723","90396104830","0.001831","1687975","5454615","30.945814","0.694777309691","0.690541862258","0.999981694753","0.999963023890","0.845261778506","0.693926068790","0.692653111288","0.691384815533","0.529815856272","0.692656348624","0.692637859933","0.000036976110","0.692634622294","0.959459328254","0.944947915186","0.952148333884","0.952148333884"
+"SNOWBALL_FRENCH_DIRECT","FR_FR","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","57698","421231","2133","421231","0","1","421231","84526","3758589","1646111","1681970","88710480395","1646111","88712126506","0.001856","1681970","5440559","30.915389","0.695429718578","0.690846106071","0.999981444352","0.999962486787","0.845413775212","0.694508136723","0.693130334647","0.691757988461","0.530374491828","0.693134123475","0.693115366288","0.000037513213","0.693111577099","0.959520798119","0.944537159644","0.951970023370","0.951970023370"
+"SNOWBALL_FRENCH_LUCENE_FILTER","FR_FR","ALL_WORDS","PRIMARY_OUTPUT","59240","425210","2301","425210","0","1","425210","85202","3763777","1661388","1690838","90394443442","1661388","90396104830","0.001838","1690838","5454615","30.998301","0.693762678186","0.690016985617","0.999981621022","0.999962918494","0.844999303320","0.693010289898","0.691884762376","0.690762884895","0.528917286853","0.691887297134","0.691868755638","0.000037081506","0.691866220643","0.958697792387","0.944714715363","0.951654891797","0.951654891797"
+"SNOWBALL_FRENCH_LUCENE_FILTER","FR_FR","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","57698","421231","2133","421231","0","1","421231","84810","3755856","1641925","1684703","88710484581","1641925","88712126506","0.001851","1684703","5440559","30.965623","0.695814817237","0.690343767984","0.999981491538","0.999962503165","0.845162629761","0.694713680979","0.693068495729","0.691431084156","0.530302080457","0.693073894149","0.693055145391","0.000037496835","0.693049746458","0.959566165512","0.944384738154","0.951914926182","0.951914926182"
+"SNOWBALL_GERMAN_DIRECT","DE_DE","ALL_WORDS","PRIMARY_OUTPUT","54092","296974","1474","296974","0","1","296974","81649","771138","190680","612734","44095055299","190680","44095245979","0.000432","612734","1383872","44.276783","0.801750435114","0.557232171762","0.999995675724","0.999981780603","0.778613923743","0.737064397386","0.657493530688","0.593429030432","0.489750735447","0.668401927114","0.668393541401","0.000018219397","0.657484715679","0.983724573695","0.949324273697","0.966218331938","0.966218331938"
+"SNOWBALL_GERMAN_DIRECT","DE_DE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","16007","150098","228","150098","0","1","150098","37843","516936","87697","356475","11263668645","87697","11263756342","0.000779","356475","873411","40.814118","0.854958297017","0.591858815609","0.999992214231","0.999960569321","0.795925514920","0.785153327381","0.699486618802","0.630674793334","0.537854226580","0.711347035607","0.711329191110","0.000039430679","0.699467554207","0.988417636496","0.932451723900","0.959619376607","0.959619376607"
+"SNOWBALL_GERMAN_LUCENE_FILTER","DE_DE","ALL_WORDS","PRIMARY_OUTPUT","54092","296974","1474","296974","0","1","296974","86669","751056","295701","632816","44094950278","295701","44095245979","0.000671","632816","1383872","45.727929","0.717507501741","0.542720714054","0.999993294039","0.999978943584","0.771357004047","0.674088567377","0.617993120299","0.570516594262","0.447170798768","0.624024185176","0.624014089276","0.000021056416","0.617982794334","0.975844522648","0.942925157079","0.959102449371","0.959102449371"
+"SNOWBALL_GERMAN_LUCENE_FILTER","DE_DE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","16007","150098","228","150098","0","1","150098","46077","481501","77653","391910","11263678689","77653","11263756342","0.000689","391910","873411","44.871200","0.861124126806","0.551287996144","0.999993105941","0.999958315274","0.775640551042","0.774110642769","0.672222202832","0.594035281304","0.506276128631","0.689004640259","0.688986412764","0.000041684726","0.672202362239","0.989021274644","0.919542244735","0.953017108147","0.953017108147"
+"SNOWBALL_HUNGARIAN_DIRECT","HU_HU","ALL_WORDS","PRIMARY_OUTPUT","19406","916344","1","916344","0","1","916344","116105","14287912","1506056","7874191","419819036837","1506056","419820542893","0.000359","7874191","22162103","35.529981","0.904643595580","0.644700189328","0.999996412620","0.999977657711","0.822348300974","0.837136808086","0.752865700984","0.684009307333","0.603676525918","0.763690969794","0.763680928295","0.000022342289","0.752854843818","0.991947513126","0.924304257644","0.956931989551","0.956931989551"
+"SNOWBALL_HUNGARIAN_DIRECT","HU_HU","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","18360","878513","1","878513","0","1","878513","111379","13776526","1496670","7634885","385869198247","1496670","385870694917","0.000388","7634885","21411411","35.658019","0.902006757459","0.643419810119","0.999996121317","0.999976336507","0.821707965718","0.834898516372","0.751079383241","0.682554714263","0.601382804609","0.761819543337","0.761808891723","0.000023663493","0.751067882221","0.991609896137","0.923288102987","0.956230170303","0.956230170303"
+"SNOWBALL_HUNGARIAN_LUCENE_FILTER","HU_HU","ALL_WORDS","PRIMARY_OUTPUT","19406","916344","1","916344","0","1","916344","114867","14299358","1792049","7862745","419818750844","1792049","419820542893","0.000427","7862745","22162103","35.478334","0.888633169244","0.645216656560","0.999995731393","0.999977003783","0.822606193976","0.826287586346","0.747610245439","0.682613266689","0.596946950992","0.757205997314","0.757195513225","0.000022996217","0.747599036407","0.990687085622","0.924490230693","0.956444632598","0.956444632598"
+"SNOWBALL_HUNGARIAN_LUCENE_FILTER","HU_HU","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","18360","878513","1","878513","0","1","878513","111379","13776526","1496670","7634885","385869198247","1496670","385870694917","0.000388","7634885","21411411","35.658019","0.902006757459","0.643419810119","0.999996121317","0.999976336507","0.821707965718","0.834898516372","0.751079383241","0.682554714263","0.601382804609","0.761819543337","0.761808891723","0.000023663493","0.751067882221","0.991609896137","0.923288102987","0.956230170303","0.956230170303"
+"SNOWBALL_ITALIAN_DIRECT","IT_IT","ALL_WORDS","PRIMARY_OUTPUT","10009","327551","0","327551","0","1","327551","46828","4499650","504775","1644164","53638016436","504775","53638521211","0.000941","1644164","6143814","26.761292","0.899134266174","0.732387080729","0.999990589319","0.999959941236","0.866188835024","0.859975076366","0.807239600802","0.760598128154","0.676782697802","0.811488952720","0.811469907061","0.000040058764","0.807219778599","0.987993752409","0.933407841859","0.959925420024","0.959925420024"
+"SNOWBALL_ITALIAN_DIRECT","IT_IT","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","10007","327469","0","327469","0","1","327469","46814","4498652","504774","1643522","53611162298","504774","53611667072","0.000942","1643522","6142174","26.757985","0.899114326863","0.732420149608","0.999990584624","0.999959933163","0.866205367116","0.859969602244","0.807251650876","0.760623806435","0.676799637969","0.811498274672","0.811479224597","0.000040066837","0.807231824542","0.987990915845","0.933413003102","0.959926810498","0.959926810498"
+"SNOWBALL_ITALIAN_LUCENE_FILTER","IT_IT","ALL_WORDS","PRIMARY_OUTPUT","10009","327551","0","327551","0","1","327551","46828","4499650","504775","1644164","53638016436","504775","53638521211","0.000941","1644164","6143814","26.761292","0.899134266174","0.732387080729","0.999990589319","0.999959941236","0.866188835024","0.859975076366","0.807239600802","0.760598128154","0.676782697802","0.811488952720","0.811469907061","0.000040058764","0.807219778599","0.987993752409","0.933407841859","0.959925420024","0.959925420024"
+"SNOWBALL_ITALIAN_LUCENE_FILTER","IT_IT","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","10007","327469","0","327469","0","1","327469","46814","4498652","504774","1643522","53611162298","504774","53611667072","0.000942","1643522","6142174","26.757985","0.899114326863","0.732420149608","0.999990584624","0.999959933163","0.866205367116","0.859969602244","0.807251650876","0.760623806435","0.676799637969","0.811498274672","0.811479224597","0.000040066837","0.807231824542","0.987990915845","0.933413003102","0.959926810498","0.959926810498"
+"SNOWBALL_NORWEGIAN_BOKMAL_DIRECT","NB_NO","ALL_WORDS","PRIMARY_OUTPUT","17929","75310","252","75310","0","1","75310","24394","106626","23997","35554","2835594218","23997","2835618215","0.000846","35554","142180","25.006330","0.816288096277","0.749936699958","0.999991537295","0.999978999989","0.874964118626","0.802094867845","0.781706946038","0.762329786671","0.641641141674","0.782409356499","0.782398932962","0.000021000011","0.781696464373","0.988328173631","0.971119668400","0.979648355684","0.979648355684"
+"SNOWBALL_NORWEGIAN_BOKMAL_DIRECT","NB_NO","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","17914","75251","252","75251","0","1","75251","24367","106567","23997","35524","2831152787","23997","2831176784","0.000848","35524","142091","25.000880","0.816205079501","0.749991202821","0.999991524019","0.999978977642","0.874991363420","0.802043209347","0.781698483431","0.762360357576","0.641629738452","0.782397999310","0.782387564360","0.000021022358","0.781687990535","0.988317966243","0.971127035574","0.979647089734","0.979647089734"
+"SNOWBALL_NORWEGIAN_BOKMAL_LUCENE_FILTER","NB_NO","ALL_WORDS","PRIMARY_OUTPUT","17929","75310","252","75310","0","1","75310","24396","106589","24046","35591","2835594169","24046","2835618215","0.000848","35591","142180","25.032353","0.815929880966","0.749676466451","0.999991520015","0.999978969662","0.874833993233","0.801758635215","0.781401315910","0.762052176648","0.641229410562","0.782101930719","0.782091491842","0.000021030338","0.781390819068","0.988295184140","0.971086244692","0.979615142710","0.979615142710"
+"SNOWBALL_NORWEGIAN_BOKMAL_LUCENE_FILTER","NB_NO","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","17914","75251","252","75251","0","1","75251","24381","106512","23993","35579","2831152791","23993","2831176784","0.000847","35579","142091","25.039587","0.816152637830","0.749604126933","0.999991525432","0.999978959629","0.874797826182","0.801914137847","0.781464144742","0.762031224736","0.641314033862","0.782170943927","0.782160500766","0.000021040371","0.781453643056","0.988310445474","0.971073741466","0.979616277822","0.979616277822"
+"SNOWBALL_NORWEGIAN_NYNORSK_DIRECT","NN_NO","ALL_WORDS","PRIMARY_OUTPUT","4688","18250","23","18250","0","1","18250","6138","22004","8274","8648","166483199","8274","166491473","0.004970","8648","30652","28.213493","0.726732280864","0.717865065901","0.999950303761","0.999898379870","0.858907684831","0.724941356316","0.722271459051","0.719621155632","0.565277706417","0.722285066089","0.722234252664","0.000101620130","0.722220641604","0.980542486408","0.964998187466","0.972708239744","0.972708239744"
+"SNOWBALL_NORWEGIAN_NYNORSK_DIRECT","NN_NO","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4681","18219","23","18219","0","1","18219","6120","21971","8274","8624","165918002","8274","165926276","0.004987","8624","30595","28.187612","0.726434121342","0.718123876450","0.999950134480","0.999898178365","0.859037005465","0.724756721095","0.722255095332","0.719770679771","0.565257660346","0.722267047015","0.722216132089","0.000101821635","0.722204176866","0.980505813025","0.965064509051","0.972723884952","0.972723884952"
+"SNOWBALL_NORWEGIAN_NYNORSK_LUCENE_FILTER","NN_NO","ALL_WORDS","PRIMARY_OUTPUT","4688","18250","23","18250","0","1","18250","6144","21978","8295","8674","166483178","8295","166491473","0.004982","8674","30652","28.298317","0.725993459518","0.717016834138","0.999950177629","0.999898097625","0.858483505883","0.724180198229","0.721477226098","0.718794356395","0.564305338023","0.721491186328","0.721440231913","0.000101902375","0.721426267560","0.980461058483","0.964862123312","0.972599049418","0.972599049418"
+"SNOWBALL_NORWEGIAN_NYNORSK_LUCENE_FILTER","NN_NO","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4681","18219","23","18219","0","1","18219","6130","21948","8274","8647","165918002","8274","165926276","0.004987","8647","30595","28.262788","0.726225928132","0.717372119627","0.999950134480","0.999898039774","0.858661127054","0.724437725685","0.721771872996","0.719125568472","0.564665929147","0.721785448310","0.721734464790","0.000101960226","0.721720885459","0.980505813025","0.964944952705","0.972663150405","0.972663150405"
+"SNOWBALL_PORTUGUESE_DIRECT","PT_PT","ALL_WORDS","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","11315","4817239","167230","671821","22358036526","167230","22358203756","0.000748","671821","5489060","12.239272","0.966449786326","0.877607277020","0.999992520419","0.999962481554","0.938799898719","0.947270839082","0.919888415834","0.894044585092","0.851660540743","0.920957852105","0.920939611009","0.000037518446","0.919869695779","0.996663145176","0.967923515462","0.982083116554","0.982083116554"
+"SNOWBALL_PORTUGUESE_DIRECT","PT_PT","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","11315","4817239","167230","671821","22358036526","167230","22358203756","0.000748","671821","5489060","12.239272","0.966449786326","0.877607277020","0.999992520419","0.999962481554","0.938799898719","0.947270839082","0.919888415834","0.894044585092","0.851660540743","0.920957852105","0.920939611009","0.000037518446","0.919869695779","0.996663145176","0.967923515462","0.982083116554","0.982083116554"
+"SNOWBALL_PORTUGUESE_LUCENE_FILTER","PT_PT","ALL_WORDS","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","11315","4817239","167230","671821","22358036526","167230","22358203756","0.000748","671821","5489060","12.239272","0.966449786326","0.877607277020","0.999992520419","0.999962481554","0.938799898719","0.947270839082","0.919888415834","0.894044585092","0.851660540743","0.920957852105","0.920939611009","0.000037518446","0.919869695779","0.996663145176","0.967923515462","0.982083116554","0.982083116554"
+"SNOWBALL_PORTUGUESE_LUCENE_FILTER","PT_PT","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","4001","211489","0","211489","0","1","211489","11315","4817239","167230","671821","22358036526","167230","22358203756","0.000748","671821","5489060","12.239272","0.966449786326","0.877607277020","0.999992520419","0.999962481554","0.938799898719","0.947270839082","0.919888415834","0.894044585092","0.851660540743","0.920957852105","0.920939611009","0.000037518446","0.919869695779","0.996663145176","0.967923515462","0.982083116554","0.982083116554"
+"SNOWBALL_RUSSIAN_DIRECT","RU_RU","ALL_WORDS","PRIMARY_OUTPUT","37410","768882","10","768882","0","1","768882","64358","8766656","3782908","4322849","295572508108","3782908","295576291016","0.001280","4322849","13089505","33.025305","0.698562595481","0.669746946122","0.999987201585","0.999972577645","0.834867073854","0.692602792505","0.683851352013","0.675318311031","0.519585195076","0.684003044583","0.683989349009","0.000027422355","0.683837646322","0.974179960240","0.953661001039","0.963811283954","0.963811283954"
+"SNOWBALL_RUSSIAN_DIRECT","RU_RU","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","37297","768133","10","768133","0","1","768133","64159","8764719","3782908","4322407","294996898744","3782908","295000681652","0.001282","4322407","13087126","33.027931","0.698516062041","0.669720685810","0.999987176613","0.999972525638","0.834853931211","0.692560581516","0.683815365804","0.675288254087","0.519543647630","0.683966853085","0.683953131513","0.000027474362","0.683801634112","0.974148936252","0.953633561396","0.963782086975","0.963782086975"
+"SNOWBALL_RUSSIAN_LUCENE_FILTER","RU_RU","ALL_WORDS","PRIMARY_OUTPUT","37410","768882","10","768882","0","1","768882","64266","8766889","3785790","4322616","295572505226","3785790","295576291016","0.001281","4322616","13089505","33.023525","0.698407806015","0.669764746642","0.999987191835","0.999972568683","0.834875969239","0.692484865100","0.683786451263","0.675303850911","0.519510266339","0.683936347366","0.683922647122","0.000027431317","0.683772741022","0.974131099393","0.953673855106","0.963793934411","0.963793934411"
+"SNOWBALL_RUSSIAN_LUCENE_FILTER","RU_RU","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","37297","768133","10","768133","0","1","768133","64159","8764719","3782908","4322407","294996898744","3782908","295000681652","0.001282","4322407","13087126","33.027931","0.698516062041","0.669720685810","0.999987176613","0.999972525638","0.834853931211","0.692560581516","0.683815365804","0.675288254087","0.519543647630","0.683966853085","0.683953131513","0.000027474362","0.683801634112","0.974148936252","0.953633561396","0.963782086975","0.963782086975"
+"SNOWBALL_SPANISH_DIRECT","ES_ES","ALL_WORDS","PRIMARY_OUTPUT","65059","871332","3589","871332","0","1","871332","195021","12811687","2228819","29161649","379565089291","2228819","379567318110","0.000587","29161649","41973336","69.476605","0.851812232913","0.305233946618","0.999994128001","0.999917308483","0.652614037309","0.627191552465","0.449423738186","0.350172671706","0.289843040458","0.509903921959","0.509876023351","0.000082691517","0.449391616998","0.981405614580","0.852462513401","0.912400934512","0.912400934512"
+"SNOWBALL_SPANISH_DIRECT","ES_ES","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","64918","869371","3525","869371","0","1","869371","194444","12787018","2201196","29076352","377858468569","2201196","377860669765","0.000583","29076352","41863370","69.455354","0.853138205793","0.305446455935","0.999994174583","0.999917233823","0.652720315259","0.627945981812","0.449838583213","0.350441220963","0.290188220622","0.510478247707","0.510450359519","0.000082766177","0.449806446076","0.981468761133","0.852555702466","0.912481600659","0.912481600659"
+"SNOWBALL_SPANISH_LUCENE_FILTER","ES_ES","ALL_WORDS","PRIMARY_OUTPUT","65059","871332","3589","871332","0","1","871332","194971","12811693","2230481","29161643","379565087629","2230481","379567318110","0.000588","29161643","41973336","69.476591","0.851718175843","0.305234089566","0.999994123622","0.999917304121","0.652614106594","0.627150877515","0.449410800675","0.350169642836","0.289832278511","0.509875888791","0.509847985527","0.000082695879","0.449378676109","0.981386049614","0.852460744495","0.912391466047","0.912391466047"
+"SNOWBALL_SPANISH_LUCENE_FILTER","ES_ES","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","64918","869371","3525","869371","0","1","869371","194444","12787018","2201196","29076352","377858468569","2201196","377860669765","0.000583","29076352","41863370","69.455354","0.853138205793","0.305446455935","0.999994174583","0.999917233823","0.652720315259","0.627945981812","0.449838583213","0.350441220963","0.290188220622","0.510478247707","0.510450359519","0.000082766177","0.449806446076","0.981468761133","0.852555702466","0.912481600659","0.912481600659"
+"SNOWBALL_SWEDISH_DIRECT","SV_SE","ALL_WORDS","PRIMARY_OUTPUT","12371","98108","68","98108","0","1","98108","25915","237017","67105","148325","4812088331","67105","4812155436","0.001394","148325","385342","38.491781","0.779348419384","0.615082186733","0.999986055105","0.999955235704","0.807534120919","0.739831942216","0.687539886056","0.642151948805","0.523855832838","0.692360693585","0.692339154006","0.000044764296","0.687517812951","0.984860422704","0.942685282143","0.963311451566","0.963311451566"
+"SNOWBALL_SWEDISH_DIRECT","SV_SE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","12342","97881","68","97881","0","1","97881","25840","236588","67105","147975","4789844472","67105","4789911577","0.001401","147975","384563","38.478741","0.779036724587","0.615212591955","0.999985990347","0.999955100897","0.807599291151","0.739644914918","0.687500000000","0.642223301999","0.523809523810","0.692295603454","0.692273994517","0.000044899103","0.687477858823","0.984821307273","0.942694565973","0.963297587020","0.963297587020"
+"SNOWBALL_SWEDISH_LUCENE_FILTER","SV_SE","ALL_WORDS","PRIMARY_OUTPUT","12371","98108","68","98108","0","1","98108","26781","230676","64262","154666","4812091174","64262","4812155436","0.001335","154666","385342","40.137333","0.782116919488","0.598626674487","0.999986645901","0.999954508853","0.799306660194","0.736939762085","0.678179573117","0.628097931391","0.513064830384","0.684248529829","0.684226838572","0.000045491147","0.678157227687","0.985207247898","0.939659207875","0.961894327137","0.961894327137"
+"SNOWBALL_SWEDISH_LUCENE_FILTER","SV_SE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","12342","97881","68","97881","0","1","97881","26706","230247","64262","154316","4789847315","64262","4789911577","0.001342","154316","384563","40.127625","0.781799537535","0.598723746174","0.999986583886","0.999954370671","0.799355165030","0.736743719918","0.678122496584","0.628142458291","0.512999498691","0.684165146635","0.684143384784","0.000045629329","0.678100081584","0.985169028543","0.939660518299","0.961876797342","0.961876797342"
+"SNOWBALL_YIDDISH_DIRECT","YI","ALL_WORDS","PRIMARY_OUTPUT","802","3578","0","3578","0","1","3578","1087","4962","1151","1382","6391758","1151","6392909","0.018004","1382","6344","21.784363","0.811712743334","0.782156368222","0.999819956768","0.999604172550","0.890988162495","0.805624107027","0.796660512162","0.787894185271","0.662041360907","0.796797522188","0.796599716782","0.000395827450","0.796462473806","0.982918530193","0.962014249878","0.972354049666","0.972354049666"
+"SNOWBALL_YIDDISH_DIRECT","YI","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","802","3578","0","3578","0","1","3578","1087","4962","1151","1382","6391758","1151","6392909","0.018004","1382","6344","21.784363","0.811712743334","0.782156368222","0.999819956768","0.999604172550","0.890988162495","0.805624107027","0.796660512162","0.787894185271","0.662041360907","0.796797522188","0.796599716782","0.000395827450","0.796462473806","0.982918530193","0.962014249878","0.972354049666","0.972354049666"
+"SNOWBALL_YIDDISH_LUCENE_FILTER","YI","ALL_WORDS","PRIMARY_OUTPUT","802","3578","0","3578","0","1","3578","1087","4962","1151","1382","6391758","1151","6392909","0.018004","1382","6344","21.784363","0.811712743334","0.782156368222","0.999819956768","0.999604172550","0.890988162495","0.805624107027","0.796660512162","0.787894185271","0.662041360907","0.796797522188","0.796599716782","0.000395827450","0.796462473806","0.982918530193","0.962014249878","0.972354049666","0.972354049666"
+"SNOWBALL_YIDDISH_LUCENE_FILTER","YI","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","802","3578","0","3578","0","1","3578","1087","4962","1151","1382","6391758","1151","6392909","0.018004","1382","6344","21.784363","0.811712743334","0.782156368222","0.999819956768","0.999604172550","0.890988162495","0.805624107027","0.796660512162","0.787894185271","0.662041360907","0.796797522188","0.796599716782","0.000395827450","0.796462473806","0.982918530193","0.962014249878","0.972354049666","0.972354049666"
+"SPANISH_LUCENE_SPANISH_LIGHT_STEM_FILTER","ES_ES","ALL_WORDS","PRIMARY_OUTPUT","65059","871332","3589","871332","0","1","871332","405552","1244317","147956","40729019","379567170154","147956","379567318110","0.000039","40729019","41973336","97.035458","0.893730611741","0.029645415842","0.999999610198","0.999892318297","0.514822513020","0.130863846499","0.057387272020","0.036752000024","0.029541282827","0.162772895888","0.162762055080","0.000107681703","0.057380579619","0.993823553768","0.756690454887","0.859195405761","0.859195405761"
+"SPANISH_LUCENE_SPANISH_LIGHT_STEM_FILTER","ES_ES","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","64918","869371","3525","869371","0","1","869371","404617","1241848","146613","40621522","377860523152","146613","377860669765","0.000039","40621522","41863370","97.033569","0.894406108634","0.029664310351","0.999999611992","0.999892119974","0.514831961171","0.130949068412","0.057424066047","0.036775459718","0.029560783207","0.162886280533","0.162875426941","0.000107880026","0.057417362063","0.993866319748","0.756725425267","0.859233931168","0.859233931168"
+"SPANISH_LUCENE_SPANISH_MINIMAL_STEM_FILTER","ES_ES","ALL_WORDS","PRIMARY_OUTPUT","65059","871332","3589","871332","0","1","871332","718633","148463","47859","41824873","379567270251","47859","379567318110","0.000013","41824873","41973336","99.646292","0.756221921130","0.003537078873","0.999999873912","0.999889695187","0.501768476392","0.017360591398","0.007041223811","0.004416184633","0.003533050405","0.051718628951","0.051713939443","0.000110304813","0.007040201537","0.995635307295","0.710609719902","0.829315972283","0.829315972283"
+"SPANISH_LUCENE_SPANISH_MINIMAL_STEM_FILTER","ES_ES","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","64918","869371","3525","869371","0","1","869371","717093","148226","47148","41715144","377860622617","47148","377860669765","0.000012","41715144","41863370","99.645929","0.758678227400","0.003540708739","0.999999875224","0.999889489251","0.501770291981","0.017379114288","0.007048522419","0.004420728101","0.003536725554","0.051829129163","0.051824445330","0.000110510749","0.007047500484","0.995675746140","0.710626433954","0.829341382913","0.829341382913"
+"SPANISH_LUCENE_SPANISH_PLURAL_STEM_FILTER","ES_ES","ALL_WORDS","PRIMARY_OUTPUT","65059","871332","3589","871332","0","1","871332","578805","325245","58578","41648091","379567259532","58578","379567318110","0.000015","41648091","41973336","99.225115","0.847382777999","0.007748847983","0.999999845672","0.999890132644","0.503874346827","0.037377069210","0.015357262275","0.009663967067","0.007738048760","0.081032341260","0.081026288458","0.000109867356","0.015355289170","0.995442321769","0.723731297627","0.838115191065","0.838115191065"
+"SPANISH_LUCENE_SPANISH_PLURAL_STEM_FILTER","ES_ES","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","64918","869371","3525","869371","0","1","869371","577533","324656","57716","41538714","377860612049","57716","377860669765","0.000015","41538714","41863370","99.224487","0.849057985417","0.007755132948","0.999999847256","0.999889928152","0.503877490102","0.037408921072","0.015369880354","0.009671831022","0.007744455857","0.081145286724","0.081139234944","0.000110071848","0.015367905834","0.995484117647","0.723752971494","0.838144538380","0.838144538380"
+"SPANISH_RADIXOR","ES_ES","ALL_WORDS","PRIMARY_OUTPUT","65059","871332","3589","871332","0","1","871332","64995","41074684","288483","898652","379567029627","288483","379567318110","0.000076","898652","41973336","2.141007","0.993025606574","0.978589931475","0.999999239969","0.999996872745","0.989294585722","0.990104500109","0.985754921826","0.981443392220","0.971909988067","0.985781345071","0.985779787115","0.000003127255","0.985753358111","0.995417814373","0.993266303762","0.994340895233","0.994340895233"
+"SPANISH_RADIXOR","ES_ES","ALL_WORDS","ANY_CANDIDATE","65059","871332","3589","828695","42637","21","916797","65118","41972710","2","626","379567318108","2","379567318110","0.000000","626","41973336","0.001491","0.999999952350","0.999985085770","0.999999999995","0.999999998346","0.999992542882","0.999996978999","0.999992519005","0.999988059050","0.999985038121","0.999992519032","0.999992518205","0.000000001654","","","","",""
+"SPANISH_RADIXOR","ES_ES","ALL_WORDS","ALL_CANDIDATES","65059","871332","3589","828695","42637","21","916797","65118","41972710","1349800","626","379565968310","1349800","379567318110","0.000356","626","41973336","0.001491","0.968842987168","0.999985085770","0.999996443846","0.999996442590","0.999990764808","0.974915259157","0.984167740127","0.993597526064","0.968828987818","0.984290880594","0.984289129583","0.000003557410","","","","",""
+"SPANISH_RADIXOR","ES_ES","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","64918","869371","3525","869371","0","1","869371","64814","40978337","276044","885033","377860393721","276044","377860669765","0.000073","885033","41863370","2.114099","0.993308734895","0.978859012067","0.999999269456","0.999996927576","0.989429140761","0.990384762162","0.986030938205","0.981715226337","0.972446769193","0.986057405488","0.986055874970","0.000003072424","0.986029401906","0.995463637710","0.993323040564","0.994392187139","0.994392187139"
+"SPANISH_RADIXOR","ES_ES","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","64918","869371","3525","826968","42403","21","914127","64933","41863370","0","0","377860669765","0","377860669765","0.000000","0","41863370","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"SPANISH_RADIXOR","ES_ES","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","64918","869371","3525","826968","42403","21","914127","64933","41863370","1255381","0","377859414384","1255381","377860669765","0.000332","0","41863370","0.000000","0.970885497124","1.000000000000","0.999996677662","0.999996678030","0.999998338831","0.976571978660","0.985227704543","0.994038240493","0.970885497124","0.985335220686","0.985333583876","0.000003321970","","","","",""
+"SWEDISH_LUCENE_SWEDISH_LIGHT_STEM_FILTER","SV_SE","ALL_WORDS","PRIMARY_OUTPUT","12371","98108","68","98108","0","1","98108","22392","218635","45941","166707","4812109495","45941","4812155436","0.000955","166707","385342","43.262089","0.826359911708","0.567379107390","0.999990453135","0.999955813777","0.783684780262","0.757232036109","0.672807954234","0.605320541501","0.506940918144","0.684733049508","0.684712936280","0.000044186223","0.672786622564","0.986795482859","0.942302523776","0.964035907715","0.964035907715"
+"SWEDISH_LUCENE_SWEDISH_LIGHT_STEM_FILTER","SV_SE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","12342","97881","68","97881","0","1","97881","22338","218126","45941","166437","4789865636","45941","4789911577","0.000959","166437","384563","43.279515","0.826025213298","0.567204853301","0.999990408800","0.999955664954","0.783597631051","0.756945124029","0.672574503184","0.605125951621","0.506675896159","0.684489232882","0.684469049184","0.000044335046","0.672553099274","0.986761366955","0.942264620928","0.963999791833","0.963999791833"
+"SWEDISH_LUCENE_SWEDISH_MINIMAL_STEM_FILTER","SV_SE","ALL_WORDS","PRIMARY_OUTPUT","12371","98108","68","98108","0","1","98108","23360","228181","40227","157161","4812115209","40227","4812155436","0.000836","157161","385342","40.784809","0.850127417961","0.592151906618","0.999991640544","0.999958984659","0.796071773581","0.781991317186","0.698068068834","0.630412272016","0.536178622033","0.709510092538","0.709491456160","0.000041015341","0.698048215965","0.988492665376","0.944581755622","0.966038479572","0.966038479572"
+"SWEDISH_LUCENE_SWEDISH_MINIMAL_STEM_FILTER","SV_SE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","12342","97881","68","97881","0","1","97881","23312","227624","40227","156939","4789871350","40227","4789911577","0.000840","156939","384563","40.809698","0.849815755775","0.591903017191","0.999991601724","0.999958840540","0.795947309457","0.781693541131","0.697790053555","0.630152322431","0.535850655618","0.709230928471","0.709212225121","0.000041159460","0.697770131116","0.988462934404","0.944527865197","0.965996098328","0.965996098328"
+"SWEDISH_RADIXOR","SV_SE","ALL_WORDS","PRIMARY_OUTPUT","12371","98108","68","98108","0","1","98108","12330","365796","24473","19546","4812130963","24473","4812155436","0.000509","19546","385342","5.072377","0.937291970410","0.949276227351","0.999994914337","0.999990853272","0.974635570844","0.939664553041","0.943246034417","0.946854921499","0.892588119029","0.943265066457","0.943260495884","0.000009146728","0.943241460869","0.992630770222","0.993394969179","0.993012722673","0.993012722673"
+"SWEDISH_RADIXOR","SV_SE","ALL_WORDS","ANY_CANDIDATE","12371","98108","68","92341","5767","5","104148","12371","385342","0","0","4812155436","0","4812155436","0.000000","0","385342","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"SWEDISH_RADIXOR","SV_SE","ALL_WORDS","ALL_CANDIDATES","12371","98108","68","92341","5767","5","104148","12371","385342","47848","0","4812107588","47848","4812155436","0.000994","0","385342","0.000000","0.889545003347","1.000000000000","0.999990056847","0.999990057643","0.999995028423","0.909639856815","0.941544130223","0.975767741439","0.889545003347","0.943156934634","0.943152245645","0.000009942357","","","","",""
+"SWEDISH_RADIXOR","SV_SE","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","12342","97881","68","97881","0","1","97881","12301","365017","24473","19546","4789887104","24473","4789911577","0.000511","19546","384563","5.082652","0.937166551131","0.949173477428","0.999994890720","0.999990810798","0.974584184074","0.939543572972","0.943131801052","0.946747541943","0.892383555482","0.943150907472","0.943146315681","0.000009189202","0.943127206266","0.992611730682","0.993377890892","0.992994663001","0.992994663001"
+"SWEDISH_RADIXOR","SV_SE","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","12342","97881","68","92114","5767","5","103921","12342","384563","0","0","4789911577","0","4789911577","0.000000","0","384563","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"SWEDISH_RADIXOR","SV_SE","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","12342","97881","68","92114","5767","5","103921","12342","384563","47848","0","4789863729","47848","4789911577","0.000999","0","384563","0.000000","0.889346015712","1.000000000000","0.999990010672","0.999990011473","0.999995005336","0.909473386475","0.941432652692","0.975719846569","0.889346015712","0.943051438529","0.943046728292","0.000009988527","","","","",""
+"UKRAINIAN_LUCENE_MORFOLOGIK_FILTER","UK_UA","ALL_WORDS","PRIMARY_OUTPUT","1493","14245","4","14245","0","1","14245","2358","56032","828","9308","101386722","828","101387550","0.000817","9308","65340","14.245485","0.985437917693","0.857545148454","0.999991833317","0.999900091560","0.928768490886","0.956895962839","0.917054009820","0.880397209478","0.846814169992","0.919270093835","0.919222898475","0.000099908440","0.917004266345","0.997989675681","0.970999849348","0.984309781661","0.984309781661"
+"UKRAINIAN_LUCENE_MORFOLOGIK_FILTER","UK_UA","ALL_WORDS","ANY_CANDIDATE","1493","14245","4","12038","2207","6","16937","2912","60394","122","4946","101387428","122","101387550","0.000120","4946","65340","7.569636","0.997984004230","0.924303642485","0.999998796696","0.999950045780","0.962151219591","0.982322936592","0.959731756929","0.938156308641","0.922581039382","0.960437530635","0.960413432420","0.000049954220","","","","",""
+"UKRAINIAN_LUCENE_MORFOLOGIK_FILTER","UK_UA","ALL_WORDS","ALL_CANDIDATES","1493","14245","4","12038","2207","6","16937","2912","60394","1368","4946","101386182","1368","101387550","0.001349","4946","65340","7.569636","0.977850458211","0.924303642485","0.999986507219","0.999937764217","0.962145074852","0.966650447520","0.950323362339","0.934538657225","0.905348683816","0.950700131656","0.950669478973","0.000062235783","","","","",""
+"UKRAINIAN_LUCENE_MORFOLOGIK_FILTER","UK_UA","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","1491","14236","4","14236","0","1","14236","2356","56016","828","9308","101258578","828","101259406","0.000818","9308","65324","14.248974","0.985433818873","0.857510256567","0.999991822982","0.999899965191","0.928751039775","0.956884181756","0.917032283413","0.880367133966","0.846777119361","0.919249480202","0.919202225823","0.000100034809","0.916982477117","0.997988093697","0.970977645923","0.984297603943","0.984297603943"
+"UKRAINIAN_LUCENE_MORFOLOGIK_FILTER","UK_UA","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","1491","14236","4","12029","2207","6","16928","2910","60378","122","4946","101259284","122","101259406","0.000120","4946","65324","7.571490","0.997983471074","0.924285101953","0.999998795174","0.999949982596","0.962141948563","0.982318335047","0.959721515768","0.938140934008","0.922562112276","0.960427641371","0.960403512884","0.000050017404","","","","",""
+"UKRAINIAN_LUCENE_MORFOLOGIK_FILTER","UK_UA","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","1491","14236","4","12029","2207","6","16928","2910","60378","1368","4946","101258038","1368","101259406","0.001351","4946","65324","7.571490","0.977844718686","0.924285101953","0.999986490144","0.999937685499","0.962135796049","0.966641904786","0.950310852286","0.934522445998","0.905325976129","0.950687806541","0.950657115187","0.000062314501","","","","",""
+"UKRAINIAN_MORFOLOGIK_DIRECT","UK_UA","ALL_WORDS","PRIMARY_OUTPUT","1493","14245","4","14245","0","1","14245","2365","56016","828","9324","101386722","828","101387550","0.000817","9324","65340","14.269972","0.985433818873","0.857300275482","0.999991833317","0.999899933851","0.928646054399","0.956831877998","0.916912197996","0.880190066750","0.846572361262","0.919136923635","0.919089660097","0.000100066149","0.916862376978","0.997989675681","0.970875945953","0.984246115917","0.984246115917"
+"UKRAINIAN_MORFOLOGIK_DIRECT","UK_UA","ALL_WORDS","ANY_CANDIDATE","1493","14245","4","12038","2207","6","16937","2919","60378","122","4962","101387428","122","101387550","0.000120","4962","65340","7.594123","0.997983471074","0.924058769513","0.999998796696","0.999949888071","0.962028783105","0.982267195939","0.959599491418","0.937954390108","0.922336622774","0.960310042786","0.960285871670","0.000050111929","","","","",""
+"UKRAINIAN_MORFOLOGIK_DIRECT","UK_UA","ALL_WORDS","ALL_CANDIDATES","1493","14245","4","12038","2207","6","16937","2919","60378","1368","4962","101386182","1368","101387550","0.001349","4962","65340","7.594123","0.977844718686","0.924058769513","0.999986507219","0.999937606509","0.962022638366","0.966592384831","0.950191209102","0.934337338211","0.905108832524","0.950571400540","0.950540673331","0.000062393491","","","","",""
+"UKRAINIAN_MORFOLOGIK_DIRECT","UK_UA","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","1491","14236","4","14236","0","1","14236","2356","56016","828","9308","101258578","828","101259406","0.000818","9308","65324","14.248974","0.985433818873","0.857510256567","0.999991822982","0.999899965191","0.928751039775","0.956884181756","0.917032283413","0.880367133966","0.846777119361","0.919249480202","0.919202225823","0.000100034809","0.916982477117","0.997988093697","0.970977645923","0.984297603943","0.984297603943"
+"UKRAINIAN_MORFOLOGIK_DIRECT","UK_UA","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","1491","14236","4","12029","2207","6","16928","2910","60378","122","4946","101259284","122","101259406","0.000120","4946","65324","7.571490","0.997983471074","0.924285101953","0.999998795174","0.999949982596","0.962141948563","0.982318335047","0.959721515768","0.938140934008","0.922562112276","0.960427641371","0.960403512884","0.000050017404","","","","",""
+"UKRAINIAN_MORFOLOGIK_DIRECT","UK_UA","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","1491","14236","4","12029","2207","6","16928","2910","60378","1368","4946","101258038","1368","101259406","0.001351","4946","65324","7.571490","0.977844718686","0.924285101953","0.999986490144","0.999937685499","0.962135796049","0.966641904786","0.950310852286","0.934522445998","0.905325976129","0.950687806541","0.950657115187","0.000062314501","","","","",""
+"UKRAINIAN_RADIXOR","UK_UA","ALL_WORDS","PRIMARY_OUTPUT","1493","14245","4","14245","0","1","14245","1493","64732","880","608","101386670","880","101387550","0.000868","608","65340","0.930517","0.986587819301","0.990694827058","0.999991320433","0.999985333094","0.995343073746","0.987406494442","0.988637057853","0.989870692292","0.977529447297","0.988639190514","0.988631855097","0.000014666906","0.988629719696","0.997993591453","0.998265712624","0.998129633491","0.998129633491"
+"UKRAINIAN_RADIXOR","UK_UA","ALL_WORDS","ANY_CANDIDATE","1493","14245","4","14055","190","2","14435","1493","65340","0","0","101387550","0","101387550","0.000000","0","65340","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"UKRAINIAN_RADIXOR","UK_UA","ALL_WORDS","ALL_CANDIDATES","1493","14245","4","14055","190","2","14435","1493","65340","1490","0","101386060","1490","101387550","0.001470","0","65340","0.000000","0.977704623672","1.000000000000","0.999985303916","0.999985313380","0.999992651958","0.982083809295","0.988726639933","0.995459946982","0.977704623672","0.988789473888","0.988782208195","0.000014686620","","","","",""
+"UKRAINIAN_RADIXOR","UK_UA","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","1491","14236","4","14236","0","1","14236","1491","64716","880","608","101258526","880","101259406","0.000869","608","65324","0.930745","0.986584547838","0.990692547915","0.999991309449","0.999985314543","0.995341928682","0.987403420118","0.988634280477","0.989868213355","0.977524016676","0.988636414174","0.988629069474","0.000014685457","0.988626933033","0.997992012551","0.998264347489","0.998128161444","0.998128161444"
+"UKRAINIAN_RADIXOR","UK_UA","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","1491","14236","4","14046","190","2","14426","1491","65324","0","0","101259406","0","101259406","0.000000","0","65324","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"UKRAINIAN_RADIXOR","UK_UA","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","1491","14236","4","14046","190","2","14426","1491","65324","1490","0","101257916","1490","101259406","0.001471","0","65324","0.000000","0.977699284581","1.000000000000","0.999985285318","0.999985294804","0.999992642659","0.982079499669","0.988723909852","0.995458840023","0.977699284581","0.988786774073","0.988779499204","0.000014705196","","","","",""
+"YI_RADIXOR","YI","ALL_WORDS","PRIMARY_OUTPUT","802","3578","0","3578","0","1","3578","802","6195","195","149","6392714","195","6392909","0.003050","149","6344","2.348676","0.969483568075","0.976513240858","0.999969497454","0.999946243726","0.988241369156","0.970881394183","0.972985707555","0.975099162627","0.947392567671","0.972992055990","0.972965163911","0.000053756274","0.972958803000","0.995691103897","0.996142223728","0.995916612726","0.995916612726"
+"YI_RADIXOR","YI","ALL_WORDS","ANY_CANDIDATE","802","3578","0","3489","89","3","3676","802","6344","0","0","6392909","0","6392909","0.000000","0","6344","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"YI_RADIXOR","YI","ALL_WORDS","ALL_CANDIDATES","802","3578","0","3489","89","3","3676","802","6344","389","0","6392520","389","6392909","0.006085","0","6344","0.000000","0.942224862617","1.000000000000","0.999939151332","0.999939211655","0.999969575666","0.953239572064","0.970253116158","0.987885016662","0.942224862617","0.970682678643","0.970653145819","0.000060788345","","","","",""
+"YI_RADIXOR","YI","LOWERCASE_GROUPS_ONLY","PRIMARY_OUTPUT","802","3578","0","3578","0","1","3578","802","6195","195","149","6392714","195","6392909","0.003050","149","6344","2.348676","0.969483568075","0.976513240858","0.999969497454","0.999946243726","0.988241369156","0.970881394183","0.972985707555","0.975099162627","0.947392567671","0.972992055990","0.972965163911","0.000053756274","0.972958803000","0.995691103897","0.996142223728","0.995916612726","0.995916612726"
+"YI_RADIXOR","YI","LOWERCASE_GROUPS_ONLY","ANY_CANDIDATE","802","3578","0","3489","89","3","3676","802","6344","0","0","6392909","0","6392909","0.000000","0","6344","0.000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","1.000000000000","0.000000000000","","","","",""
+"YI_RADIXOR","YI","LOWERCASE_GROUPS_ONLY","ALL_CANDIDATES","802","3578","0","3489","89","3","3676","802","6344","389","0","6392520","389","6392909","0.006085","0","6344","0.000000","0.942224862617","1.000000000000","0.999939151332","0.999939211655","0.999969575666","0.953239572064","0.970253116158","0.987885016662","0.942224862617","0.970682678643","0.970653145819","0.000060788345","","","","",""

docs/benchmarks/data/stemming-quality.sha256

@@ -0,0 +1 @@
+5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28  stemming-quality.csv

docs/benchmarks/index.md

@@ -0,0 +1,327 @@
+# Benchmark Results
+
+This section contains the published Radixor benchmark reference set. It is intentionally split into
+two layers:
+
+- **benchmark reference pages**, which explain methodology, corpora, environment, candidate
+  selection, and the English dictionary coverage experiment;
+- **language result pages**, which contain the actual same-language accuracy and throughput tables.
+- **pairwise quality pages and generated sections**, which publish over-stemming, under-stemming,
+  candidate-policy, classification, and partition measurements from one checked result snapshot.
+
+This structure keeps methodology separate from per-language result pages, while preserving all
+measured data and the command-class analysis for each Radixor language resource.
+
+## Read This First
+
+Start with [Benchmarking](../benchmarking.md) for the high-level interpretation model. The most
+important rule is that speed and exact-root quality must be read together. Many competing stemmers
+are intentionally light, minimal, or aggressive; they can be fast because they are not trying to
+match dictionary roots with the same precision.
+
+Radixor rows in the refreshed tables use contracted compiled patch tries. Contraction collapses
+uniform preferred-command subtrees into accepting leaves, reducing hot lookup depth while preserving
+the preferred result measured by the accuracy pass.
+
+## Reference Pages
+
+| Page | Purpose |
+| --- | --- |
+| [Methodology](reference/methodology.md) | Workload design, normalization, speed metrics, quality metrics, and interpretation rules. |
+| [Linguistic quality methodology](reference/linguistic-quality.md) | Gold-standard groups, output policies, pairwise formulas, ranking rules, aggregation, and limitations. |
+| [Tested stemmers](reference/tested-stemmers.md) | Versions, upstream attribution, evaluated coverage, adapters, preprocessing, and output capability. |
+| [Reproducibility and raw data](reference/reproducibility.md) | Commands, versioned CSV snapshot, checksum, generated artifacts, and unavailable provenance. |
+| [Corpora](reference/corpora.md) | Dictionary row counts, complete quality tokens, already-root tokens, changed speed tokens, and timing token counts. |
+| [Environment and reports](reference/environment.md) | Hardware, JVM, JMH settings, report files, and badge/report policy. |
+| [English dictionary coverage](reference/english-coverage.md) | Quality/speed operating curve for contracted Radixor tries built from 100% down to 10% of English dictionary rows. |
+| [Candidate evaluation](reference/candidates.md) | Included benchmark families and evaluated candidates that were skipped. |
+
+## Language Results
+
+Each language page contains:
+
+- the dictionary corpus size,
+- the Radixor patch-command distribution,
+- exact-root quality metrics,
+- throughput metrics,
+- interpretation notes for the compared stemmers.
+
+Open [Language Benchmark Pages](languages/index.md) for the complete language list.
+
+## Key Published Result
+
+The English dictionary coverage benchmark shows the current contracted-trie operating curve. With
+the full English dictionary, Radixor reaches `97.478%` all-token exactness and `97.197%`
+changed-token exactness at `135.8 ns/token`. Even with a deterministic 10% dictionary slice, it
+keeps `92.868%` all-token exactness and `76.516%` changed-token exactness at `86.0 ns/token`.
+
+Those figures should not be reduced to a single speed badge. The professional interpretation is a
+quality/speed envelope: the amount and quality of dictionary knowledge affect stemming precision,
+while contracted tries reduce lookup cost in uniform regions of the compiled graph.
+
+## Quality versus performance
+
+Each language page keeps exact-root accuracy, JMH latency, and pairwise linguistic-quality results in separate tables. No undocumented scalar combines them. The current repository checkout does not contain the dated machine-readable JMH CSV files named by the performance provenance page, so this revision preserves the existing performance tables but does not regenerate a cross-language Pareto frontier from rounded Markdown values. A defensible Pareto analysis requires the original unrounded JMH snapshot on the same hardware and JVM. Readers can still inspect the quality and speed dimensions side by side on every language page.
+
+<!-- STEMMING-QUALITY-OVERVIEW:START -->
+
+## Pairwise Quality Findings
+
+The validated snapshot is a broad multilingual comparison covering the complete 20-language Radixor dictionary universe; 19 languages have existing benchmark pages. The direct ranking below uses only deterministic `PRIMARY_OUTPUT` rows over identical per-language inputs. Candidate-aware rows are intentionally excluded from this claim.
+
+!!! success "Evidence-based primary-output result"
+    Radixor achieved the highest balanced accuracy among the evaluated deterministic stemmers for every documented language in both `ALL_WORDS` and `LOWERCASE_GROUPS_ONLY`: **38 wins in 38 language-mode comparisons, with no exact first-place ties**. This statement is limited to the evaluated implementations, versions, dictionaries, adapters, and balanced-accuracy metric; it is not a universal claim about every stemming use case.
+
+### Per-language winner matrix
+
+| Language | Dictionary mode | Winner | Balanced accuracy | Runner-up | Difference | Exact tie | Deterministic stemmers |
+|---|---|---|---:|---|---:|---|---:|
+|Czech (`CS_CZ`)|ALL_WORDS|Radixor|0.996565|HUNSPELL CZECH LUCENE FILTER|0.142812638|no|3|
+|Czech (`CS_CZ`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997139|HUNSPELL CZECH LUCENE FILTER|0.144369049|no|3|
+|Danish (`DA_DK`)|ALL_WORDS|Radixor|0.996066|SNOWBALL DANISH LUCENE FILTER|0.058096771|no|3|
+|Danish (`DA_DK`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996305|SNOWBALL DANISH DIRECT|0.058230346|no|3|
+|Dutch (`NL_NL`)|ALL_WORDS|Radixor|0.988661|SNOWBALL DUTCH DIRECT|0.261574077|no|4|
+|Dutch (`NL_NL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.989040|SNOWBALL DUTCH DIRECT|0.258544404|no|4|
+|English (`US_UK`)|ALL_WORDS|Radixor|0.965159|ENGLISH LUCENE PORTER COPIED|0.010532535|no|11|
+|English (`US_UK`)|LOWERCASE_GROUPS_ONLY|Radixor|0.965820|ENGLISH LUCENE PORTER COPIED|0.010920064|no|11|
+|Finnish (`FI_FI`)|ALL_WORDS|Radixor|0.984594|SNOWBALL FINNISH LUCENE FILTER|0.244241861|no|4|
+|Finnish (`FI_FI`)|LOWERCASE_GROUPS_ONLY|Radixor|0.988068|SNOWBALL FINNISH DIRECT|0.249668284|no|4|
+|French (`FR_FR`)|ALL_WORDS|Radixor|0.956992|SNOWBALL FRENCH DIRECT|0.111730673|no|6|
+|French (`FR_FR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.957224|SNOWBALL FRENCH DIRECT|0.111809799|no|6|
+|German (`DE_DE`)|ALL_WORDS|Radixor|0.907901|GERMAN CISTEM|0.027131083|no|8|
+|German (`DE_DE`)|LOWERCASE_GROUPS_ONLY|Radixor|0.966157|GERMAN CISTEM|0.050868631|no|8|
+|Hungarian (`HU_HU`)|ALL_WORDS|Radixor|0.995491|SNOWBALL HUNGARIAN LUCENE FILTER|0.172884951|no|4|
+|Hungarian (`HU_HU`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996163|SNOWBALL HUNGARIAN DIRECT|0.174455479|no|4|
+|Italian (`IT_IT`)|ALL_WORDS|Radixor|0.996507|SNOWBALL ITALIAN DIRECT|0.130318040|no|4|
+|Italian (`IT_IT`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996512|SNOWBALL ITALIAN DIRECT|0.130307087|no|4|
+|Norwegian Bokmal (`NB_NO`)|ALL_WORDS|Radixor|0.974783|SNOWBALL NORWEGIAN BOKMAL DIRECT|0.099819340|no|5|
+|Norwegian Bokmal (`NB_NO`)|LOWERCASE_GROUPS_ONLY|Radixor|0.975000|SNOWBALL NORWEGIAN BOKMAL DIRECT|0.100008544|no|5|
+|Norwegian Nynorsk (`NN_NO`)|ALL_WORDS|Radixor|0.935777|SNOWBALL NORWEGIAN NYNORSK DIRECT|0.076868986|no|3|
+|Norwegian Nynorsk (`NN_NO`)|LOWERCASE_GROUPS_ONLY|Radixor|0.935853|SNOWBALL NORWEGIAN NYNORSK DIRECT|0.076816096|no|3|
+|Persian (`FA_IR`)|ALL_WORDS|Radixor|0.974922|PERSIAN LUCENE PERSIAN STEM FILTER|0.472751327|no|2|
+|Persian (`FA_IR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.974922|PERSIAN LUCENE PERSIAN STEM FILTER|0.472751327|no|2|
+|Polish (`PL_PL`)|ALL_WORDS|Radixor|0.990388|POLISH LUCENE MORFOLOGIK FILTER|0.042233990|no|5|
+|Polish (`PL_PL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.990579|POLISH LUCENE MORFOLOGIK FILTER|0.042401633|no|5|
+|Portuguese (`PT_PT`)|ALL_WORDS|Radixor|0.998502|SNOWBALL PORTUGUESE DIRECT|0.059701750|no|6|
+|Portuguese (`PT_PT`)|LOWERCASE_GROUPS_ONLY|Radixor|0.998502|SNOWBALL PORTUGUESE DIRECT|0.059701750|no|6|
+|Russian (`RU_RU`)|ALL_WORDS|Radixor|0.989827|SNOWBALL RUSSIAN LUCENE FILTER|0.154951419|no|4|
+|Russian (`RU_RU`)|LOWERCASE_GROUPS_ONLY|Radixor|0.989852|SNOWBALL RUSSIAN DIRECT|0.154997931|no|4|
+|Spanish (`ES_ES`)|ALL_WORDS|Radixor|0.989295|SNOWBALL SPANISH LUCENE FILTER|0.336680479|no|7|
+|Spanish (`ES_ES`)|LOWERCASE_GROUPS_ONLY|Radixor|0.989429|SNOWBALL SPANISH DIRECT|0.336708826|no|7|
+|Swedish (`SV_SE`)|ALL_WORDS|Radixor|0.974636|SNOWBALL SWEDISH DIRECT|0.167101450|no|5|
+|Swedish (`SV_SE`)|LOWERCASE_GROUPS_ONLY|Radixor|0.974584|SNOWBALL SWEDISH DIRECT|0.166984893|no|5|
+|Ukrainian (`UK_UA`)|ALL_WORDS|Radixor|0.995343|UKRAINIAN LUCENE MORFOLOGIK FILTER|0.066574583|no|4|
+|Ukrainian (`UK_UA`)|LOWERCASE_GROUPS_ONLY|Radixor|0.995342|UKRAINIAN LUCENE MORFOLOGIK FILTER|0.066590889|no|4|
+|Yiddish (`YI`)|ALL_WORDS|Radixor|0.988241|SNOWBALL YIDDISH DIRECT|0.097253207|no|3|
+|Yiddish (`YI`)|LOWERCASE_GROUPS_ONLY|Radixor|0.988241|SNOWBALL YIDDISH DIRECT|0.097253207|no|3|
+
+### Secondary-metric trade-offs
+
+Balanced-accuracy leadership does not imply leadership on every error trade-off. The table below lists all **15** deterministic primary-output language-mode-metric cases where a non-Radixor adapter has the best displayed value. Equal values are resolved by the authoritative row ordering and should be read as ties when the unrounded values are equal. Throughput leadership remains in the separate performance tables.
+
+<details class="quality-details" markdown="1"><summary>Non-Radixor secondary-metric leaders</summary>
+
+| Language | Dictionary mode | Metric | Leader | Value |
+|---|---|---|---|---:|
+|English|ALL_WORDS|Over-stemming percentage|ENGLISH LUCENE POSSESSIVE FILTER|0.000604|
+|English|LOWERCASE_GROUPS_ONLY|Over-stemming percentage|ENGLISH LUCENE POSSESSIVE FILTER|0.000653|
+|French|ALL_WORDS|Over-stemming percentage|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|0.000177|
+|French|LOWERCASE_GROUPS_ONLY|Over-stemming percentage|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|0.000166|
+|German|LOWERCASE_GROUPS_ONLY|Over-stemming percentage|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|0.000188|
+|Italian|ALL_WORDS|Over-stemming percentage|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|0.000020|
+|Italian|LOWERCASE_GROUPS_ONLY|Over-stemming percentage|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|0.000020|
+|Persian|ALL_WORDS|Over-stemming percentage|PERSIAN LUCENE PERSIAN STEM FILTER|0.002652|
+|Persian|LOWERCASE_GROUPS_ONLY|Over-stemming percentage|PERSIAN LUCENE PERSIAN STEM FILTER|0.002652|
+|Portuguese|ALL_WORDS|Over-stemming percentage|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|0.000003|
+|Portuguese|LOWERCASE_GROUPS_ONLY|Over-stemming percentage|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|0.000003|
+|Spanish|ALL_WORDS|Over-stemming percentage|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|0.000013|
+|Spanish|LOWERCASE_GROUPS_ONLY|Over-stemming percentage|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|0.000012|
+|Ukrainian|ALL_WORDS|Over-stemming percentage|HUNSPELL UKRAINIAN LUCENE FILTER|0.000783|
+|Ukrainian|LOWERCASE_GROUPS_ONLY|Over-stemming percentage|HUNSPELL UKRAINIAN LUCENE FILTER|0.000784|
+
+</details>
+
+### Win, tie, and placement summary
+
+Counts use `PRIMARY_OUTPUT` only and retain each adapter configuration as a separate stemmer except that language-specific Radixor identifiers are combined as Radixor. Coverage is displayed explicitly; unsupported languages are absent, not losses.
+
+<details class="quality-details" markdown="1"><summary>ALL_WORDS placements</summary>
+
+| Stemmer | Evaluated languages | Wins | Exact first-place ties | Top-three placements | Average rank | Median rank |
+|---|---:|---:|---:|---:|---:|---:|
+|Radixor|19|19|0|19|1.000|1.000|
+|CZECH LUCENE CZECH STEM FILTER|1|0|0|1|3.000|3.000|
+|ENGLISH LUCENE KSTEM FILTER|1|0|0|0|8.000|8.000|
+|ENGLISH LUCENE MINIMAL FILTER|1|0|0|0|9.000|9.000|
+|ENGLISH LUCENE PORTER COPIED|1|0|0|1|2.000|2.000|
+|ENGLISH LUCENE PORTER FILTER|1|0|0|1|3.000|3.000|
+|ENGLISH LUCENE POSSESSIVE FILTER|1|0|0|0|11.000|11.000|
+|ENGLISH OPENNLP PORTER|1|0|0|0|4.000|4.000|
+|ENGLISH PAICE HUSK LANCASTER|1|0|0|0|7.000|7.000|
+|ENGLISH SNOWBALL ORIGINAL PORTER|1|0|0|0|6.000|6.000|
+|ENGLISH SNOWBALL PORTER2|1|0|0|0|5.000|5.000|
+|FINNISH LUCENE FINNISH LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|FRENCH LUCENE FRENCH LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
+|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|1|0|0|0|6.000|6.000|
+|GERMAN CISTEM|1|0|0|1|2.000|2.000|
+|GERMAN LUCENE GERMAN LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
+|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|1|0|0|0|8.000|8.000|
+|GERMAN LUCENE GERMAN STEM FILTER|1|0|0|0|6.000|6.000|
+|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|HUNSPELL CZECH LUCENE FILTER|1|0|0|1|2.000|2.000|
+|HUNSPELL DUTCH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|HUNSPELL ENGLISH LUCENE FILTER|1|0|0|0|10.000|10.000|
+|HUNSPELL FRENCH LUCENE FILTER|1|0|0|0|4.000|4.000|
+|HUNSPELL GERMAN LUCENE FILTER|1|0|0|0|7.000|7.000|
+|HUNSPELL POLISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|HUNSPELL SPANISH LUCENE FILTER|1|0|0|0|4.000|4.000|
+|HUNSPELL UKRAINIAN LUCENE FILTER|1|0|0|0|4.000|4.000|
+|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|1|0|0|0|5.000|5.000|
+|PERSIAN LUCENE PERSIAN STEM FILTER|1|0|0|1|2.000|2.000|
+|POLISH LUCENE MORFOLOGIK FILTER|1|0|0|1|2.000|2.000|
+|POLISH LUCENE STEMPEL DIRECT|1|0|0|0|4.000|4.000|
+|POLISH LUCENE STEMPEL FILTER|1|0|0|0|5.000|5.000|
+|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
+|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|1|0|0|0|6.000|6.000|
+|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|1|0|0|0|4.000|4.000|
+|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL DANISH DIRECT|1|0|0|1|3.000|3.000|
+|SNOWBALL DANISH LUCENE FILTER|1|0|0|1|2.000|2.000|
+|SNOWBALL DUTCH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL DUTCH LUCENE FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL FINNISH DIRECT|1|0|0|1|3.000|3.000|
+|SNOWBALL FINNISH LUCENE FILTER|1|0|0|1|2.000|2.000|
+|SNOWBALL FRENCH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL FRENCH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL GERMAN DIRECT|1|0|0|1|3.000|3.000|
+|SNOWBALL GERMAN LUCENE FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL HUNGARIAN DIRECT|1|0|0|1|3.000|3.000|
+|SNOWBALL HUNGARIAN LUCENE FILTER|1|0|0|1|2.000|2.000|
+|SNOWBALL ITALIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL ITALIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL NORWEGIAN BOKMAL DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL NORWEGIAN NYNORSK DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL PORTUGUESE DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL PORTUGUESE LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL RUSSIAN DIRECT|1|0|0|1|3.000|3.000|
+|SNOWBALL RUSSIAN LUCENE FILTER|1|0|0|1|2.000|2.000|
+|SNOWBALL SPANISH DIRECT|1|0|0|1|3.000|3.000|
+|SNOWBALL SPANISH LUCENE FILTER|1|0|0|1|2.000|2.000|
+|SNOWBALL SWEDISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL SWEDISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL YIDDISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL YIDDISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SPANISH LUCENE SPANISH LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
+|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|1|0|0|0|7.000|7.000|
+|SPANISH LUCENE SPANISH PLURAL STEM FILTER|1|0|0|0|6.000|6.000|
+|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
+|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|1|0|0|0|4.000|4.000|
+|UKRAINIAN LUCENE MORFOLOGIK FILTER|1|0|0|1|2.000|2.000|
+|UKRAINIAN MORFOLOGIK DIRECT|1|0|0|1|3.000|3.000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>LOWERCASE_GROUPS_ONLY placements</summary>
+
+| Stemmer | Evaluated languages | Wins | Exact first-place ties | Top-three placements | Average rank | Median rank |
+|---|---:|---:|---:|---:|---:|---:|
+|Radixor|19|19|0|19|1.000|1.000|
+|CZECH LUCENE CZECH STEM FILTER|1|0|0|1|3.000|3.000|
+|ENGLISH LUCENE KSTEM FILTER|1|0|0|0|8.000|8.000|
+|ENGLISH LUCENE MINIMAL FILTER|1|0|0|0|9.000|9.000|
+|ENGLISH LUCENE PORTER COPIED|1|0|0|1|2.000|2.000|
+|ENGLISH LUCENE PORTER FILTER|1|0|0|1|3.000|3.000|
+|ENGLISH LUCENE POSSESSIVE FILTER|1|0|0|0|11.000|11.000|
+|ENGLISH OPENNLP PORTER|1|0|0|0|4.000|4.000|
+|ENGLISH PAICE HUSK LANCASTER|1|0|0|0|7.000|7.000|
+|ENGLISH SNOWBALL ORIGINAL PORTER|1|0|0|0|6.000|6.000|
+|ENGLISH SNOWBALL PORTER2|1|0|0|0|5.000|5.000|
+|FINNISH LUCENE FINNISH LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|FRENCH LUCENE FRENCH LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
+|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|1|0|0|0|6.000|6.000|
+|GERMAN CISTEM|1|0|0|1|2.000|2.000|
+|GERMAN LUCENE GERMAN LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
+|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|1|0|0|0|8.000|8.000|
+|GERMAN LUCENE GERMAN STEM FILTER|1|0|0|0|6.000|6.000|
+|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|HUNSPELL CZECH LUCENE FILTER|1|0|0|1|2.000|2.000|
+|HUNSPELL DUTCH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|HUNSPELL ENGLISH LUCENE FILTER|1|0|0|0|10.000|10.000|
+|HUNSPELL FRENCH LUCENE FILTER|1|0|0|0|4.000|4.000|
+|HUNSPELL GERMAN LUCENE FILTER|1|0|0|0|7.000|7.000|
+|HUNSPELL POLISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|HUNSPELL SPANISH LUCENE FILTER|1|0|0|0|4.000|4.000|
+|HUNSPELL UKRAINIAN LUCENE FILTER|1|0|0|0|4.000|4.000|
+|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|1|0|0|0|5.000|5.000|
+|PERSIAN LUCENE PERSIAN STEM FILTER|1|0|0|1|2.000|2.000|
+|POLISH LUCENE MORFOLOGIK FILTER|1|0|0|1|2.000|2.000|
+|POLISH LUCENE STEMPEL DIRECT|1|0|0|0|4.000|4.000|
+|POLISH LUCENE STEMPEL FILTER|1|0|0|0|5.000|5.000|
+|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
+|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|1|0|0|0|6.000|6.000|
+|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|1|0|0|0|4.000|4.000|
+|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL DANISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL DANISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL DUTCH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL DUTCH LUCENE FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL FINNISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL FINNISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL FRENCH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL FRENCH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL GERMAN DIRECT|1|0|0|1|3.000|3.000|
+|SNOWBALL GERMAN LUCENE FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL HUNGARIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL HUNGARIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL ITALIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL ITALIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL NORWEGIAN BOKMAL DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL NORWEGIAN NYNORSK DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL PORTUGUESE DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL PORTUGUESE LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL RUSSIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL RUSSIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL SPANISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL SPANISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL SWEDISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL SWEDISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL YIDDISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL YIDDISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SPANISH LUCENE SPANISH LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
+|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|1|0|0|0|7.000|7.000|
+|SPANISH LUCENE SPANISH PLURAL STEM FILTER|1|0|0|0|6.000|6.000|
+|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
+|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|1|0|0|0|4.000|4.000|
+|UKRAINIAN LUCENE MORFOLOGIK FILTER|1|0|0|1|2.000|2.000|
+|UKRAINIAN MORFOLOGIK DIRECT|1|0|0|1|3.000|3.000|
+
+</details>
+
+
+### Radixor full-coverage aggregates
+
+These aggregates cover all 19 documented languages. Macro balanced accuracy gives each language equal weight. Micro metrics first sum raw pair counts across languages. Unsupported third-party languages are never inserted as zero results, so this full-coverage table is not presented as a cross-stemmer common-language ranking.
+
+| Dictionary mode | Languages | Macro balanced accuracy | Micro balanced accuracy | Micro precision | Micro recall | Micro F1 |
+|---|---:|---:|---:|---:|---:|---:|
+|ALL_WORDS|19|0.978929|0.987664|0.975113|0.975328|0.975221|
+|LOWERCASE_GROUPS_ONLY|19|0.982354|0.989366|0.975322|0.978734|0.977025|
+
+### Reproducible data
+
+- [Machine-readable quality snapshot](data/stemming-quality.csv)
+- SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- [Linguistic quality methodology](reference/linguistic-quality.md)
+- [Tested stemmer inventory](reference/tested-stemmers.md)
+- [Reproducibility and raw data](reference/reproducibility.md)
+- Pearson and Spearman correlation files are generated under `build/reports/stemming-quality/`; they are separated by dictionary mode and output policy. Correlation does not establish metric equivalence.
+
+<!-- STEMMING-QUALITY-OVERVIEW:END -->

docs/benchmarks/languages/czech.md

@@ -0,0 +1,418 @@
+# Czech Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Czech. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `CS_CZ` | 5,113 | 56,612 | 10,049 | 46,563 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **56,612**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 675 | 1.192% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 22,681 | 40.064% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 14,980 | 26.461% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 10,109 | 17.857% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 8,167 | 14.426% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 99.465% | 99.439% | 99.582% | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | 84.850% | 82.269% | 96.806% | Benchmark-only Czech Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene CzechStemFilter | 16.784% | 15.538% | 22.559% | Lucene Czech suffix stemmer implemented as a TokenFilter. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `czechRadixor` | 3.332 | 0.240 | 71.6 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 346.819 | 3.622 | 7448.4 | 104.091 | Benchmark-only Czech Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene CzechStemFilter | `czechLuceneCzechStemFilter` | 3.163 | 0.253 | 67.9 | 0.949 | Czech suffix stemmer implemented as a Lucene TokenFilter. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `CS_CZ` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/cs_cz/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.996565** among 3 deterministic stemmers. The runner-up is `HUNSPELL CZECH LUCENE FILTER` at 0.853752, a difference of 0.142813. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.997139** among 3 deterministic stemmers. The runner-up is `HUNSPELL CZECH LUCENE FILTER` at 0.852770, a difference of 0.144369. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **7 result rows**, **3 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996565|3867 / 1334876815 (0.000290%)|2073 / 301835 (0.686799%)|0.988432|0.990189|0.990191|
+|2|HUNSPELL CZECH LUCENE FILTER|PRIMARY_OUTPUT|0.853752|11408 / 1334876815 (0.000855%)|88283 / 301835 (29.248762%)|0.888560|0.810759|0.819499|
+|3|CZECH LUCENE CZECH STEM FILTER|PRIMARY_OUTPUT|0.793614|14480 / 1334876815 (0.001085%)|124586 / 301835 (41.276194%)|0.829234|0.718241|0.736765|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987264|0.993132|0.999997|0.996565|0.999996|0.000004|
+|2|HUNSPELL CZECH LUCENE FILTER|PRIMARY_OUTPUT|0.949289|0.707512|0.999991|0.853752|0.999925|0.000075|
+|3|CZECH LUCENE CZECH STEM FILTER|PRIMARY_OUTPUT|0.924477|0.587238|0.999989|0.793614|0.999896|0.000104|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988432|0.990189|0.991953|0.980569|0.990194|0.990191|
+|2|HUNSPELL CZECH LUCENE FILTER|PRIMARY_OUTPUT|0.888560|0.810759|0.745486|0.681745|0.819533|0.819499|
+|3|CZECH LUCENE CZECH STEM FILTER|PRIMARY_OUTPUT|0.829234|0.718241|0.633453|0.560356|0.736809|0.736765|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.990187|0.998733|0.998686|0.998709|0.998709|
+|2|HUNSPELL CZECH LUCENE FILTER|PRIMARY_OUTPUT|0.810723|0.995777|0.952852|0.973842|0.973842|
+|3|CZECH LUCENE CZECH STEM FILTER|PRIMARY_OUTPUT|0.718192|0.993801|0.944977|0.968774|0.968774|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|299762|3867|2073|1334872948|3867 / 1334876815|2073 / 301835|
+|2|HUNSPELL CZECH LUCENE FILTER|PRIMARY_OUTPUT|213552|11408|88283|1334865407|11408 / 1334876815|88283 / 301835|
+|3|CZECH LUCENE CZECH STEM FILTER|PRIMARY_OUTPUT|177249|14480|124586|1334862335|14480 / 1334876815|124586 / 301835|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 1334876815 (0.000000%)|0 / 301835 (0.000000%)|1.000000|1.000000|1.000000|
+|2|HUNSPELL CZECH LUCENE FILTER|ANY_CANDIDATE|0.871577|10102 / 1334876815 (0.000757%)|77523 / 301835 (25.683900%)|0.904855|0.836596|0.843258|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|HUNSPELL CZECH LUCENE FILTER|ANY_CANDIDATE|0.956905|0.743161|0.999992|0.871577|0.999934|0.000066|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|HUNSPELL CZECH LUCENE FILTER|ANY_CANDIDATE|0.904855|0.836596|0.777914|0.719094|0.843288|0.843258|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL CZECH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|301835|0|0|1334876815|0 / 1334876815|0 / 301835|
+|2|HUNSPELL CZECH LUCENE FILTER|ANY_CANDIDATE|224312|10102|77523|1334866713|10102 / 1334876815|77523 / 301835|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999998|5850 / 1334876815 (0.000438%)|0 / 301835 (0.000000%)|0.984732|0.990402|0.990446|
+|2|HUNSPELL CZECH LUCENE FILTER|ALL_CANDIDATES|0.871575|13917 / 1334876815 (0.001043%)|77523 / 301835 (25.683900%)|0.893851|0.830687|0.836477|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.980987|1.000000|0.999996|0.999998|0.999996|0.000004|
+|2|HUNSPELL CZECH LUCENE FILTER|ALL_CANDIDATES|0.941581|0.743161|0.999990|0.871575|0.999932|0.000068|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.984732|0.990402|0.996139|0.980987|0.990448|0.990446|
+|2|HUNSPELL CZECH LUCENE FILTER|ALL_CANDIDATES|0.893851|0.830687|0.775861|0.710406|0.836509|0.836477|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL CZECH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|301835|5850|0|1334870965|5850 / 1334876815|0 / 301835|
+|2|HUNSPELL CZECH LUCENE FILTER|ALL_CANDIDATES|224312|13917|77523|1334862898|13917 / 1334876815|77523 / 301835|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|2073|3867|1983|596|1.153340%|4|52319|
+|HUNSPELL CZECH LUCENE FILTER|10760|1306|2509|3317|6.418840%|5|55596|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **7 result rows**, **3 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.997139|3863 / 1298544215 (0.000297%)|1709 / 298813 (0.571930%)|0.988580|0.990710|0.990714|
+|2|HUNSPELL CZECH LUCENE FILTER|PRIMARY_OUTPUT|0.852770|11239 / 1298544215 (0.000866%)|87986 / 298813 (29.445171%)|0.888009|0.809505|0.818403|
+|3|CZECH LUCENE CZECH STEM FILTER|PRIMARY_OUTPUT|0.791794|13950 / 1298544215 (0.001074%)|124426 / 298813 (41.640089%)|0.828709|0.715948|0.735055|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987165|0.994281|0.999997|0.997139|0.999996|0.000004|
+|2|HUNSPELL CZECH LUCENE FILTER|PRIMARY_OUTPUT|0.949389|0.705548|0.999991|0.852770|0.999924|0.000076|
+|3|CZECH LUCENE CZECH STEM FILTER|PRIMARY_OUTPUT|0.925931|0.583599|0.999989|0.791794|0.999893|0.000107|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988580|0.990710|0.992849|0.981591|0.990716|0.990714|
+|2|HUNSPELL CZECH LUCENE FILTER|PRIMARY_OUTPUT|0.888009|0.809505|0.743753|0.679973|0.818437|0.818403|
+|3|CZECH LUCENE CZECH STEM FILTER|PRIMARY_OUTPUT|0.828709|0.715948|0.630198|0.557569|0.735100|0.735055|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.990708|0.998726|0.999030|0.998878|0.998878|
+|2|HUNSPELL CZECH LUCENE FILTER|PRIMARY_OUTPUT|0.809467|0.995812|0.952394|0.973619|0.973619|
+|3|CZECH LUCENE CZECH STEM FILTER|PRIMARY_OUTPUT|0.715897|0.993897|0.944297|0.968463|0.968463|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|297104|3863|1709|1298540352|3863 / 1298544215|1709 / 298813|
+|2|HUNSPELL CZECH LUCENE FILTER|PRIMARY_OUTPUT|210827|11239|87986|1298532976|11239 / 1298544215|87986 / 298813|
+|3|CZECH LUCENE CZECH STEM FILTER|PRIMARY_OUTPUT|174387|13950|124426|1298530265|13950 / 1298544215|124426 / 298813|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 1298544215 (0.000000%)|0 / 298813 (0.000000%)|1.000000|1.000000|1.000000|
+|2|HUNSPELL CZECH LUCENE FILTER|ANY_CANDIDATE|0.870432|10028 / 1298544215 (0.000772%)|77431 / 298813 (25.912862%)|0.904004|0.835052|0.841852|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|HUNSPELL CZECH LUCENE FILTER|ANY_CANDIDATE|0.956666|0.740871|0.999992|0.870432|0.999933|0.000067|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|HUNSPELL CZECH LUCENE FILTER|ANY_CANDIDATE|0.904004|0.835052|0.775874|0.716815|0.841883|0.841852|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL CZECH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|298813|0|0|1298544215|0 / 1298544215|0 / 298813|
+|2|HUNSPELL CZECH LUCENE FILTER|ANY_CANDIDATE|221382|10028|77431|1298534187|10028 / 1298544215|77431 / 298813|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999998|5782 / 1298544215 (0.000445%)|0 / 298813 (0.000000%)|0.984756|0.990418|0.990461|
+|2|HUNSPELL CZECH LUCENE FILTER|ALL_CANDIDATES|0.870430|13601 / 1298544215 (0.001047%)|77431 / 298813 (25.912862%)|0.893574|0.829463|0.835425|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.981017|1.000000|0.999996|0.999998|0.999996|0.000004|
+|2|HUNSPELL CZECH LUCENE FILTER|ALL_CANDIDATES|0.942119|0.740871|0.999990|0.870430|0.999930|0.000070|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.984756|0.990418|0.996145|0.981017|0.990463|0.990461|
+|2|HUNSPELL CZECH LUCENE FILTER|ALL_CANDIDATES|0.893574|0.829463|0.773936|0.708617|0.835457|0.835425|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL CZECH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|298813|5782|0|1298538433|5782 / 1298544215|0 / 298813|
+|2|HUNSPELL CZECH LUCENE FILTER|ALL_CANDIDATES|221382|13601|77431|1298530614|13601 / 1298544215|77431 / 298813|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|1709|3863|1919|540|1.059488%|4|51543|
+|HUNSPELL CZECH LUCENE FILTER|10555|1211|2362|3237|6.351044%|5|54804|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `CS_CZ`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/danish.md

@@ -0,0 +1,396 @@
+# Danish Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Danish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `DA_DK` | 4,179 | 32,256 | 8,356 | 23,900 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **32,256**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 137 | 0.425% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 1,127 | 3.494% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 22,586 | 70.021% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 8,405 | 26.057% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 1 | 0.003% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 99.371% | 99.527% | 98.923% | Full Radixor dictionary patch-command stemmer. |
+| Lucene SnowballFilter | 55.509% | 54.159% | 59.371% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Official Snowball direct | 55.509% | 54.159% | 59.371% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `radixor[DANISH]` | 1.143 | 0.017 | 47.8 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Official Snowball direct | `snowballDirect[DANISH]` | 2.168 | 0.058 | 90.7 | 1.896 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[DANISH]` | 2.975 | 0.143 | 124.5 | 2.602 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `DA_DK` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/da_dk/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.996066** among 3 deterministic stemmers. The runner-up is `SNOWBALL DANISH LUCENE FILTER` at 0.937969, a difference of 0.058097. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.996305** among 3 deterministic stemmers. The runner-up is `SNOWBALL DANISH DIRECT` at 0.938074, a difference of 0.058230. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **5 result rows**, **3 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996066|1165 / 394111186 (0.000296%)|707 / 89895 (0.786473%)|0.988108|0.989614|0.989615|
+|2|SNOWBALL DANISH LUCENE FILTER|PRIMARY_OUTPUT|0.937969|6507 / 394111186 (0.001651%)|11151 / 89895 (12.404472%)|0.913718|0.899181|0.899475|
+|3|SNOWBALL DANISH DIRECT|PRIMARY_OUTPUT|0.937903|6341 / 394111186 (0.001609%)|11163 / 89895 (12.417821%)|0.915090|0.899959|0.900279|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987106|0.992135|0.999997|0.996066|0.999995|0.000005|
+|2|SNOWBALL DANISH LUCENE FILTER|PRIMARY_OUTPUT|0.923672|0.875955|0.999983|0.937969|0.999955|0.000045|
+|3|SNOWBALL DANISH DIRECT|PRIMARY_OUTPUT|0.925464|0.875822|0.999984|0.937903|0.999956|0.000044|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988108|0.989614|0.991125|0.979442|0.989618|0.989615|
+|2|SNOWBALL DANISH LUCENE FILTER|PRIMARY_OUTPUT|0.913718|0.899181|0.885100|0.816830|0.899498|0.899475|
+|3|SNOWBALL DANISH DIRECT|PRIMARY_OUTPUT|0.915090|0.899959|0.885320|0.818114|0.900301|0.900279|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.989612|0.998466|0.998719|0.998592|0.998592|
+|2|SNOWBALL DANISH LUCENE FILTER|PRIMARY_OUTPUT|0.899159|0.994053|0.978603|0.986268|0.986268|
+|3|SNOWBALL DANISH DIRECT|PRIMARY_OUTPUT|0.899937|0.994196|0.978579|0.986326|0.986326|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|89188|1165|707|394110021|1165 / 394111186|707 / 89895|
+|2|SNOWBALL DANISH LUCENE FILTER|PRIMARY_OUTPUT|78744|6507|11151|394104679|6507 / 394111186|11151 / 89895|
+|3|SNOWBALL DANISH DIRECT|PRIMARY_OUTPUT|78732|6341|11163|394104845|6341 / 394111186|11163 / 89895|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 394111186 (0.000000%)|0 / 89895 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|89895|0|0|394111186|0 / 394111186|0 / 89895|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999998|1849 / 394111186 (0.000469%)|0 / 89895 (0.000000%)|0.983812|0.989820|0.989869|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.979846|1.000000|0.999995|0.999998|0.999995|0.000005|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.983812|0.989820|0.995903|0.979846|0.989872|0.989869|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|89895|1849|0|394109337|1849 / 394111186|0 / 89895|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|707|1165|684|323|1.150326%|3|28405|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **5 result rows**, **3 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996305|1165 / 392820788 (0.000297%)|663 / 89740 (0.738801%)|0.988190|0.989843|0.989845|
+|2|SNOWBALL DANISH DIRECT|PRIMARY_OUTPUT|0.938074|6341 / 392820788 (0.001614%)|11113 / 89740 (12.383552%)|0.915093|0.900096|0.900410|
+|3|SNOWBALL DANISH LUCENE FILTER|PRIMARY_OUTPUT|0.938074|6341 / 392820788 (0.001614%)|11113 / 89740 (12.383552%)|0.915093|0.900096|0.900410|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987090|0.992612|0.999997|0.996305|0.999995|0.000005|
+|2|SNOWBALL DANISH DIRECT|PRIMARY_OUTPUT|0.925372|0.876164|0.999984|0.938074|0.999956|0.000044|
+|3|SNOWBALL DANISH LUCENE FILTER|PRIMARY_OUTPUT|0.925372|0.876164|0.999984|0.938074|0.999956|0.000044|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988190|0.989843|0.991503|0.979891|0.989847|0.989845|
+|2|SNOWBALL DANISH DIRECT|PRIMARY_OUTPUT|0.915093|0.900096|0.885583|0.818341|0.900432|0.900410|
+|3|SNOWBALL DANISH LUCENE FILTER|PRIMARY_OUTPUT|0.915093|0.900096|0.885583|0.818341|0.900432|0.900410|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.989841|0.998463|0.998812|0.998637|0.998637|
+|2|SNOWBALL DANISH DIRECT|PRIMARY_OUTPUT|0.900074|0.994185|0.978644|0.986354|0.986354|
+|3|SNOWBALL DANISH LUCENE FILTER|PRIMARY_OUTPUT|0.900074|0.994185|0.978644|0.986354|0.986354|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|89077|1165|663|392819623|1165 / 392820788|663 / 89740|
+|2|SNOWBALL DANISH DIRECT|PRIMARY_OUTPUT|78627|6341|11113|392814447|6341 / 392820788|11113 / 89740|
+|3|SNOWBALL DANISH LUCENE FILTER|PRIMARY_OUTPUT|78627|6341|11113|392814447|6341 / 392820788|11113 / 89740|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 392820788 (0.000000%)|0 / 89740 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|89740|0|0|392820788|0 / 392820788|0 / 89740|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999998|1849 / 392820788 (0.000471%)|0 / 89740 (0.000000%)|0.983784|0.989803|0.989852|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.979812|1.000000|0.999995|0.999998|0.999995|0.000005|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.983784|0.989803|0.995896|0.979812|0.989855|0.989852|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|89740|1849|0|392818939|1849 / 392820788|0 / 89740|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|663|1165|684|315|1.123676%|3|28351|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `DA_DK`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/dutch.md

@@ -0,0 +1,430 @@
+# Dutch Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Dutch. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `NL_NL` | 4,992 | 31,466 | 9,981 | 21,485 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **31,466**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 2,107 | 6.696% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 11,484 | 36.497% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 7,732 | 24.573% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 10,127 | 32.184% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 16 | 0.051% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 99.120% | 98.711% | 100.000% | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | 46.590% | 22.718% | 97.976% | Benchmark-only Dutch Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Official Snowball direct | 15.954% | 8.992% | 30.939% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+| Lucene SnowballFilter | 12.620% | 5.441% | 28.073% | Lucene TokenFilter integration path around the Snowball algorithm. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `radixor[DUTCH]` | 1.331 | 0.114 | 61.9 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 22.760 | 1.387 | 1059.3 | 17.105 | Benchmark-only Dutch Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Official Snowball direct | `snowballDirect[DUTCH]` | 4.146 | 0.291 | 193.0 | 3.116 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[DUTCH]` | 7.375 | 0.595 | 343.3 | 5.543 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `NL_NL` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/nl_nl/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.988661** among 4 deterministic stemmers. The runner-up is `SNOWBALL DUTCH DIRECT` at 0.727087, a difference of 0.261574. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.989040** among 4 deterministic stemmers. The runner-up is `SNOWBALL DUTCH DIRECT` at 0.730495, a difference of 0.258544. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **8 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988661|1214 / 350437960 (0.000346%)|1464 / 64566 (2.267447%)|0.980362|0.979221|0.979219|
+|2|SNOWBALL DUTCH DIRECT|PRIMARY_OUTPUT|0.727087|4382 / 350437960 (0.001250%)|35241 / 64566 (54.581359%)|0.735353|0.596807|0.628557|
+|3|HUNSPELL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.643123|1333 / 350437960 (0.000380%)|46084 / 64566 (71.375027%)|0.642512|0.438061|0.516674|
+|4|SNOWBALL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.618497|1588 / 350437960 (0.000453%)|49264 / 64566 (76.300220%)|0.579068|0.375712|0.463333|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.981124|0.977326|0.999997|0.988661|0.999992|0.000008|
+|2|SNOWBALL DUTCH DIRECT|PRIMARY_OUTPUT|0.869997|0.454186|0.999987|0.727087|0.999887|0.000113|
+|3|HUNSPELL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.932728|0.286250|0.999996|0.643123|0.999865|0.000135|
+|4|SNOWBALL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.905980|0.236998|0.999995|0.618497|0.999855|0.000145|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.980362|0.979221|0.978083|0.959289|0.979223|0.979219|
+|2|SNOWBALL DUTCH DIRECT|PRIMARY_OUTPUT|0.735353|0.596807|0.502190|0.425321|0.628602|0.628557|
+|3|HUNSPELL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.642512|0.438061|0.332316|0.280459|0.516714|0.516674|
+|4|SNOWBALL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.579068|0.375712|0.278062|0.231309|0.463374|0.463333|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.979217|0.997464|0.997003|0.997234|0.997234|
+|2|SNOWBALL DUTCH DIRECT|PRIMARY_OUTPUT|0.596756|0.992815|0.917346|0.953590|0.953590|
+|3|HUNSPELL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.438012|0.996932|0.889026|0.939892|0.939892|
+|4|SNOWBALL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.375664|0.995828|0.888410|0.939057|0.939057|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|63102|1214|1464|350436746|1214 / 350437960|1464 / 64566|
+|2|SNOWBALL DUTCH DIRECT|PRIMARY_OUTPUT|29325|4382|35241|350433578|4382 / 350437960|35241 / 64566|
+|3|HUNSPELL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|18482|1333|46084|350436627|1333 / 350437960|46084 / 64566|
+|4|SNOWBALL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|15302|1588|49264|350436372|1588 / 350437960|49264 / 64566|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 350437960 (0.000000%)|0 / 64566 (0.000000%)|1.000000|1.000000|1.000000|
+|2|HUNSPELL DUTCH LUCENE FILTER|ANY_CANDIDATE|0.665519|1164 / 350437960 (0.000332%)|43192 / 64566 (66.895889%)|0.690741|0.490770|0.560268|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|HUNSPELL DUTCH LUCENE FILTER|ANY_CANDIDATE|0.948354|0.331041|0.999997|0.665519|0.999873|0.000127|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|HUNSPELL DUTCH LUCENE FILTER|ANY_CANDIDATE|0.690741|0.490770|0.380588|0.325179|0.560307|0.560268|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL DUTCH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|64566|0|0|350437960|0 / 350437960|0 / 64566|
+|2|HUNSPELL DUTCH LUCENE FILTER|ANY_CANDIDATE|21374|1164|43192|350436796|1164 / 350437960|43192 / 64566|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999996|2651 / 350437960 (0.000756%)|0 / 64566 (0.000000%)|0.968198|0.979884|0.980078|
+|2|HUNSPELL DUTCH LUCENE FILTER|ALL_CANDIDATES|0.665518|1738 / 350437960 (0.000496%)|43192 / 64566 (66.895889%)|0.680640|0.487557|0.553265|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.960561|1.000000|0.999992|0.999996|0.999992|0.000008|
+|2|HUNSPELL DUTCH LUCENE FILTER|ALL_CANDIDATES|0.924801|0.331041|0.999995|0.665518|0.999872|0.000128|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.968198|0.979884|0.991855|0.960561|0.980082|0.980078|
+|2|HUNSPELL DUTCH LUCENE FILTER|ALL_CANDIDATES|0.680640|0.487557|0.379812|0.322364|0.553306|0.553265|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL DUTCH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|64566|2651|0|350435309|2651 / 350437960|0 / 64566|
+|2|HUNSPELL DUTCH LUCENE FILTER|ALL_CANDIDATES|21374|1738|43192|350436222|1738 / 350437960|43192 / 64566|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|HUNSPELL DUTCH LUCENE FILTER|2892|169|405|1254|4.736186%|3|27763|
+|Radixor|1464|1214|1437|572|2.160366%|3|27061|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **8 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.989040|1214 / 329603856 (0.000368%)|1384 / 63147 (2.191711%)|0.980194|0.979401|0.979398|
+|2|SNOWBALL DUTCH DIRECT|PRIMARY_OUTPUT|0.730495|4382 / 329603856 (0.001329%)|34036 / 63147 (53.899631%)|0.738412|0.602463|0.632953|
+|3|HUNSPELL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.645159|1310 / 329603856 (0.000397%)|44814 / 63147 (70.967742%)|0.646808|0.442880|0.520498|
+|4|SNOWBALL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.618546|1544 / 329603856 (0.000468%)|48175 / 63147 (76.290243%)|0.579362|0.375883|0.463566|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.980723|0.978083|0.999996|0.989040|0.999992|0.000008|
+|2|SNOWBALL DUTCH DIRECT|PRIMARY_OUTPUT|0.869167|0.461004|0.999987|0.730495|0.999883|0.000117|
+|3|HUNSPELL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.933310|0.290323|0.999996|0.645159|0.999860|0.000140|
+|4|SNOWBALL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.906515|0.237098|0.999995|0.618546|0.999849|0.000151|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.980194|0.979401|0.978610|0.959634|0.979402|0.979398|
+|2|SNOWBALL DUTCH DIRECT|PRIMARY_OUTPUT|0.738412|0.602463|0.508789|0.431089|0.633000|0.632953|
+|3|HUNSPELL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.646808|0.442880|0.336718|0.284422|0.520539|0.520498|
+|4|SNOWBALL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.579362|0.375883|0.278182|0.231439|0.463608|0.463566|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.979397|0.997373|0.997139|0.997256|0.997256|
+|2|SNOWBALL DUTCH DIRECT|PRIMARY_OUTPUT|0.602410|0.992557|0.918059|0.953856|0.953856|
+|3|HUNSPELL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.442829|0.996884|0.889061|0.939890|0.939890|
+|4|SNOWBALL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|0.375834|0.995817|0.887492|0.938539|0.938539|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|61763|1214|1384|329602642|1214 / 329603856|1384 / 63147|
+|2|SNOWBALL DUTCH DIRECT|PRIMARY_OUTPUT|29111|4382|34036|329599474|4382 / 329603856|34036 / 63147|
+|3|HUNSPELL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|18333|1310|44814|329602546|1310 / 329603856|44814 / 63147|
+|4|SNOWBALL DUTCH LUCENE FILTER|PRIMARY_OUTPUT|14972|1544|48175|329602312|1544 / 329603856|48175 / 63147|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 329603856 (0.000000%)|0 / 63147 (0.000000%)|1.000000|1.000000|1.000000|
+|2|HUNSPELL DUTCH LUCENE FILTER|ANY_CANDIDATE|0.667956|1141 / 329603856 (0.000346%)|41935 / 63147 (66.408539%)|0.695206|0.496187|0.564555|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|HUNSPELL DUTCH LUCENE FILTER|ANY_CANDIDATE|0.948955|0.335915|0.999997|0.667956|0.999869|0.000131|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|HUNSPELL DUTCH LUCENE FILTER|ANY_CANDIDATE|0.695206|0.496187|0.385755|0.329953|0.564595|0.564555|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL DUTCH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|63147|0|0|329603856|0 / 329603856|0 / 63147|
+|2|HUNSPELL DUTCH LUCENE FILTER|ANY_CANDIDATE|21212|1141|41935|329602715|1141 / 329603856|41935 / 63147|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999996|2651 / 329603856 (0.000804%)|0 / 63147 (0.000000%)|0.967506|0.979441|0.979644|
+|2|HUNSPELL DUTCH LUCENE FILTER|ALL_CANDIDATES|0.667955|1712 / 329603856 (0.000519%)|41935 / 63147 (66.408539%)|0.684952|0.492895|0.557477|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.959710|1.000000|0.999992|0.999996|0.999992|0.000008|
+|2|HUNSPELL DUTCH LUCENE FILTER|ALL_CANDIDATES|0.925318|0.335915|0.999995|0.667955|0.999868|0.000132|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.967506|0.979441|0.991674|0.959710|0.979648|0.979644|
+|2|HUNSPELL DUTCH LUCENE FILTER|ALL_CANDIDATES|0.684952|0.492895|0.384956|0.327048|0.557519|0.557477|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL DUTCH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|63147|2651|0|329601205|2651 / 329603856|0 / 63147|
+|2|HUNSPELL DUTCH LUCENE FILTER|ALL_CANDIDATES|21212|1712|41935|329602144|1712 / 329603856|41935 / 63147|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|HUNSPELL DUTCH LUCENE FILTER|2879|169|402|1186|4.618740%|3|26896|
+|Radixor|1384|1214|1437|549|2.138017%|3|26239|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `NL_NL`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/english.md

@@ -0,0 +1,514 @@
+# English Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for English. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `US_UK` | 396,939 | 1,004,374 | 793,874 | 210,500 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **1,004,374**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 28 | 0.003% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 22,493 | 2.240% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 186,764 | 18.595% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 795,024 | 79.156% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 65 | 0.006% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 97.478% | 97.197% | 97.552% | Full Radixor dictionary patch-command stemmer. |
+| Lucene EnglishMinimalStemFilter | 90.981% | 65.189% | 97.820% | Minimal English plural reduction, not a full stemmer. |
+| Lucene KStemFilter | 80.076% | 76.608% | 80.996% | Krovetz-style English stemming TokenFilter; broader than minimal suffix reducers. |
+| Lucene HunspellStemFilter | 80.243% | 12.750% | 98.139% | Benchmark-only English Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene EnglishPossessiveFilter | 79.032% | 0.003% | 99.987% | Possessive-ending remover only, not a full stemmer. |
+| Snowball English / Porter2 | 40.342% | 46.296% | 38.763% | Porter2 rule-based suffix stemmer, distinct from original Porter. |
+| Lucene PorterStemFilter | 39.538% | 46.201% | 37.772% | Lucene TokenFilter path for Porter suffix rules; not dictionary-root equivalent. |
+| Lucene PorterStemmer direct copy | 39.538% | 46.201% | 37.772% | Direct Porter suffix-rule implementation generated under build for benchmark-only use. |
+| OpenNLP PorterStemmer | 39.538% | 46.201% | 37.772% | Apache OpenNLP Porter suffix-rule implementation. |
+| Snowball original Porter | 39.529% | 46.179% | 37.766% | Classic Porter rule-based suffix stemmer. |
+| Paice/Husk Lancaster | 28.055% | 37.039% | 25.673% | Aggressive Paice/Husk rule stemmer that often produces shorter stems. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `radixorUsUkProfiPreferredStem` | 21.987 | 8.707 | 104.5 | 1.000 | Full dictionary patch-command stemmer using compiled patch commands. |
+| Lucene EnglishPossessiveFilter | `luceneEnglishPossessiveFilter` | 24.539 | 1.515 | 116.6 | 1.116 | Possessive-ending remover only; not a full stemmer. |
+| Lucene EnglishMinimalStemFilter | `luceneEnglishMinimalStemFilter` | 22.702 | 1.195 | 107.8 | 1.032 | Narrow plural reduction filter; not a full stemmer. |
+| Lucene PorterStemmer direct copy | `lucenePorterStemmerCopied` | 24.696 | 13.235 | 117.3 | 1.123 | Benchmark-only generated copy of Lucene package-private Porter implementation. |
+| OpenNLP PorterStemmer | `opennlpPorterStemmer` | 23.121 | 12.528 | 109.8 | 1.052 | Apache OpenNLP Porter implementation. |
+| Snowball original Porter | `snowballOriginalPorter` | 38.904 | 10.353 | 184.8 | 1.769 | Classic Porter suffix-rule stemmer; historical English baseline, not a dictionary-equivalent stemmer. |
+| Lucene PorterStemFilter | `lucenePorterStemFilter` | 37.021 | 1.196 | 175.9 | 1.684 | Lucene TokenFilter integration path for Porter; includes TokenStream overhead. |
+| Lucene KStemFilter | `luceneKStemFilter` | 51.640 | 2.591 | 245.3 | 2.349 | Krovetz-style English TokenFilter; broader than minimal suffix filters. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 79.785 | 1.347 | 379.0 | 3.629 | Benchmark-only English Hunspell comparison using the benchmark Hunspell corpus. |
+| Snowball English / Porter2 | `snowballEnglishPorter2` | 52.437 | 0.773 | 249.1 | 2.385 | Porter2 suffix-rule stemmer, distinct from original Porter. |
+| Paice/Husk Lancaster | `paiceHuskLancaster` | 141.556 | 12.324 | 672.5 | 6.438 | Aggressive rule-based English stemmer. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `US_UK` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/us_uk/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.965159** among 11 deterministic stemmers. The runner-up is `ENGLISH LUCENE PORTER COPIED` at 0.954627, a difference of 0.010533. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.965820** among 11 deterministic stemmers. The runner-up is `ENGLISH LUCENE PORTER COPIED` at 0.954900, a difference of 0.010920. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **15 result rows**, **11 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.965159|1149886 / 184490451771 (0.000623%)|21869 / 313870 (6.967534%)|0.240076|0.332621|0.434052|
+|2|ENGLISH LUCENE PORTER COPIED|PRIMARY_OUTPUT|0.954627|1557406 / 184490451771 (0.000844%)|28480 / 313870 (9.073820%)|0.185679|0.264659|0.375252|
+|3|ENGLISH LUCENE PORTER FILTER|PRIMARY_OUTPUT|0.954627|1557406 / 184490451771 (0.000844%)|28480 / 313870 (9.073820%)|0.185679|0.264659|0.375252|
+|4|ENGLISH OPENNLP PORTER|PRIMARY_OUTPUT|0.954627|1557406 / 184490451771 (0.000844%)|28480 / 313870 (9.073820%)|0.185679|0.264659|0.375252|
+|5|ENGLISH SNOWBALL PORTER2|PRIMARY_OUTPUT|0.954537|1566711 / 184490451771 (0.000849%)|28536 / 313870 (9.091662%)|0.184753|0.263477|0.374240|
+|6|ENGLISH SNOWBALL ORIGINAL PORTER|PRIMARY_OUTPUT|0.954490|1555293 / 184490451771 (0.000843%)|28566 / 313870 (9.101220%)|0.185835|0.264849|0.375363|
+|7|ENGLISH PAICE HUSK LANCASTER|PRIMARY_OUTPUT|0.952394|3062661 / 184490451771 (0.001660%)|29879 / 313870 (9.519546%)|0.103643|0.155164|0.277089|
+|8|ENGLISH LUCENE KSTEM FILTER|PRIMARY_OUTPUT|0.878441|1368501 / 184490451771 (0.000742%)|76305 / 313870 (24.311020%)|0.176284|0.247472|0.334598|
+|9|ENGLISH LUCENE MINIMAL FILTER|PRIMARY_OUTPUT|0.718599|1122264 / 184490451771 (0.000608%)|176645 / 313870 (56.279670%)|0.128204|0.174436|0.218251|
+|10|HUNSPELL ENGLISH LUCENE FILTER|PRIMARY_OUTPUT|0.573277|1981986 / 184490451771 (0.001074%)|267868 / 313870 (85.343614%)|0.027298|0.039287|0.057655|
+|11|ENGLISH LUCENE POSSESSIVE FILTER|PRIMARY_OUTPUT|0.500008|1115154 / 184490451771 (0.000604%)|313863 / 313870 (99.997770%)|0.000007|0.000010|0.000009|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.202513|0.930325|0.999994|0.965159|0.999994|0.000006|
+|2|ENGLISH LUCENE PORTER COPIED|PRIMARY_OUTPUT|0.154868|0.909262|0.999992|0.954627|0.999991|0.000009|
+|3|ENGLISH LUCENE PORTER FILTER|PRIMARY_OUTPUT|0.154868|0.909262|0.999992|0.954627|0.999991|0.000009|
+|4|ENGLISH OPENNLP PORTER|PRIMARY_OUTPUT|0.154868|0.909262|0.999992|0.954627|0.999991|0.000009|
+|5|ENGLISH SNOWBALL PORTER2|PRIMARY_OUTPUT|0.154064|0.909083|0.999992|0.954537|0.999991|0.000009|
+|6|ENGLISH SNOWBALL ORIGINAL PORTER|PRIMARY_OUTPUT|0.155006|0.908988|0.999992|0.954490|0.999991|0.000009|
+|7|ENGLISH PAICE HUSK LANCASTER|PRIMARY_OUTPUT|0.084858|0.904805|0.999983|0.952394|0.999983|0.000017|
+|8|ENGLISH LUCENE KSTEM FILTER|PRIMARY_OUTPUT|0.147917|0.756890|0.999993|0.878441|0.999992|0.000008|
+|9|ENGLISH LUCENE MINIMAL FILTER|PRIMARY_OUTPUT|0.108953|0.437203|0.999994|0.718599|0.999993|0.000007|
+|10|HUNSPELL ENGLISH LUCENE FILTER|PRIMARY_OUTPUT|0.022684|0.146564|0.999989|0.573277|0.999988|0.000012|
+|11|ENGLISH LUCENE POSSESSIVE FILTER|PRIMARY_OUTPUT|0.000006|0.000022|0.999994|0.500008|0.999992|0.000008|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.240076|0.332621|0.541270|0.199487|0.434054|0.434052|
+|2|ENGLISH LUCENE PORTER COPIED|PRIMARY_OUTPUT|0.185679|0.264659|0.460563|0.152511|0.375254|0.375252|
+|3|ENGLISH LUCENE PORTER FILTER|PRIMARY_OUTPUT|0.185679|0.264659|0.460563|0.152511|0.375254|0.375252|
+|4|ENGLISH OPENNLP PORTER|PRIMARY_OUTPUT|0.185679|0.264659|0.460563|0.152511|0.375254|0.375252|
+|5|ENGLISH SNOWBALL PORTER2|PRIMARY_OUTPUT|0.184753|0.263477|0.459102|0.151727|0.374242|0.374240|
+|6|ENGLISH SNOWBALL ORIGINAL PORTER|PRIMARY_OUTPUT|0.185835|0.264849|0.460751|0.152637|0.375365|0.375363|
+|7|ENGLISH PAICE HUSK LANCASTER|PRIMARY_OUTPUT|0.103643|0.155164|0.308543|0.084107|0.277092|0.277089|
+|8|ENGLISH LUCENE KSTEM FILTER|PRIMARY_OUTPUT|0.176284|0.247472|0.415099|0.141208|0.334600|0.334598|
+|9|ENGLISH LUCENE MINIMAL FILTER|PRIMARY_OUTPUT|0.128204|0.174436|0.272816|0.095552|0.218253|0.218251|
+|10|HUNSPELL ENGLISH LUCENE FILTER|PRIMARY_OUTPUT|0.027298|0.039287|0.070051|0.020037|0.057659|0.057655|
+|11|ENGLISH LUCENE POSSESSIVE FILTER|PRIMARY_OUTPUT|0.000007|0.000010|0.000015|0.000005|0.000012|0.000009|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.332619|0.994215|0.997770|0.995989|0.995989|
+|2|ENGLISH LUCENE PORTER COPIED|PRIMARY_OUTPUT|0.264656|0.969648|0.997199|0.983231|0.983231|
+|3|ENGLISH LUCENE PORTER FILTER|PRIMARY_OUTPUT|0.264656|0.969648|0.997199|0.983231|0.983231|
+|4|ENGLISH OPENNLP PORTER|PRIMARY_OUTPUT|0.264656|0.969648|0.997199|0.983231|0.983231|
+|5|ENGLISH SNOWBALL PORTER2|PRIMARY_OUTPUT|0.263474|0.969037|0.997182|0.982908|0.982908|
+|6|ENGLISH SNOWBALL ORIGINAL PORTER|PRIMARY_OUTPUT|0.264847|0.969891|0.997193|0.983353|0.983353|
+|7|ENGLISH PAICE HUSK LANCASTER|PRIMARY_OUTPUT|0.155162|0.937768|0.996600|0.966289|0.966289|
+|8|ENGLISH LUCENE KSTEM FILTER|PRIMARY_OUTPUT|0.247470|0.980687|0.992108|0.986364|0.986364|
+|9|ENGLISH LUCENE MINIMAL FILTER|PRIMARY_OUTPUT|0.174433|0.995202|0.981174|0.988138|0.988138|
+|10|HUNSPELL ENGLISH LUCENE FILTER|PRIMARY_OUTPUT|0.039284|0.993096|0.963677|0.978166|0.978166|
+|11|ENGLISH LUCENE POSSESSIVE FILTER|PRIMARY_OUTPUT|0.000007|0.995789|0.958019|0.976539|0.976539|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|292001|1149886|21869|184489301885|1149886 / 184490451771|21869 / 313870|
+|2|ENGLISH LUCENE PORTER COPIED|PRIMARY_OUTPUT|285390|1557406|28480|184488894365|1557406 / 184490451771|28480 / 313870|
+|3|ENGLISH LUCENE PORTER FILTER|PRIMARY_OUTPUT|285390|1557406|28480|184488894365|1557406 / 184490451771|28480 / 313870|
+|4|ENGLISH OPENNLP PORTER|PRIMARY_OUTPUT|285390|1557406|28480|184488894365|1557406 / 184490451771|28480 / 313870|
+|5|ENGLISH SNOWBALL PORTER2|PRIMARY_OUTPUT|285334|1566711|28536|184488885060|1566711 / 184490451771|28536 / 313870|
+|6|ENGLISH SNOWBALL ORIGINAL PORTER|PRIMARY_OUTPUT|285304|1555293|28566|184488896478|1555293 / 184490451771|28566 / 313870|
+|7|ENGLISH PAICE HUSK LANCASTER|PRIMARY_OUTPUT|283991|3062661|29879|184487389110|3062661 / 184490451771|29879 / 313870|
+|8|ENGLISH LUCENE KSTEM FILTER|PRIMARY_OUTPUT|237565|1368501|76305|184489083270|1368501 / 184490451771|76305 / 313870|
+|9|ENGLISH LUCENE MINIMAL FILTER|PRIMARY_OUTPUT|137225|1122264|176645|184489329507|1122264 / 184490451771|176645 / 313870|
+|10|HUNSPELL ENGLISH LUCENE FILTER|PRIMARY_OUTPUT|46002|1981986|267868|184488469785|1981986 / 184490451771|267868 / 313870|
+|11|ENGLISH LUCENE POSSESSIVE FILTER|PRIMARY_OUTPUT|7|1115154|313863|184489336617|1115154 / 184490451771|313863 / 313870|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.999976|12 / 184490451771 (0.000000%)|15 / 313870 (0.004779%)|0.999960|0.999957|0.999957|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ANY_CANDIDATE|0.581603|1978852 / 184490451771 (0.001073%)|262641 / 313870 (83.678274%)|0.030370|0.043712|0.064174|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.999962|0.999952|1.000000|0.999976|1.000000|0.000000|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ANY_CANDIDATE|0.025235|0.163217|0.999989|0.581603|0.999988|0.000012|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.999960|0.999957|0.999954|0.999914|0.999957|0.999957|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ANY_CANDIDATE|0.030370|0.043712|0.077961|0.022344|0.064178|0.064174|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|313855|12|15|184490451759|12 / 184490451771|15 / 313870|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ANY_CANDIDATE|51229|1978852|262641|184488472919|1978852 / 184490451771|262641 / 313870|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999945|11482166 / 184490451771 (0.006224%)|15 / 313870 (0.004779%)|0.033039|0.051834|0.163107|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ALL_CANDIDATES|0.581603|2008917 / 184490451771 (0.001089%)|262641 / 313870 (83.678274%)|0.029943|0.043158|0.063704|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.026607|0.999952|0.999938|0.999945|0.999938|0.000062|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ALL_CANDIDATES|0.024867|0.163217|0.999989|0.581603|0.999988|0.000012|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.033039|0.051834|0.120237|0.026607|0.163112|0.163107|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ALL_CANDIDATES|0.029943|0.043158|0.077254|0.022055|0.063708|0.063704|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|313855|11482166|15|184478969605|11482166 / 184490451771|15 / 313870|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ALL_CANDIDATES|51229|2008917|262641|184488442854|2008917 / 184490451771|262641 / 313870|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|21854|1149874|10332280|29208|4.808384%|1355|2838145|
+|HUNSPELL ENGLISH LUCENE FILTER|5227|3134|26931|6837|1.125545%|4|614296|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **15 result rows**, **11 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.965820|1148489 / 170474840204 (0.000674%)|21319 / 311891 (6.835401%)|0.239424|0.331902|0.433722|
+|2|ENGLISH LUCENE PORTER COPIED|PRIMARY_OUTPUT|0.954900|1552702 / 170474840204 (0.000911%)|28130 / 311891 (9.019177%)|0.185277|0.264166|0.374937|
+|3|ENGLISH LUCENE PORTER FILTER|PRIMARY_OUTPUT|0.954900|1552702 / 170474840204 (0.000911%)|28130 / 311891 (9.019177%)|0.185277|0.264166|0.374937|
+|4|ENGLISH OPENNLP PORTER|PRIMARY_OUTPUT|0.954900|1552702 / 170474840204 (0.000911%)|28130 / 311891 (9.019177%)|0.185277|0.264166|0.374937|
+|5|ENGLISH SNOWBALL PORTER2|PRIMARY_OUTPUT|0.954850|1561891 / 170474840204 (0.000916%)|28161 / 311891 (9.029116%)|0.184375|0.263016|0.373964|
+|6|ENGLISH SNOWBALL ORIGINAL PORTER|PRIMARY_OUTPUT|0.954762|1550615 / 170474840204 (0.000910%)|28216 / 311891 (9.046750%)|0.185431|0.264353|0.375045|
+|7|ENGLISH PAICE HUSK LANCASTER|PRIMARY_OUTPUT|0.952710|3045870 / 170474840204 (0.001787%)|29493 / 311891 (9.456188%)|0.103633|0.155157|0.277170|
+|8|ENGLISH LUCENE KSTEM FILTER|PRIMARY_OUTPUT|0.880820|1367069 / 170474840204 (0.000802%)|74340 / 311891 (23.835250%)|0.176477|0.247899|0.335789|
+|9|ENGLISH LUCENE MINIMAL FILTER|PRIMARY_OUTPUT|0.719516|1120871 / 170474840204 (0.000657%)|174959 / 311891 (56.096200%)|0.128139|0.174470|0.218621|
+|10|HUNSPELL ENGLISH LUCENE FILTER|PRIMARY_OUTPUT|0.573619|1978041 / 170474840204 (0.001160%)|265965 / 311891 (85.274984%)|0.027312|0.039323|0.057799|
+|11|ENGLISH LUCENE POSSESSIVE FILTER|PRIMARY_OUTPUT|0.500005|1113773 / 170474840204 (0.000653%)|311886 / 311891 (99.998397%)|0.000005|0.000007|0.000005|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.201918|0.931646|0.999993|0.965820|0.999993|0.000007|
+|2|ENGLISH LUCENE PORTER COPIED|PRIMARY_OUTPUT|0.154515|0.909808|0.999991|0.954900|0.999991|0.000009|
+|3|ENGLISH LUCENE PORTER FILTER|PRIMARY_OUTPUT|0.154515|0.909808|0.999991|0.954900|0.999991|0.000009|
+|4|ENGLISH OPENNLP PORTER|PRIMARY_OUTPUT|0.154515|0.909808|0.999991|0.954900|0.999991|0.000009|
+|5|ENGLISH SNOWBALL PORTER2|PRIMARY_OUTPUT|0.153731|0.909709|0.999991|0.954850|0.999991|0.000009|
+|6|ENGLISH SNOWBALL ORIGINAL PORTER|PRIMARY_OUTPUT|0.154651|0.909532|0.999991|0.954762|0.999991|0.000009|
+|7|ENGLISH PAICE HUSK LANCASTER|PRIMARY_OUTPUT|0.084848|0.905438|0.999982|0.952710|0.999982|0.000018|
+|8|ENGLISH LUCENE KSTEM FILTER|PRIMARY_OUTPUT|0.148042|0.761647|0.999992|0.880820|0.999992|0.000008|
+|9|ENGLISH LUCENE MINIMAL FILTER|PRIMARY_OUTPUT|0.108866|0.439038|0.999993|0.719516|0.999992|0.000008|
+|10|HUNSPELL ENGLISH LUCENE FILTER|PRIMARY_OUTPUT|0.022691|0.147250|0.999988|0.573619|0.999987|0.000013|
+|11|ENGLISH LUCENE POSSESSIVE FILTER|PRIMARY_OUTPUT|0.000004|0.000016|0.999993|0.500005|0.999992|0.000008|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.239424|0.331902|0.540775|0.198970|0.433723|0.433722|
+|2|ENGLISH LUCENE PORTER COPIED|PRIMARY_OUTPUT|0.185277|0.264166|0.460049|0.152184|0.374939|0.374937|
+|3|ENGLISH LUCENE PORTER FILTER|PRIMARY_OUTPUT|0.185277|0.264166|0.460049|0.152184|0.374939|0.374937|
+|4|ENGLISH OPENNLP PORTER|PRIMARY_OUTPUT|0.185277|0.264166|0.460049|0.152184|0.374939|0.374937|
+|5|ENGLISH SNOWBALL PORTER2|PRIMARY_OUTPUT|0.184375|0.263016|0.458637|0.151421|0.373966|0.373964|
+|6|ENGLISH SNOWBALL ORIGINAL PORTER|PRIMARY_OUTPUT|0.185431|0.264353|0.460234|0.152308|0.375047|0.375045|
+|7|ENGLISH PAICE HUSK LANCASTER|PRIMARY_OUTPUT|0.103633|0.155157|0.308576|0.084103|0.277173|0.277170|
+|8|ENGLISH LUCENE KSTEM FILTER|PRIMARY_OUTPUT|0.176477|0.247899|0.416437|0.141487|0.335791|0.335789|
+|9|ENGLISH LUCENE MINIMAL FILTER|PRIMARY_OUTPUT|0.128139|0.174470|0.273277|0.095572|0.218624|0.218621|
+|10|HUNSPELL ENGLISH LUCENE FILTER|PRIMARY_OUTPUT|0.027312|0.039323|0.070190|0.020056|0.057804|0.057799|
+|11|ENGLISH LUCENE POSSESSIVE FILTER|PRIMARY_OUTPUT|0.000005|0.000007|0.000011|0.000004|0.000008|0.000005|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.331900|0.993959|0.997731|0.995842|0.995842|
+|2|ENGLISH LUCENE PORTER COPIED|PRIMARY_OUTPUT|0.264164|0.968645|0.997109|0.982671|0.982671|
+|3|ENGLISH LUCENE PORTER FILTER|PRIMARY_OUTPUT|0.264164|0.968645|0.997109|0.982671|0.982671|
+|4|ENGLISH OPENNLP PORTER|PRIMARY_OUTPUT|0.264164|0.968645|0.997109|0.982671|0.982671|
+|5|ENGLISH SNOWBALL PORTER2|PRIMARY_OUTPUT|0.263014|0.968020|0.997096|0.982343|0.982343|
+|6|ENGLISH SNOWBALL ORIGINAL PORTER|PRIMARY_OUTPUT|0.264351|0.968894|0.997102|0.982795|0.982795|
+|7|ENGLISH PAICE HUSK LANCASTER|PRIMARY_OUTPUT|0.155154|0.936077|0.996487|0.965338|0.965338|
+|8|ENGLISH LUCENE KSTEM FILTER|PRIMARY_OUTPUT|0.247897|0.979822|0.991991|0.985869|0.985869|
+|9|ENGLISH LUCENE MINIMAL FILTER|PRIMARY_OUTPUT|0.174467|0.994994|0.980520|0.987704|0.987704|
+|10|HUNSPELL ENGLISH LUCENE FILTER|PRIMARY_OUTPUT|0.039320|0.993066|0.962317|0.977450|0.977450|
+|11|ENGLISH LUCENE POSSESSIVE FILTER|PRIMARY_OUTPUT|0.000004|0.995605|0.956423|0.975621|0.975621|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|290572|1148489|21319|170473691715|1148489 / 170474840204|21319 / 311891|
+|2|ENGLISH LUCENE PORTER COPIED|PRIMARY_OUTPUT|283761|1552702|28130|170473287502|1552702 / 170474840204|28130 / 311891|
+|3|ENGLISH LUCENE PORTER FILTER|PRIMARY_OUTPUT|283761|1552702|28130|170473287502|1552702 / 170474840204|28130 / 311891|
+|4|ENGLISH OPENNLP PORTER|PRIMARY_OUTPUT|283761|1552702|28130|170473287502|1552702 / 170474840204|28130 / 311891|
+|5|ENGLISH SNOWBALL PORTER2|PRIMARY_OUTPUT|283730|1561891|28161|170473278313|1561891 / 170474840204|28161 / 311891|
+|6|ENGLISH SNOWBALL ORIGINAL PORTER|PRIMARY_OUTPUT|283675|1550615|28216|170473289589|1550615 / 170474840204|28216 / 311891|
+|7|ENGLISH PAICE HUSK LANCASTER|PRIMARY_OUTPUT|282398|3045870|29493|170471794334|3045870 / 170474840204|29493 / 311891|
+|8|ENGLISH LUCENE KSTEM FILTER|PRIMARY_OUTPUT|237551|1367069|74340|170473473135|1367069 / 170474840204|74340 / 311891|
+|9|ENGLISH LUCENE MINIMAL FILTER|PRIMARY_OUTPUT|136932|1120871|174959|170473719333|1120871 / 170474840204|174959 / 311891|
+|10|HUNSPELL ENGLISH LUCENE FILTER|PRIMARY_OUTPUT|45926|1978041|265965|170472862163|1978041 / 170474840204|265965 / 311891|
+|11|ENGLISH LUCENE POSSESSIVE FILTER|PRIMARY_OUTPUT|5|1113773|311886|170473726431|1113773 / 170474840204|311886 / 311891|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 170474840204 (0.000000%)|0 / 311891 (0.000000%)|1.000000|1.000000|1.000000|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ANY_CANDIDATE|0.581994|1974950 / 170474840204 (0.001158%)|260741 / 311891 (83.600040%)|0.030387|0.043756|0.064341|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ANY_CANDIDATE|0.025246|0.164000|0.999988|0.581994|0.999987|0.000013|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ANY_CANDIDATE|0.030387|0.043756|0.078123|0.022367|0.064345|0.064341|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|311891|0|0|170474840204|0 / 170474840204|0 / 311891|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ANY_CANDIDATE|51150|1974950|260741|170472865254|1974950 / 170474840204|260741 / 311891|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999966|11470018 / 170474840204 (0.006728%)|0 / 311891 (0.000000%)|0.032872|0.051579|0.162697|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ALL_CANDIDATES|0.581994|2004598 / 170474840204 (0.001176%)|260741 / 311891 (83.600040%)|0.029965|0.043208|0.063875|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.026472|1.000000|0.999933|0.999966|0.999933|0.000067|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ALL_CANDIDATES|0.024881|0.164000|0.999988|0.581994|0.999987|0.000013|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.032872|0.051579|0.119687|0.026472|0.162702|0.162697|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ALL_CANDIDATES|0.029965|0.043208|0.077422|0.022081|0.063879|0.063875|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|311891|11470018|0|170463370186|11470018 / 170474840204|0 / 311891|
+|2|HUNSPELL ENGLISH LUCENE FILTER|ALL_CANDIDATES|51150|2004598|260741|170472835606|2004598 / 170474840204|260741 / 311891|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|21319|1148489|10321529|28826|4.936720%|1355|2812871|
+|HUNSPELL ENGLISH LUCENE FILTER|5224|3091|26557|6786|1.162165%|4|590716|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `US_UK`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/finnish.md

@@ -0,0 +1,408 @@
+# Finnish Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Finnish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `FI_FI` | 57,027 | 1,865,215 | 110,525 | 1,754,690 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **1,865,215**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 745 | 0.040% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 1,176,003 | 63.049% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 565,585 | 30.323% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 116,946 | 6.270% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 5,936 | 0.318% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 98.661% | 98.803% | 96.408% | Full Radixor dictionary patch-command stemmer. |
+| Lucene SnowballFilter | 10.991% | 10.268% | 22.471% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Official Snowball direct | 10.991% | 10.268% | 22.471% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+| Lucene FinnishLightStemFilter | 4.351% | 4.294% | 5.264% | Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `finnishRadixor` | 308.076 | 15.529 | 175.6 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene FinnishLightStemFilter | `finnishLuceneFinnishLightStemFilter` | 175.250 | 46.995 | 99.9 | 0.569 | Light Finnish suffix stemmer. |
+| Official Snowball direct | `snowballDirect[FINNISH]` | 264.652 | 63.054 | 150.8 | 0.859 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[FINNISH]` | 374.883 | 238.157 | 213.6 | 1.217 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `FI_FI` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/fi_fi/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.984594** among 4 deterministic stemmers. The runner-up is `SNOWBALL FINNISH LUCENE FILTER` at 0.740353, a difference of 0.244242. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.988068** among 4 deterministic stemmers. The runner-up is `SNOWBALL FINNISH DIRECT` at 0.738400, a difference of 0.249668. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **6 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.984594|731279 / 1641126814491 (0.000045%)|971268 / 31523695 (3.081073%)|0.975128|0.972893|0.972899|
+|2|SNOWBALL FINNISH LUCENE FILTER|PRIMARY_OUTPUT|0.740353|1922153 / 1641126814491 (0.000117%)|16370057 / 31523695 (51.929372%)|0.758996|0.623613|0.653138|
+|3|SNOWBALL FINNISH DIRECT|PRIMARY_OUTPUT|0.739729|1544812 / 1641126814491 (0.000094%)|16409363 / 31523695 (52.054060%)|0.769880|0.627374|0.659540|
+|4|FINNISH LUCENE FINNISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.695969|2223150 / 1641126814491 (0.000135%)|19168306 / 31523695 (60.806025%)|0.687649|0.536000|0.576338|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.976624|0.969189|1.000000|0.984594|0.999999|0.000001|
+|2|SNOWBALL FINNISH LUCENE FILTER|PRIMARY_OUTPUT|0.887434|0.480706|0.999999|0.740353|0.999989|0.000011|
+|3|SNOWBALL FINNISH DIRECT|PRIMARY_OUTPUT|0.907269|0.479459|0.999999|0.739729|0.999989|0.000011|
+|4|FINNISH LUCENE FINNISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.847505|0.391940|0.999999|0.695969|0.999987|0.000013|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.975128|0.972893|0.970667|0.947216|0.972900|0.972899|
+|2|SNOWBALL FINNISH LUCENE FILTER|PRIMARY_OUTPUT|0.758996|0.623613|0.529216|0.453080|0.653142|0.653138|
+|3|SNOWBALL FINNISH DIRECT|PRIMARY_OUTPUT|0.769880|0.627374|0.529384|0.457061|0.659544|0.659540|
+|4|FINNISH LUCENE FINNISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.687649|0.536000|0.439152|0.366120|0.576343|0.576338|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.972892|0.996085|0.993746|0.994914|0.994914|
+|2|SNOWBALL FINNISH LUCENE FILTER|PRIMARY_OUTPUT|0.623608|0.990718|0.904385|0.945585|0.945585|
+|3|SNOWBALL FINNISH DIRECT|PRIMARY_OUTPUT|0.627369|0.991872|0.904139|0.945975|0.945975|
+|4|FINNISH LUCENE FINNISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.535994|0.988126|0.886473|0.934544|0.934544|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|30552427|731279|971268|1641126083212|731279 / 1641126814491|971268 / 31523695|
+|2|SNOWBALL FINNISH LUCENE FILTER|PRIMARY_OUTPUT|15153638|1922153|16370057|1641124892338|1922153 / 1641126814491|16370057 / 31523695|
+|3|SNOWBALL FINNISH DIRECT|PRIMARY_OUTPUT|15114332|1544812|16409363|1641125269679|1544812 / 1641126814491|16409363 / 31523695|
+|4|FINNISH LUCENE FINNISH LIGHT STEM FILTER|PRIMARY_OUTPUT|12355389|2223150|19168306|1641124591341|2223150 / 1641126814491|19168306 / 31523695|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 1641126814491 (0.000000%)|0 / 31523695 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|31523695|0|0|1641126814491|0 / 1641126814491|0 / 31523695|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999999|1683575 / 1641126814491 (0.000103%)|0 / 31523695 (0.000000%)|0.959025|0.973991|0.974320|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.949301|1.000000|0.999999|0.999999|0.999999|0.000001|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.959025|0.973991|0.989432|0.949301|0.974321|0.974320|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|31523695|1683575|0|1641125130916|1683575 / 1641126814491|0 / 31523695|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|971268|731279|952296|57328|3.164291%|6|1876272|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **6 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988068|730145 / 1543589444152 (0.000047%)|735305 / 30813833 (2.386282%)|0.976268|0.976219|0.976218|
+|2|SNOWBALL FINNISH DIRECT|PRIMARY_OUTPUT|0.738400|1513705 / 1543589444152 (0.000098%)|16121763 / 30813833 (52.319888%)|0.768117|0.624934|0.657464|
+|3|SNOWBALL FINNISH LUCENE FILTER|PRIMARY_OUTPUT|0.738400|1513705 / 1543589444152 (0.000098%)|16121763 / 30813833 (52.319888%)|0.768117|0.624934|0.657464|
+|4|FINNISH LUCENE FINNISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.694529|1806392 / 1543589444152 (0.000117%)|18825444 / 30813833 (61.094133%)|0.697056|0.537492|0.581469|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.976301|0.976137|1.000000|0.988068|0.999999|0.000001|
+|2|SNOWBALL FINNISH DIRECT|PRIMARY_OUTPUT|0.906595|0.476801|0.999999|0.738400|0.999989|0.000011|
+|3|SNOWBALL FINNISH LUCENE FILTER|PRIMARY_OUTPUT|0.906595|0.476801|0.999999|0.738400|0.999989|0.000011|
+|4|FINNISH LUCENE FINNISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.869053|0.389059|0.999999|0.694529|0.999987|0.000013|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.976268|0.976219|0.976170|0.953543|0.976219|0.976218|
+|2|SNOWBALL FINNISH DIRECT|PRIMARY_OUTPUT|0.768117|0.624934|0.526744|0.454475|0.657469|0.657464|
+|3|SNOWBALL FINNISH LUCENE FILTER|PRIMARY_OUTPUT|0.768117|0.624934|0.526744|0.454475|0.657469|0.657464|
+|4|FINNISH LUCENE FINNISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.697056|0.537492|0.437372|0.367514|0.581474|0.581469|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.976218|0.996000|0.996069|0.996035|0.996035|
+|2|SNOWBALL FINNISH DIRECT|PRIMARY_OUTPUT|0.624929|0.991732|0.902933|0.945252|0.945252|
+|3|SNOWBALL FINNISH LUCENE FILTER|PRIMARY_OUTPUT|0.624929|0.991732|0.902933|0.945252|0.945252|
+|4|FINNISH LUCENE FINNISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.537486|0.989268|0.885294|0.934397|0.934397|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|30078528|730145|735305|1543588714007|730145 / 1543589444152|735305 / 30813833|
+|2|SNOWBALL FINNISH DIRECT|PRIMARY_OUTPUT|14692070|1513705|16121763|1543587930447|1513705 / 1543589444152|16121763 / 30813833|
+|3|SNOWBALL FINNISH LUCENE FILTER|PRIMARY_OUTPUT|14692070|1513705|16121763|1543587930447|1513705 / 1543589444152|16121763 / 30813833|
+|4|FINNISH LUCENE FINNISH LIGHT STEM FILTER|PRIMARY_OUTPUT|11988389|1806392|18825444|1543587637760|1806392 / 1543589444152|18825444 / 30813833|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 1543589444152 (0.000000%)|0 / 30813833 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|30813833|0|0|1543589444152|0 / 1543589444152|0 / 30813833|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999999|1653320 / 1543589444152 (0.000107%)|0 / 30813833 (0.000000%)|0.958843|0.973873|0.974205|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.949077|1.000000|0.999999|0.999999|0.999999|0.000001|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.958843|0.973873|0.989383|0.949077|0.974206|0.974205|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|30813833|1653320|0|1543587790832|1653320 / 1543589444152|0 / 30813833|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|735305|730145|923175|44331|2.523029%|6|1805864|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `FI_FI`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/french.md

@@ -0,0 +1,454 @@
+# French Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for French. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `FR_FR` | 59,240 | 474,110 | 108,141 | 365,969 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **474,110**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 5,370 | 1.133% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 185,263 | 39.076% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 153,886 | 32.458% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 116,519 | 24.576% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 13,072 | 2.757% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 94.831% | 94.859% | 94.734% | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | 68.923% | 63.617% | 86.876% | Benchmark-only French Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene FrenchMinimalStemFilter | 11.472% | 6.236% | 29.192% | Minimal suffix reducer; narrow baseline, not a full stemmer. |
+| Lucene SnowballFilter | 8.551% | 5.183% | 19.952% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Official Snowball direct | 8.462% | 5.067% | 19.952% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+| Lucene FrenchLightStemFilter | 6.377% | 3.965% | 14.540% | Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `frenchRadixor` | 47.033 | 4.146 | 128.5 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 1664.935 | 65.928 | 4549.4 | 35.399 | Benchmark-only French Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene FrenchMinimalStemFilter | `frenchLuceneFrenchMinimalStemFilter` | 19.234 | 2.098 | 52.6 | 0.409 | Minimal French suffix reducer; narrow baseline. |
+| Lucene FrenchLightStemFilter | `frenchLuceneFrenchLightStemFilter` | 30.560 | 3.680 | 83.5 | 0.650 | Light French suffix stemmer. |
+| Official Snowball direct | `snowballDirect[FRENCH]` | 111.057 | 8.172 | 303.5 | 2.361 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[FRENCH]` | 123.648 | 3.500 | 337.9 | 2.629 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `FR_FR` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/fr_fr/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.956992** among 6 deterministic stemmers. The runner-up is `SNOWBALL FRENCH DIRECT` at 0.845262, a difference of 0.111731. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.957224** among 6 deterministic stemmers. The runner-up is `SNOWBALL FRENCH DIRECT` at 0.845414, a difference of 0.111810. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **10 result rows**, **6 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.956992|318767 / 90396104830 (0.000353%)|469160 / 5454615 (8.601157%)|0.934603|0.926765|0.926851|
+|2|SNOWBALL FRENCH DIRECT|PRIMARY_OUTPUT|0.845262|1654723 / 90396104830 (0.001831%)|1687975 / 5454615 (30.945814%)|0.693926|0.692653|0.692638|
+|3|SNOWBALL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.844999|1661388 / 90396104830 (0.001838%)|1690838 / 5454615 (30.998301%)|0.693010|0.691885|0.691869|
+|4|HUNSPELL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.813742|776728 / 90396104830 (0.000859%)|2031881 / 5454615 (37.250677%)|0.769069|0.709075|0.715131|
+|5|FRENCH LUCENE FRENCH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.518587|276403 / 90396104830 (0.000306%)|5251833 / 5454615 (96.282377%)|0.137547|0.068348|0.125415|
+|6|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.516830|160438 / 90396104830 (0.000177%)|5271003 / 5454615 (96.633823%)|0.134400|0.063329|0.134021|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.939903|0.913988|0.999996|0.956992|0.999991|0.000009|
+|2|SNOWBALL FRENCH DIRECT|PRIMARY_OUTPUT|0.694777|0.690542|0.999982|0.845262|0.999963|0.000037|
+|3|SNOWBALL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.693763|0.690017|0.999982|0.844999|0.999963|0.000037|
+|4|HUNSPELL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.815041|0.627493|0.999991|0.813742|0.999969|0.000031|
+|5|FRENCH LUCENE FRENCH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.423181|0.037176|0.999997|0.518587|0.999939|0.000061|
+|6|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.533678|0.033662|0.999998|0.516830|0.999940|0.000060|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.934603|0.926765|0.919056|0.863524|0.926855|0.926851|
+|2|SNOWBALL FRENCH DIRECT|PRIMARY_OUTPUT|0.693926|0.692653|0.691385|0.529816|0.692656|0.692638|
+|3|SNOWBALL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.693010|0.691885|0.690763|0.528917|0.691887|0.691869|
+|4|HUNSPELL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.769069|0.709075|0.657765|0.549277|0.715145|0.715131|
+|5|FRENCH LUCENE FRENCH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.137547|0.068348|0.045472|0.035383|0.125428|0.125415|
+|6|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.134400|0.063329|0.041424|0.032700|0.134032|0.134021|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.926760|0.988772|0.985214|0.986990|0.986990|
+|2|SNOWBALL FRENCH DIRECT|PRIMARY_OUTPUT|0.692635|0.959459|0.944948|0.952148|0.952148|
+|3|SNOWBALL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.691866|0.958698|0.944715|0.951655|0.951655|
+|4|HUNSPELL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.709060|0.978337|0.913706|0.944918|0.944918|
+|5|FRENCH LUCENE FRENCH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.068339|0.974110|0.812376|0.885922|0.885922|
+|6|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.063322|0.984019|0.810979|0.889158|0.889158|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|4985455|318767|469160|90395786063|318767 / 90396104830|469160 / 5454615|
+|2|SNOWBALL FRENCH DIRECT|PRIMARY_OUTPUT|3766640|1654723|1687975|90394450107|1654723 / 90396104830|1687975 / 5454615|
+|3|SNOWBALL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|3763777|1661388|1690838|90394443442|1661388 / 90396104830|1690838 / 5454615|
+|4|HUNSPELL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|3422734|776728|2031881|90395328102|776728 / 90396104830|2031881 / 5454615|
+|5|FRENCH LUCENE FRENCH LIGHT STEM FILTER|PRIMARY_OUTPUT|202782|276403|5251833|90395828427|276403 / 90396104830|5251833 / 5454615|
+|6|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|PRIMARY_OUTPUT|183612|160438|5271003|90395944392|160438 / 90396104830|5271003 / 5454615|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.999979|12 / 90396104830 (0.000000%)|232 / 5454615 (0.004253%)|0.999990|0.999978|0.999978|
+|2|HUNSPELL FRENCH LUCENE FILTER|ANY_CANDIDATE|0.830964|745831 / 90396104830 (0.000825%)|1844003 / 5454615 (33.806291%)|0.789019|0.736029|0.740670|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.999998|0.999957|1.000000|0.999979|1.000000|0.000000|
+|2|HUNSPELL FRENCH LUCENE FILTER|ANY_CANDIDATE|0.828798|0.661937|0.999992|0.830964|0.999971|0.000029|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.999990|0.999978|0.999966|0.999955|0.999978|0.999978|
+|2|HUNSPELL FRENCH LUCENE FILTER|ANY_CANDIDATE|0.789019|0.736029|0.689709|0.582315|0.740684|0.740670|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL FRENCH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|5454383|12|232|90396104818|12 / 90396104830|232 / 5454615|
+|2|HUNSPELL FRENCH LUCENE FILTER|ANY_CANDIDATE|3610612|745831|1844003|90395358999|745831 / 90396104830|1844003 / 5454615|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999973|1056255 / 90396104830 (0.001168%)|232 / 5454615 (0.004253%)|0.865853|0.911704|0.915270|
+|2|HUNSPELL FRENCH LUCENE FILTER|ALL_CANDIDATES|0.830963|1043199 / 90396104830 (0.001154%)|1844003 / 5454615 (33.806291%)|0.750028|0.714377|0.716613|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.837765|0.999957|0.999988|0.999973|0.999988|0.000012|
+|2|HUNSPELL FRENCH LUCENE FILTER|ALL_CANDIDATES|0.775840|0.661937|0.999988|0.830963|0.999968|0.000032|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.865853|0.911704|0.962682|0.837735|0.915275|0.915270|
+|2|HUNSPELL FRENCH LUCENE FILTER|ALL_CANDIDATES|0.750028|0.714377|0.681961|0.555666|0.716629|0.716613|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL FRENCH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|5454383|1056255|232|90395048575|1056255 / 90396104830|232 / 5454615|
+|2|HUNSPELL FRENCH LUCENE FILTER|ALL_CANDIDATES|3610612|1043199|1844003|90395061631|1043199 / 90396104830|1844003 / 5454615|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|468928|318755|737488|43040|10.122057%|56|477024|
+|HUNSPELL FRENCH LUCENE FILTER|187878|30897|266471|13511|3.177489%|4|439015|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **10 result rows**, **6 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.957224|315266 / 88712126506 (0.000355%)|465436 / 5440559 (8.554930%)|0.935099|0.927248|0.927334|
+|2|SNOWBALL FRENCH DIRECT|PRIMARY_OUTPUT|0.845414|1646111 / 88712126506 (0.001856%)|1681970 / 5440559 (30.915389%)|0.694508|0.693130|0.693115|
+|3|SNOWBALL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.845163|1641925 / 88712126506 (0.001851%)|1684703 / 5440559 (30.965623%)|0.694714|0.693068|0.693055|
+|4|HUNSPELL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.813617|763305 / 88712126506 (0.000860%)|2028011 / 5440559 (37.275784%)|0.770537|0.709734|0.715938|
+|5|FRENCH LUCENE FRENCH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.518442|262689 / 88712126506 (0.000296%)|5239869 / 5440559 (96.311225%)|0.137571|0.067985|0.126383|
+|6|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.516697|147476 / 88712126506 (0.000166%)|5258873 / 5440559 (96.660527%)|0.134439|0.062979|0.135757|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.940408|0.914451|0.999996|0.957224|0.999991|0.000009|
+|2|SNOWBALL FRENCH DIRECT|PRIMARY_OUTPUT|0.695430|0.690846|0.999981|0.845414|0.999962|0.000038|
+|3|SNOWBALL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.695815|0.690344|0.999981|0.845163|0.999963|0.000037|
+|4|HUNSPELL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.817210|0.627242|0.999991|0.813617|0.999969|0.000031|
+|5|FRENCH LUCENE FRENCH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.433101|0.036888|0.999997|0.518442|0.999938|0.000062|
+|6|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.551965|0.033395|0.999998|0.516697|0.999939|0.000061|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.935099|0.927248|0.919527|0.864363|0.927338|0.927334|
+|2|SNOWBALL FRENCH DIRECT|PRIMARY_OUTPUT|0.694508|0.693130|0.691758|0.530374|0.693134|0.693115|
+|3|SNOWBALL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.694714|0.693068|0.691431|0.530302|0.693074|0.693055|
+|4|HUNSPELL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.770537|0.709734|0.657826|0.550068|0.715953|0.715938|
+|5|FRENCH LUCENE FRENCH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.137571|0.067985|0.045148|0.035189|0.126397|0.126383|
+|6|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.134439|0.062979|0.041121|0.032513|0.135767|0.135757|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.927243|0.988916|0.985550|0.987230|0.987230|
+|2|SNOWBALL FRENCH DIRECT|PRIMARY_OUTPUT|0.693112|0.959521|0.944537|0.951970|0.951970|
+|3|SNOWBALL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.693050|0.959566|0.944385|0.951915|0.951915|
+|4|HUNSPELL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|0.709719|0.979328|0.913162|0.945088|0.945088|
+|5|FRENCH LUCENE FRENCH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.067976|0.975086|0.811144|0.885591|0.885591|
+|6|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.062973|0.985086|0.809774|0.888868|0.888868|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|4975123|315266|465436|88711811240|315266 / 88712126506|465436 / 5440559|
+|2|SNOWBALL FRENCH DIRECT|PRIMARY_OUTPUT|3758589|1646111|1681970|88710480395|1646111 / 88712126506|1681970 / 5440559|
+|3|SNOWBALL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|3755856|1641925|1684703|88710484581|1641925 / 88712126506|1684703 / 5440559|
+|4|HUNSPELL FRENCH LUCENE FILTER|PRIMARY_OUTPUT|3412548|763305|2028011|88711363201|763305 / 88712126506|2028011 / 5440559|
+|5|FRENCH LUCENE FRENCH LIGHT STEM FILTER|PRIMARY_OUTPUT|200690|262689|5239869|88711863817|262689 / 88712126506|5239869 / 5440559|
+|6|FRENCH LUCENE FRENCH MINIMAL STEM FILTER|PRIMARY_OUTPUT|181686|147476|5258873|88711979030|147476 / 88712126506|5258873 / 5440559|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 88712126506 (0.000000%)|0 / 5440559 (0.000000%)|1.000000|1.000000|1.000000|
+|2|HUNSPELL FRENCH LUCENE FILTER|ANY_CANDIDATE|0.830852|733584 / 88712126506 (0.000827%)|1840476 / 5440559 (33.828803%)|0.790351|0.736648|0.741404|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|HUNSPELL FRENCH LUCENE FILTER|ANY_CANDIDATE|0.830724|0.661712|0.999992|0.830852|0.999971|0.000029|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|HUNSPELL FRENCH LUCENE FILTER|ANY_CANDIDATE|0.790351|0.736648|0.689779|0.583090|0.741418|0.741404|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL FRENCH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|5440559|0|0|88712126506|0 / 88712126506|0 / 5440559|
+|2|HUNSPELL FRENCH LUCENE FILTER|ANY_CANDIDATE|3600083|733584|1840476|88711392922|733584 / 88712126506|1840476 / 5440559|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999995|938985 / 88712126506 (0.001058%)|0 / 5440559 (0.000000%)|0.878679|0.920560|0.923474|
+|2|HUNSPELL FRENCH LUCENE FILTER|ALL_CANDIDATES|0.830850|1027635 / 88712126506 (0.001158%)|1840476 / 5440559 (33.828803%)|0.751538|0.715134|0.717460|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.852813|1.000000|0.999989|0.999995|0.999989|0.000011|
+|2|HUNSPELL FRENCH LUCENE FILTER|ALL_CANDIDATES|0.777939|0.661712|0.999988|0.830850|0.999968|0.000032|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.878679|0.920560|0.966634|0.852813|0.923479|0.923474|
+|2|HUNSPELL FRENCH LUCENE FILTER|ALL_CANDIDATES|0.751538|0.715134|0.682093|0.556582|0.717476|0.717460|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL FRENCH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|5440559|938985|0|88711187521|938985 / 88712126506|0 / 5440559|
+|2|HUNSPELL FRENCH LUCENE FILTER|ALL_CANDIDATES|3600083|1027635|1840476|88711098871|1027635 / 88712126506|1840476 / 5440559|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|465436|315266|623719|41130|9.764239%|56|468574|
+|HUNSPELL FRENCH LUCENE FILTER|187535|29721|264330|13437|3.189936%|4|434961|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `FR_FR`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/german.md

@@ -0,0 +1,478 @@
+# German Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for German. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `DE_DE` | 39,315 | 213,440 | 73,799 | 139,641 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **213,440**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 3,627 | 1.699% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 48,605 | 22.772% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 80,443 | 37.689% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 75,717 | 35.475% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 5,048 | 2.365% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 92.725% | 92.847% | 92.396% | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | 47.064% | 29.661% | 93.678% | Benchmark-only German Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| CISTEM (German) | 24.675% | 23.724% | 27.222% | Benchmark-only CISTEM implementation. |
+| Lucene GermanLightStemFilter | 37.434% | 35.465% | 42.707% | Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer. |
+| Lucene GermanMinimalStemFilter | 27.640% | 24.951% | 34.844% | Minimal suffix reducer; narrow baseline, not a full stemmer. |
+| Lucene SnowballFilter | 30.956% | 28.853% | 36.589% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Official Snowball direct | 30.481% | 29.027% | 34.376% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+| Lucene GermanStemFilter | 21.559% | 19.312% | 27.576% | German Lucene stemming TokenFilter; broader than minimal/light variants. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `germanRadixor` | 41.166 | 2.396 | 294.8 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| CISTEM | `germanCistem` | 248.392 | 12.294 | 1778.8 | 6.034 | Benchmark-only CISTEM implementation. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 281.322 | 3.411 | 2014.6 | 6.834 | Benchmark-only German Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene GermanMinimalStemFilter | `germanLuceneGermanMinimalStemFilter` | 23.562 | 0.969 | 168.7 | 0.572 | Minimal German suffix reduction; narrow baseline. |
+| Lucene GermanLightStemFilter | `germanLuceneGermanLightStemFilter` | 24.410 | 1.034 | 174.8 | 0.593 | Light German suffix stemmer; narrower than a dictionary stemmer. |
+| Lucene GermanStemFilter | `germanLuceneGermanStemFilter` | 71.039 | 4.443 | 508.7 | 1.726 | Older German stemming TokenFilter with normalization requirements. |
+| Lucene SnowballFilter | `luceneSnowballFilter[GERMAN]` | 105.771 | 9.617 | 757.4 | 2.569 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Official Snowball direct | `snowballDirect[GERMAN]` | 100.688 | 9.018 | 721.0 | 2.446 | Official Snowball generated Java stemmer; direct API. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `DE_DE` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/de_de/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.907901** among 8 deterministic stemmers. The runner-up is `GERMAN CISTEM` at 0.880770, a difference of 0.027131. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.966157** among 8 deterministic stemmers. The runner-up is `GERMAN CISTEM` at 0.915288, a difference of 0.050869. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **12 result rows**, **8 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.907901|98192 / 44095245979 (0.000223%)|254903 / 1383872 (18.419550%)|0.897073|0.864768|0.866326|
+|2|GERMAN CISTEM|PRIMARY_OUTPUT|0.880770|477122 / 44095245979 (0.001082%)|329983 / 1383872 (23.844908%)|0.701852|0.723109|0.724023|
+|3|SNOWBALL GERMAN DIRECT|PRIMARY_OUTPUT|0.778614|190680 / 44095245979 (0.000432%)|612734 / 1383872 (44.276783%)|0.737064|0.657494|0.668394|
+|4|SNOWBALL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.771357|295701 / 44095245979 (0.000671%)|632816 / 1383872 (45.727929%)|0.674089|0.617993|0.624014|
+|5|GERMAN LUCENE GERMAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.756258|205740 / 44095245979 (0.000467%)|674609 / 1383872 (48.747933%)|0.703092|0.617052|0.630292|
+|6|GERMAN LUCENE GERMAN STEM FILTER|PRIMARY_OUTPUT|0.723772|331871 / 44095245979 (0.000753%)|764518 / 1383872 (55.244849%)|0.596821|0.530474|0.539809|
+|7|HUNSPELL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.641579|203883 / 44095245979 (0.000462%)|992010 / 1383872 (71.683653%)|0.520145|0.395897|0.431563|
+|8|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.598139|110840 / 44095245979 (0.000251%)|1112246 / 1383872 (80.372029%)|0.466113|0.307558|0.373350|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.919984|0.815804|0.999998|0.907901|0.999992|0.000008|
+|2|GERMAN CISTEM|PRIMARY_OUTPUT|0.688361|0.761551|0.999989|0.880770|0.999982|0.000018|
+|3|SNOWBALL GERMAN DIRECT|PRIMARY_OUTPUT|0.801750|0.557232|0.999996|0.778614|0.999982|0.000018|
+|4|SNOWBALL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.717508|0.542721|0.999993|0.771357|0.999979|0.000021|
+|5|GERMAN LUCENE GERMAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.775148|0.512521|0.999995|0.756258|0.999980|0.000020|
+|6|GERMAN LUCENE GERMAN STEM FILTER|PRIMARY_OUTPUT|0.651112|0.447552|0.999992|0.723772|0.999975|0.000025|
+|7|HUNSPELL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.657768|0.283163|0.999995|0.641579|0.999973|0.000027|
+|8|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.710196|0.196280|0.999997|0.598139|0.999972|0.000028|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.897073|0.864768|0.834709|0.761755|0.866330|0.866326|
+|2|GERMAN CISTEM|PRIMARY_OUTPUT|0.701852|0.723109|0.745694|0.566304|0.724032|0.724023|
+|3|SNOWBALL GERMAN DIRECT|PRIMARY_OUTPUT|0.737064|0.657494|0.593429|0.489751|0.668402|0.668394|
+|4|SNOWBALL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.674089|0.617993|0.570517|0.447171|0.624024|0.624014|
+|5|GERMAN LUCENE GERMAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.703092|0.617052|0.549774|0.446186|0.630301|0.630292|
+|6|GERMAN LUCENE GERMAN STEM FILTER|PRIMARY_OUTPUT|0.596821|0.530474|0.477402|0.360983|0.539820|0.539809|
+|7|HUNSPELL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.520145|0.395897|0.319562|0.246803|0.431574|0.431563|
+|8|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.466113|0.307558|0.229493|0.181725|0.373359|0.373350|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.864764|0.989946|0.975085|0.982460|0.982460|
+|2|GERMAN CISTEM|PRIMARY_OUTPUT|0.723100|0.974048|0.975147|0.974597|0.974597|
+|3|SNOWBALL GERMAN DIRECT|PRIMARY_OUTPUT|0.657485|0.983725|0.949324|0.966218|0.966218|
+|4|SNOWBALL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.617983|0.975845|0.942925|0.959102|0.959102|
+|5|GERMAN LUCENE GERMAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.617043|0.980753|0.936533|0.958133|0.958133|
+|6|GERMAN LUCENE GERMAN STEM FILTER|PRIMARY_OUTPUT|0.530462|0.975550|0.942890|0.958942|0.958942|
+|7|HUNSPELL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.395885|0.980463|0.886873|0.931322|0.931322|
+|8|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.307549|0.983615|0.896264|0.937910|0.937910|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|1128969|98192|254903|44095147787|98192 / 44095245979|254903 / 1383872|
+|2|GERMAN CISTEM|PRIMARY_OUTPUT|1053889|477122|329983|44094768857|477122 / 44095245979|329983 / 1383872|
+|3|SNOWBALL GERMAN DIRECT|PRIMARY_OUTPUT|771138|190680|612734|44095055299|190680 / 44095245979|612734 / 1383872|
+|4|SNOWBALL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|751056|295701|632816|44094950278|295701 / 44095245979|632816 / 1383872|
+|5|GERMAN LUCENE GERMAN LIGHT STEM FILTER|PRIMARY_OUTPUT|709263|205740|674609|44095040239|205740 / 44095245979|674609 / 1383872|
+|6|GERMAN LUCENE GERMAN STEM FILTER|PRIMARY_OUTPUT|619354|331871|764518|44094914108|331871 / 44095245979|764518 / 1383872|
+|7|HUNSPELL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|391862|203883|992010|44095042096|203883 / 44095245979|992010 / 1383872|
+|8|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|271626|110840|1112246|44095135139|110840 / 44095245979|1112246 / 1383872|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.959835|1375 / 44095245979 (0.000003%)|111167 / 1383872 (8.033041%)|0.981996|0.957658|0.958475|
+|2|HUNSPELL GERMAN LUCENE FILTER|ANY_CANDIDATE|0.647474|158403 / 44095245979 (0.000359%)|975697 / 1383872 (70.504859%)|0.559116|0.418544|0.460956|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.998921|0.919670|1.000000|0.959835|0.999997|0.000003|
+|2|HUNSPELL GERMAN LUCENE FILTER|ANY_CANDIDATE|0.720422|0.294951|0.999996|0.647474|0.999974|0.000026|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.981996|0.957658|0.934498|0.918757|0.958476|0.958475|
+|2|HUNSPELL GERMAN LUCENE FILTER|ANY_CANDIDATE|0.559116|0.418544|0.334456|0.264658|0.460966|0.460956|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL GERMAN LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1272705|1375|111167|44095244604|1375 / 44095245979|111167 / 1383872|
+|2|HUNSPELL GERMAN LUCENE FILTER|ANY_CANDIDATE|408175|158403|975697|44095087576|158403 / 44095245979|975697 / 1383872|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.959832|244817 / 44095245979 (0.000555%)|111167 / 1383872 (8.033041%)|0.853711|0.877306|0.878234|
+|2|HUNSPELL GERMAN LUCENE FILTER|ALL_CANDIDATES|0.647473|242551 / 44095245979 (0.000550%)|975697 / 1383872 (70.504859%)|0.511911|0.401234|0.430118|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.838673|0.919670|0.999994|0.959832|0.999992|0.000008|
+|2|HUNSPELL GERMAN LUCENE FILTER|ALL_CANDIDATES|0.627261|0.294951|0.999994|0.647473|0.999972|0.000028|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.853711|0.877306|0.902242|0.781429|0.878238|0.878234|
+|2|HUNSPELL GERMAN LUCENE FILTER|ALL_CANDIDATES|0.511911|0.401234|0.329907|0.250965|0.430130|0.430118|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL GERMAN LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|1272705|244817|111167|44095001162|244817 / 44095245979|111167 / 1383872|
+|2|HUNSPELL GERMAN LUCENE FILTER|ALL_CANDIDATES|408175|242551|975697|44095003428|242551 / 44095245979|975697 / 1383872|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|143736|96817|146625|48574|16.356314%|8|361016|
+|HUNSPELL GERMAN LUCENE FILTER|16313|45480|38668|7891|2.657135%|3|305052|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **12 result rows**, **8 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.966157|47898 / 11263756342 (0.000425%)|59114 / 873411 (6.768177%)|0.941996|0.938343|0.938358|
+|2|GERMAN CISTEM|PRIMARY_OUTPUT|0.915288|156784 / 11263756342 (0.001392%)|147964 / 873411 (16.940936%)|0.823934|0.826418|0.826415|
+|3|SNOWBALL GERMAN DIRECT|PRIMARY_OUTPUT|0.795926|87697 / 11263756342 (0.000779%)|356475 / 873411 (40.814118%)|0.785153|0.699487|0.711329|
+|4|SNOWBALL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.775641|77653 / 11263756342 (0.000689%)|391910 / 873411 (44.871200%)|0.774111|0.672222|0.688986|
+|5|GERMAN LUCENE GERMAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.769953|55477 / 11263756342 (0.000493%)|401846 / 873411 (46.008809%)|0.790797|0.673446|0.695023|
+|6|GERMAN LUCENE GERMAN STEM FILTER|PRIMARY_OUTPUT|0.716810|78723 / 11263756342 (0.000699%)|494677 / 873411 (56.637368%)|0.700519|0.569153|0.599149|
+|7|HUNSPELL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.659196|84679 / 11263756342 (0.000752%)|595318 / 873411 (68.160122%)|0.598178|0.449922|0.494019|
+|8|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.575691|21214 / 11263756342 (0.000188%)|741190 / 873411 (84.861537%)|0.444545|0.257528|0.361168|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.944446|0.932318|0.999996|0.966157|0.999991|0.000009|
+|2|GERMAN CISTEM|PRIMARY_OUTPUT|0.822287|0.830591|0.999986|0.915288|0.999973|0.000027|
+|3|SNOWBALL GERMAN DIRECT|PRIMARY_OUTPUT|0.854958|0.591859|0.999992|0.795926|0.999961|0.000039|
+|4|SNOWBALL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.861124|0.551288|0.999993|0.775641|0.999958|0.000042|
+|5|GERMAN LUCENE GERMAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.894739|0.539912|0.999995|0.769953|0.999959|0.000041|
+|6|GERMAN LUCENE GERMAN STEM FILTER|PRIMARY_OUTPUT|0.827912|0.433626|0.999993|0.716810|0.999949|0.000051|
+|7|HUNSPELL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.766578|0.318399|0.999992|0.659196|0.999940|0.000060|
+|8|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.861739|0.151385|0.999998|0.575691|0.999932|0.000068|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.941996|0.938343|0.934719|0.883848|0.938363|0.938358|
+|2|GERMAN CISTEM|PRIMARY_OUTPUT|0.823934|0.826418|0.828917|0.704184|0.826428|0.826415|
+|3|SNOWBALL GERMAN DIRECT|PRIMARY_OUTPUT|0.785153|0.699487|0.630675|0.537854|0.711347|0.711329|
+|4|SNOWBALL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.774111|0.672222|0.594035|0.506276|0.689005|0.688986|
+|5|GERMAN LUCENE GERMAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.790797|0.673446|0.586424|0.507666|0.695040|0.695023|
+|6|GERMAN LUCENE GERMAN STEM FILTER|PRIMARY_OUTPUT|0.700519|0.569153|0.479277|0.397774|0.599170|0.599149|
+|7|HUNSPELL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.598178|0.449922|0.360559|0.290258|0.494042|0.494019|
+|8|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.444545|0.257528|0.181270|0.147795|0.361184|0.361168|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.938338|0.994062|0.990664|0.992360|0.992360|
+|2|GERMAN CISTEM|PRIMARY_OUTPUT|0.826404|0.985936|0.973570|0.979714|0.979714|
+|3|SNOWBALL GERMAN DIRECT|PRIMARY_OUTPUT|0.699468|0.988418|0.932452|0.959619|0.959619|
+|4|SNOWBALL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.672202|0.989021|0.919542|0.953017|0.953017|
+|5|GERMAN LUCENE GERMAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.673427|0.991320|0.915070|0.951670|0.951670|
+|6|GERMAN LUCENE GERMAN STEM FILTER|PRIMARY_OUTPUT|0.569130|0.988584|0.918718|0.952371|0.952371|
+|7|HUNSPELL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|0.449897|0.988041|0.865581|0.922766|0.922766|
+|8|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.257511|0.992643|0.854403|0.918349|0.918349|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|814297|47898|59114|11263708444|47898 / 11263756342|59114 / 873411|
+|2|GERMAN CISTEM|PRIMARY_OUTPUT|725447|156784|147964|11263599558|156784 / 11263756342|147964 / 873411|
+|3|SNOWBALL GERMAN DIRECT|PRIMARY_OUTPUT|516936|87697|356475|11263668645|87697 / 11263756342|356475 / 873411|
+|4|SNOWBALL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|481501|77653|391910|11263678689|77653 / 11263756342|391910 / 873411|
+|5|GERMAN LUCENE GERMAN LIGHT STEM FILTER|PRIMARY_OUTPUT|471565|55477|401846|11263700865|55477 / 11263756342|401846 / 873411|
+|6|GERMAN LUCENE GERMAN STEM FILTER|PRIMARY_OUTPUT|378734|78723|494677|11263677619|78723 / 11263756342|494677 / 873411|
+|7|HUNSPELL GERMAN LUCENE FILTER|PRIMARY_OUTPUT|278093|84679|595318|11263671663|84679 / 11263756342|595318 / 873411|
+|8|GERMAN LUCENE GERMAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|132221|21214|741190|11263735128|21214 / 11263756342|741190 / 873411|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 11263756342 (0.000000%)|0 / 873411 (0.000000%)|1.000000|1.000000|1.000000|
+|2|HUNSPELL GERMAN LUCENE FILTER|ANY_CANDIDATE|0.665363|60996 / 11263756342 (0.000542%)|584547 / 873411 (66.926911%)|0.635466|0.472281|0.522540|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|HUNSPELL GERMAN LUCENE FILTER|ANY_CANDIDATE|0.825656|0.330731|0.999995|0.665363|0.999943|0.000057|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|HUNSPELL GERMAN LUCENE FILTER|ANY_CANDIDATE|0.635466|0.472281|0.375782|0.309142|0.522561|0.522540|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL GERMAN LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|873411|0|0|11263756342|0 / 11263756342|0 / 873411|
+|2|HUNSPELL GERMAN LUCENE FILTER|ANY_CANDIDATE|288864|60996|584547|11263695346|60996 / 11263756342|584547 / 873411|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999996|97544 / 11263756342 (0.000866%)|0 / 873411 (0.000000%)|0.917983|0.947112|0.948436|
+|2|HUNSPELL GERMAN LUCENE FILTER|ALL_CANDIDATES|0.665361|96545 / 11263756342 (0.000857%)|584547 / 873411 (66.926911%)|0.598050|0.458944|0.497855|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.899538|1.000000|0.999991|0.999996|0.999991|0.000009|
+|2|HUNSPELL GERMAN LUCENE FILTER|ALL_CANDIDATES|0.749500|0.330731|0.999991|0.665361|0.999940|0.000060|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.917983|0.947112|0.978152|0.899538|0.948440|0.948436|
+|2|HUNSPELL GERMAN LUCENE FILTER|ALL_CANDIDATES|0.598050|0.458944|0.372338|0.297811|0.497878|0.497855|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL GERMAN LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|873411|97544|0|11263658798|97544 / 11263756342|0 / 873411|
+|2|HUNSPELL GERMAN LUCENE FILTER|ALL_CANDIDATES|288864|96545|584547|11263659797|96545 / 11263756342|584547 / 873411|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|59114|47898|49646|14978|9.978814%|8|167157|
+|HUNSPELL GERMAN LUCENE FILTER|10771|23683|11866|4989|3.323828%|3|155207|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `DE_DE`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/hungarian.md

@@ -0,0 +1,408 @@
+# Hungarian Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Hungarian. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `HU_HU` | 19,406 | 935,713 | 38,775 | 896,938 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **935,713**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 15 | 0.002% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 149,173 | 15.942% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 746,296 | 79.757% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 40,125 | 4.288% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 104 | 0.011% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 99.222% | 99.537% | 91.948% | Full Radixor dictionary patch-command stemmer. |
+| Lucene SnowballFilter | 66.445% | 66.938% | 55.043% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Official Snowball direct | 66.445% | 66.938% | 55.043% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+| Lucene HungarianLightStemFilter | 14.748% | 14.777% | 14.086% | Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `hungarianRadixor` | 62.232 | 6.412 | 69.4 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene HungarianLightStemFilter | `hungarianLuceneHungarianLightStemFilter` | 92.813 | 6.929 | 103.5 | 1.491 | Light Hungarian suffix stemmer. |
+| Official Snowball direct | `snowballDirect[HUNGARIAN]` | 157.765 | 13.202 | 175.9 | 2.535 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[HUNGARIAN]` | 188.863 | 15.880 | 210.6 | 3.035 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `HU_HU` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/hu_hu/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.995491** among 4 deterministic stemmers. The runner-up is `SNOWBALL HUNGARIAN LUCENE FILTER` at 0.822606, a difference of 0.172885. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.996163** among 4 deterministic stemmers. The runner-up is `SNOWBALL HUNGARIAN DIRECT` at 0.821708, a difference of 0.174455. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **6 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.995491|272900 / 419820542893 (0.000065%)|199837 / 22162103 (0.901706%)|0.988376|0.989352|0.989353|
+|2|SNOWBALL HUNGARIAN LUCENE FILTER|PRIMARY_OUTPUT|0.822606|1792049 / 419820542893 (0.000427%)|7862745 / 22162103 (35.478334%)|0.826288|0.747610|0.757196|
+|3|SNOWBALL HUNGARIAN DIRECT|PRIMARY_OUTPUT|0.822348|1506056 / 419820542893 (0.000359%)|7874191 / 22162103 (35.529981%)|0.837137|0.752866|0.763681|
+|4|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.816668|4132555 / 419820542893 (0.000984%)|8125833 / 22162103 (36.665442%)|0.740018|0.696055|0.699478|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987727|0.990983|0.999999|0.995491|0.999999|0.000001|
+|2|SNOWBALL HUNGARIAN LUCENE FILTER|PRIMARY_OUTPUT|0.888633|0.645217|0.999996|0.822606|0.999977|0.000023|
+|3|SNOWBALL HUNGARIAN DIRECT|PRIMARY_OUTPUT|0.904644|0.644700|0.999996|0.822348|0.999978|0.000022|
+|4|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.772547|0.633346|0.999990|0.816668|0.999971|0.000029|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988376|0.989352|0.990330|0.978929|0.989353|0.989353|
+|2|SNOWBALL HUNGARIAN LUCENE FILTER|PRIMARY_OUTPUT|0.826288|0.747610|0.682613|0.596947|0.757206|0.757196|
+|3|SNOWBALL HUNGARIAN DIRECT|PRIMARY_OUTPUT|0.837137|0.752866|0.684009|0.603677|0.763691|0.763681|
+|4|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.740018|0.696055|0.657023|0.533807|0.699492|0.699478|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.989352|0.998036|0.997809|0.997922|0.997922|
+|2|SNOWBALL HUNGARIAN LUCENE FILTER|PRIMARY_OUTPUT|0.747599|0.990687|0.924490|0.956445|0.956445|
+|3|SNOWBALL HUNGARIAN DIRECT|PRIMARY_OUTPUT|0.752855|0.991948|0.924304|0.956932|0.956932|
+|4|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.696040|0.982615|0.926772|0.953877|0.953877|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|21962266|272900|199837|419820269993|272900 / 419820542893|199837 / 22162103|
+|2|SNOWBALL HUNGARIAN LUCENE FILTER|PRIMARY_OUTPUT|14299358|1792049|7862745|419818750844|1792049 / 419820542893|7862745 / 22162103|
+|3|SNOWBALL HUNGARIAN DIRECT|PRIMARY_OUTPUT|14287912|1506056|7874191|419819036837|1506056 / 419820542893|7874191 / 22162103|
+|4|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|14036270|4132555|8125833|419816410338|4132555 / 419820542893|8125833 / 22162103|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 419820542893 (0.000000%)|0 / 22162103 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|22162103|0|0|419820542893|0 / 419820542893|0 / 22162103|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999999|460158 / 419820542893 (0.000110%)|0 / 22162103 (0.000000%)|0.983661|0.989725|0.989777|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.979659|1.000000|0.999999|0.999999|0.999999|0.000001|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.983661|0.989725|0.995865|0.979659|0.989777|0.989777|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|22162103|460158|0|419820082735|460158 / 419820542893|0 / 22162103|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|199837|272900|187258|12320|1.344473%|5|929326|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **6 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996163|272775 / 385870694917 (0.000071%)|164277 / 21411411 (0.767240%)|0.988321|0.989820|0.989822|
+|2|SNOWBALL HUNGARIAN DIRECT|PRIMARY_OUTPUT|0.821708|1496670 / 385870694917 (0.000388%)|7634885 / 21411411 (35.658019%)|0.834899|0.751079|0.761809|
+|3|SNOWBALL HUNGARIAN LUCENE FILTER|PRIMARY_OUTPUT|0.821708|1496670 / 385870694917 (0.000388%)|7634885 / 21411411 (35.658019%)|0.834899|0.751079|0.761809|
+|4|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.815077|3639046 / 385870694917 (0.000943%)|7918708 / 21411411 (36.983588%)|0.750108|0.700135|0.704477|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987325|0.992328|0.999999|0.996163|0.999999|0.000001|
+|2|SNOWBALL HUNGARIAN DIRECT|PRIMARY_OUTPUT|0.902007|0.643420|0.999996|0.821708|0.999976|0.000024|
+|3|SNOWBALL HUNGARIAN LUCENE FILTER|PRIMARY_OUTPUT|0.902007|0.643420|0.999996|0.821708|0.999976|0.000024|
+|4|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.787585|0.630164|0.999991|0.815077|0.999970|0.000030|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988321|0.989820|0.991323|0.979845|0.989823|0.989822|
+|2|SNOWBALL HUNGARIAN DIRECT|PRIMARY_OUTPUT|0.834899|0.751079|0.682555|0.601383|0.761820|0.761809|
+|3|SNOWBALL HUNGARIAN LUCENE FILTER|PRIMARY_OUTPUT|0.834899|0.751079|0.682555|0.601383|0.761820|0.761809|
+|4|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.750108|0.700135|0.656404|0.538621|0.704491|0.704477|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.989819|0.997945|0.998273|0.998109|0.998109|
+|2|SNOWBALL HUNGARIAN DIRECT|PRIMARY_OUTPUT|0.751068|0.991610|0.923288|0.956230|0.956230|
+|3|SNOWBALL HUNGARIAN LUCENE FILTER|PRIMARY_OUTPUT|0.751068|0.991610|0.923288|0.956230|0.956230|
+|4|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.700120|0.983687|0.925487|0.953700|0.953700|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|21247134|272775|164277|385870422142|272775 / 385870694917|164277 / 21411411|
+|2|SNOWBALL HUNGARIAN DIRECT|PRIMARY_OUTPUT|13776526|1496670|7634885|385869198247|1496670 / 385870694917|7634885 / 21411411|
+|3|SNOWBALL HUNGARIAN LUCENE FILTER|PRIMARY_OUTPUT|13776526|1496670|7634885|385869198247|1496670 / 385870694917|7634885 / 21411411|
+|4|HUNGARIAN LUCENE HUNGARIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|13492703|3639046|7918708|385867055871|3639046 / 385870694917|7918708 / 21411411|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 385870694917 (0.000000%)|0 / 21411411 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|21411411|0|0|385870694917|0 / 385870694917|0 / 21411411|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999999|458462 / 385870694917 (0.000119%)|0 / 21411411 (0.000000%)|0.983159|0.989407|0.989462|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.979037|1.000000|0.999999|0.999999|0.999999|0.000001|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.983159|0.989407|0.995736|0.979037|0.989463|0.989462|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|21411411|458462|0|385870236455|458462 / 385870694917|0 / 21411411|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|164277|272775|185687|11153|1.269532%|5|890245|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `HU_HU`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/index.md

@@ -0,0 +1,45 @@
+# Language Benchmark Pages
+
+This section splits Radixor stemmer benchmark results by language. Each language page preserves the existing exact-root accuracy and runtime-performance results and adds pairwise stemming-quality tables for both dictionary-processing modes.
+
+## Reference Pages
+
+| Page | Purpose |
+| --- | --- |
+| [Methodology](../reference/methodology.md) | Workload design, normalization, speed metrics, and exact-root quality metrics. Pairwise quality definitions are also reproduced on every language page. |
+| [Corpora](../reference/corpora.md) | Dictionary sizes and changed-token timing workloads. |
+| [Environment and reports](../reference/environment.md) | Hardware, JVM, JMH settings, report files, and badge policy. |
+| [English dictionary coverage](../reference/english-coverage.md) | Quality/speed operating curve for contracted Radixor tries built from 100% down to 10% of English dictionary rows. |
+| [Candidate evaluation](../reference/candidates.md) | Included and skipped stemmer candidates. |
+
+## Languages
+
+| Language | Resource | Benchmark page |
+| --- | --- | --- |
+| Czech | `CS_CZ` | [Czech](czech.md) |
+| Danish | `DA_DK` | [Danish](danish.md) |
+| Dutch | `NL_NL` | [Dutch](dutch.md) |
+| English | `US_UK` | [English](english.md) |
+| Finnish | `FI_FI` | [Finnish](finnish.md) |
+| French | `FR_FR` | [French](french.md) |
+| German | `DE_DE` | [German](german.md) |
+| Hungarian | `HU_HU` | [Hungarian](hungarian.md) |
+| Italian | `IT_IT` | [Italian](italian.md) |
+| Norwegian Bokmal | `NB_NO` | [Norwegian Bokmal](norwegian-bokmal.md) |
+| Norwegian Nynorsk | `NN_NO` | [Norwegian Nynorsk](norwegian-nynorsk.md) |
+| Persian | `FA_IR` | [Persian](persian.md) |
+| Polish | `PL_PL` | [Polish](polish.md) |
+| Portuguese | `PT_PT` | [Portuguese](portuguese.md) |
+| Russian | `RU_RU` | [Russian](russian.md) |
+| Spanish | `ES_ES` | [Spanish](spanish.md) |
+| Swedish | `SV_SE` | [Swedish](swedish.md) |
+| Ukrainian | `UK_UA` | [Ukrainian](ukrainian.md) |
+| Yiddish | `YI` | [Yiddish](yiddish.md) |
+
+## Methodology Notes
+
+- Speed benchmarks process only changed dictionary tokens where the surface form differs from the expected root.
+- Accuracy benchmarks process the complete dictionary and report `All exact`, `Changed exact`, and `Root preserved`.
+- Radixor speed must be interpreted together with exact-root quality. A slower Radixor row must not be read as a simple performance weakness when Radixor is also the row with accuracy close to 100% and competing stemmers are much lower. Many fast light, minimal, possessive, or aggressive rule-based stemmers are fast because they do much less linguistic work. The measured Radixor cost buys dictionary-trained precision, and that precision is what improves search quality when queries and indexed text are reduced to the same intended roots. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows this contracted-trie operating curve explicitly.
+- Results are comparable only within the same language and benchmark family.
+- The historical Porter badge is retired; no JMH badge JSON is generated.

docs/benchmarks/languages/italian.md

@@ -0,0 +1,407 @@
+# Italian Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Italian. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `IT_IT` | 10,009 | 337,546 | 20,004 | 317,542 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **337,546**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 302,171 | 89.520% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 12,348 | 3.658% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 20,013 | 5.929% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 3,014 | 0.893% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 99.056% | 98.997% | 100.000% | Full Radixor dictionary patch-command stemmer. |
+| Lucene ItalianLightStemFilter | 0.466% | 0.479% | 0.270% | Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer. |
+| Lucene SnowballFilter | 0.041% | 0.043% | 0.010% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Official Snowball direct | 0.041% | 0.043% | 0.010% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `italianRadixor` | 24.491 | 3.128 | 77.1 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene ItalianLightStemFilter | `italianLuceneItalianLightStemFilter` | 15.977 | 1.041 | 50.3 | 0.652 | Light Italian suffix stemmer. |
+| Official Snowball direct | `snowballDirect[ITALIAN]` | 109.526 | 12.572 | 344.9 | 4.472 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[ITALIAN]` | 116.260 | 7.459 | 366.1 | 4.747 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `IT_IT` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/it_it/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.996507** among 4 deterministic stemmers. The runner-up is `SNOWBALL ITALIAN DIRECT` at 0.866189, a difference of 0.130318. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.996512** among 4 deterministic stemmers. The runner-up is `SNOWBALL ITALIAN DIRECT` at 0.866205, a difference of 0.130307. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **6 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996507|124172 / 53638521211 (0.000231%)|42908 / 6143814 (0.698394%)|0.982618|0.986492|0.986512|
+|2|SNOWBALL ITALIAN DIRECT|PRIMARY_OUTPUT|0.866189|504775 / 53638521211 (0.000941%)|1644164 / 6143814 (26.761292%)|0.859975|0.807240|0.811470|
+|3|SNOWBALL ITALIAN LUCENE FILTER|PRIMARY_OUTPUT|0.866189|504775 / 53638521211 (0.000941%)|1644164 / 6143814 (26.761292%)|0.859975|0.807240|0.811470|
+|4|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.508926|10589 / 53638521211 (0.000020%)|6034130 / 6143814 (98.214725%)|0.082782|0.035020|0.127588|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.980053|0.993016|0.999998|0.996507|0.999997|0.000003|
+|2|SNOWBALL ITALIAN DIRECT|PRIMARY_OUTPUT|0.899134|0.732387|0.999991|0.866189|0.999960|0.000040|
+|3|SNOWBALL ITALIAN LUCENE FILTER|PRIMARY_OUTPUT|0.899134|0.732387|0.999991|0.866189|0.999960|0.000040|
+|4|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.911959|0.017853|1.000000|0.508926|0.999887|0.000113|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.982618|0.986492|0.990396|0.973344|0.986513|0.986512|
+|2|SNOWBALL ITALIAN DIRECT|PRIMARY_OUTPUT|0.859975|0.807240|0.760598|0.676783|0.811489|0.811470|
+|3|SNOWBALL ITALIAN LUCENE FILTER|PRIMARY_OUTPUT|0.859975|0.807240|0.760598|0.676783|0.811489|0.811470|
+|4|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.082782|0.035020|0.022207|0.017822|0.127597|0.127588|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.986490|0.995780|0.997113|0.996446|0.996446|
+|2|SNOWBALL ITALIAN DIRECT|PRIMARY_OUTPUT|0.807220|0.987994|0.933408|0.959925|0.959925|
+|3|SNOWBALL ITALIAN LUCENE FILTER|PRIMARY_OUTPUT|0.807220|0.987994|0.933408|0.959925|0.959925|
+|4|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.035016|0.997481|0.737537|0.848037|0.848037|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|6100906|124172|42908|53638397039|124172 / 53638521211|42908 / 6143814|
+|2|SNOWBALL ITALIAN DIRECT|PRIMARY_OUTPUT|4499650|504775|1644164|53638016436|504775 / 53638521211|1644164 / 6143814|
+|3|SNOWBALL ITALIAN LUCENE FILTER|PRIMARY_OUTPUT|4499650|504775|1644164|53638016436|504775 / 53638521211|1644164 / 6143814|
+|4|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|109684|10589|6034130|53638510622|10589 / 53638521211|6034130 / 6143814|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.999993|0 / 53638521211 (0.000000%)|80 / 6143814 (0.001302%)|0.999997|0.999993|0.999993|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0.999987|1.000000|0.999993|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.999997|0.999993|0.999990|0.999987|0.999993|0.999993|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|6143734|0|80|53638521211|0 / 53638521211|80 / 6143814|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999992|170950 / 53638521211 (0.000319%)|80 / 6143814 (0.001302%)|0.978222|0.986272|0.986363|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.972928|0.999987|0.999997|0.999992|0.999997|0.000003|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.978222|0.986272|0.994455|0.972916|0.986365|0.986363|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|6143734|170950|80|53638350261|170950 / 53638521211|80 / 6143814|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|42828|124172|46778|6254|1.909321%|4|334175|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **6 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996512|124171 / 53611667072 (0.000232%)|42828 / 6142174 (0.697278%)|0.982617|0.986495|0.986515|
+|2|SNOWBALL ITALIAN DIRECT|PRIMARY_OUTPUT|0.866205|504774 / 53611667072 (0.000942%)|1643522 / 6142174 (26.757985%)|0.859970|0.807252|0.811479|
+|3|SNOWBALL ITALIAN LUCENE FILTER|PRIMARY_OUTPUT|0.866205|504774 / 53611667072 (0.000942%)|1643522 / 6142174 (26.757985%)|0.859970|0.807252|0.811479|
+|4|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.508927|10588 / 53611667072 (0.000020%)|6032516 / 6142174 (98.214671%)|0.082784|0.035021|0.127589|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.980048|0.993027|0.999998|0.996512|0.999997|0.000003|
+|2|SNOWBALL ITALIAN DIRECT|PRIMARY_OUTPUT|0.899114|0.732420|0.999991|0.866205|0.999960|0.000040|
+|3|SNOWBALL ITALIAN LUCENE FILTER|PRIMARY_OUTPUT|0.899114|0.732420|0.999991|0.866205|0.999960|0.000040|
+|4|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.911947|0.017853|1.000000|0.508927|0.999887|0.000113|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.982617|0.986495|0.990404|0.973350|0.986516|0.986515|
+|2|SNOWBALL ITALIAN DIRECT|PRIMARY_OUTPUT|0.859970|0.807252|0.760624|0.676800|0.811498|0.811479|
+|3|SNOWBALL ITALIAN LUCENE FILTER|PRIMARY_OUTPUT|0.859970|0.807252|0.760624|0.676800|0.811498|0.811479|
+|4|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.082784|0.035021|0.022208|0.017823|0.127598|0.127589|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.986493|0.995780|0.997115|0.996447|0.996447|
+|2|SNOWBALL ITALIAN DIRECT|PRIMARY_OUTPUT|0.807232|0.987991|0.933413|0.959927|0.959927|
+|3|SNOWBALL ITALIAN LUCENE FILTER|PRIMARY_OUTPUT|0.807232|0.987991|0.933413|0.959927|0.959927|
+|4|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.035017|0.997481|0.737534|0.848035|0.848035|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|6099346|124171|42828|53611542901|124171 / 53611667072|42828 / 6142174|
+|2|SNOWBALL ITALIAN DIRECT|PRIMARY_OUTPUT|4498652|504774|1643522|53611162298|504774 / 53611667072|1643522 / 6142174|
+|3|SNOWBALL ITALIAN LUCENE FILTER|PRIMARY_OUTPUT|4498652|504774|1643522|53611162298|504774 / 53611667072|1643522 / 6142174|
+|4|ITALIAN LUCENE ITALIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|109658|10588|6032516|53611656484|10588 / 53611667072|6032516 / 6142174|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 53611667072 (0.000000%)|0 / 6142174 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|6142174|0|0|53611667072|0 / 53611667072|0 / 6142174|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999998|170949 / 53611667072 (0.000319%)|0 / 6142174 (0.000000%)|0.978219|0.986275|0.986366|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.972922|1.000000|0.999997|0.999998|0.999997|0.000003|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.978219|0.986275|0.994464|0.972922|0.986368|0.986366|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|6142174|170949|0|53611496123|170949 / 53611667072|0 / 6142174|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|42828|124171|46778|6252|1.909188%|4|334089|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `IT_IT`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/norwegian-bokmal.md

@@ -0,0 +1,420 @@
+# Norwegian Bokmal Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Norwegian Bokmal. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `NB_NO` | 17,929 | 90,757 | 33,376 | 57,381 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **90,757**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 1,500 | 1.653% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 4,296 | 4.734% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 47,619 | 52.469% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 34,420 | 37.925% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 2,922 | 3.220% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 96.852% | 97.637% | 95.503% | Full Radixor dictionary patch-command stemmer. |
+| Lucene NorwegianMinimalStemFilter | 57.107% | 53.913% | 62.599% | Minimal suffix reducer; narrow baseline, not a full stemmer. |
+| Official Snowball direct | 54.824% | 51.791% | 60.040% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+| Lucene SnowballFilter | 54.803% | 51.780% | 60.001% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Lucene NorwegianLightStemFilter | 52.136% | 50.616% | 54.749% | Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `norwegianBokmalRadixor` | 3.631 | 1.377 | 63.3 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene NorwegianMinimalStemFilter | `norwegianBokmalLuceneNorwegianMinimalStemFilter` | 2.910 | 0.177 | 50.7 | 0.801 | Minimal Norwegian suffix reducer. |
+| Lucene NorwegianLightStemFilter | `norwegianBokmalLuceneNorwegianLightStemFilter` | 3.335 | 0.116 | 58.1 | 0.919 | Light Norwegian suffix stemmer. |
+| Official Snowball direct | `snowballDirect[NORWEGIAN_BOKMAL]` | 4.277 | 0.082 | 74.5 | 1.178 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[NORWEGIAN_BOKMAL]` | 6.077 | 0.208 | 105.9 | 1.674 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `NB_NO` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/nb_no/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.974783** among 5 deterministic stemmers. The runner-up is `SNOWBALL NORWEGIAN BOKMAL DIRECT` at 0.874964, a difference of 0.099819. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.975000** among 5 deterministic stemmers. The runner-up is `SNOWBALL NORWEGIAN BOKMAL DIRECT` at 0.874991, a difference of 0.100009. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **7 result rows**, **5 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.974783|11482 / 2835618215 (0.000405%)|7170 / 142180 (5.042903%)|0.927078|0.935387|0.935488|
+|2|SNOWBALL NORWEGIAN BOKMAL DIRECT|PRIMARY_OUTPUT|0.874964|23997 / 2835618215 (0.000846%)|35554 / 142180 (25.006330%)|0.802095|0.781707|0.782399|
+|3|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|PRIMARY_OUTPUT|0.874834|24046 / 2835618215 (0.000848%)|35591 / 142180 (25.032353%)|0.801759|0.781401|0.782091|
+|4|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.850006|25171 / 2835618215 (0.000888%)|42651 / 142180 (29.997890%)|0.776381|0.745871|0.747464|
+|5|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.832414|14772 / 2835618215 (0.000521%)|47654 / 142180 (33.516669%)|0.815763|0.751764|0.758263|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.921620|0.949571|0.999996|0.974783|0.999993|0.000007|
+|2|SNOWBALL NORWEGIAN BOKMAL DIRECT|PRIMARY_OUTPUT|0.816288|0.749937|0.999992|0.874964|0.999979|0.000021|
+|3|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|PRIMARY_OUTPUT|0.815930|0.749676|0.999992|0.874834|0.999979|0.000021|
+|4|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.798148|0.700021|0.999991|0.850006|0.999976|0.000024|
+|5|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.864847|0.664833|0.999995|0.832414|0.999978|0.000022|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.927078|0.935387|0.943846|0.878617|0.935491|0.935488|
+|2|SNOWBALL NORWEGIAN BOKMAL DIRECT|PRIMARY_OUTPUT|0.802095|0.781707|0.762330|0.641641|0.782409|0.782399|
+|3|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|PRIMARY_OUTPUT|0.801759|0.781401|0.762052|0.641229|0.782102|0.782091|
+|4|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.776381|0.745871|0.717668|0.594732|0.747476|0.747464|
+|5|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.815763|0.751764|0.697076|0.602261|0.758274|0.758263|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.935384|0.993354|0.994615|0.993984|0.993984|
+|2|SNOWBALL NORWEGIAN BOKMAL DIRECT|PRIMARY_OUTPUT|0.781696|0.988328|0.971120|0.979648|0.979648|
+|3|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|PRIMARY_OUTPUT|0.781391|0.988295|0.971086|0.979615|0.979615|
+|4|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.745859|0.987774|0.965622|0.976573|0.976573|
+|5|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.751753|0.992089|0.962516|0.977079|0.977079|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|135010|11482|7170|2835606733|11482 / 2835618215|7170 / 142180|
+|2|SNOWBALL NORWEGIAN BOKMAL DIRECT|PRIMARY_OUTPUT|106626|23997|35554|2835594218|23997 / 2835618215|35554 / 142180|
+|3|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|PRIMARY_OUTPUT|106589|24046|35591|2835594169|24046 / 2835618215|35591 / 142180|
+|4|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|99529|25171|42651|2835593044|25171 / 2835618215|42651 / 142180|
+|5|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|94526|14772|47654|2835603443|14772 / 2835618215|47654 / 142180|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 2835618215 (0.000000%)|0 / 142180 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|142180|0|0|2835618215|0 / 2835618215|0 / 142180|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999996|20161 / 2835618215 (0.000711%)|0 / 142180 (0.000000%)|0.898118|0.933794|0.935844|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.875811|1.000000|0.999993|0.999996|0.999993|0.000007|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.898118|0.933794|0.972422|0.875811|0.935848|0.935844|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|142180|20161|0|2835598054|20161 / 2835618215|0 / 142180|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|7170|11482|8679|4237|5.626079%|9|79825|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **7 result rows**, **5 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.975000|11482 / 2831176784 (0.000406%)|7104 / 142091 (4.999613%)|0.927151|0.935591|0.935695|
+|2|SNOWBALL NORWEGIAN BOKMAL DIRECT|PRIMARY_OUTPUT|0.874991|23997 / 2831176784 (0.000848%)|35524 / 142091 (25.000880%)|0.802043|0.781698|0.782388|
+|3|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|PRIMARY_OUTPUT|0.874798|23993 / 2831176784 (0.000847%)|35579 / 142091 (25.039587%)|0.801914|0.781464|0.782161|
+|4|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.849947|25118 / 2831176784 (0.000887%)|42641 / 142091 (30.009642%)|0.776513|0.745896|0.747500|
+|5|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.832344|14719 / 2831176784 (0.000520%)|47644 / 142091 (33.530625%)|0.815950|0.751796|0.758325|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.921608|0.950004|0.999996|0.975000|0.999993|0.000007|
+|2|SNOWBALL NORWEGIAN BOKMAL DIRECT|PRIMARY_OUTPUT|0.816205|0.749991|0.999992|0.874991|0.999979|0.000021|
+|3|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|PRIMARY_OUTPUT|0.816153|0.749604|0.999992|0.874798|0.999979|0.000021|
+|4|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.798359|0.699904|0.999991|0.849947|0.999976|0.000024|
+|5|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.865169|0.664694|0.999995|0.832344|0.999978|0.000022|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.927151|0.935591|0.944186|0.878976|0.935698|0.935695|
+|2|SNOWBALL NORWEGIAN BOKMAL DIRECT|PRIMARY_OUTPUT|0.802043|0.781698|0.762360|0.641630|0.782398|0.782388|
+|3|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|PRIMARY_OUTPUT|0.801914|0.781464|0.762031|0.641314|0.782171|0.782161|
+|4|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.776513|0.745896|0.717603|0.594765|0.747512|0.747500|
+|5|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.815950|0.751796|0.696995|0.602302|0.758335|0.758325|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.935587|0.993348|0.994694|0.994020|0.994020|
+|2|SNOWBALL NORWEGIAN BOKMAL DIRECT|PRIMARY_OUTPUT|0.781688|0.988318|0.971127|0.979647|0.979647|
+|3|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|PRIMARY_OUTPUT|0.781454|0.988310|0.971074|0.979616|0.979616|
+|4|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.745885|0.987789|0.965603|0.976570|0.976570|
+|5|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.751785|0.992107|0.962494|0.977076|0.977076|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|134987|11482|7104|2831165302|11482 / 2831176784|7104 / 142091|
+|2|SNOWBALL NORWEGIAN BOKMAL DIRECT|PRIMARY_OUTPUT|106567|23997|35524|2831152787|23997 / 2831176784|35524 / 142091|
+|3|SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|PRIMARY_OUTPUT|106512|23993|35579|2831152791|23993 / 2831176784|35579 / 142091|
+|4|NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|99450|25118|42641|2831151666|25118 / 2831176784|42641 / 142091|
+|5|NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER|PRIMARY_OUTPUT|94447|14719|47644|2831162065|14719 / 2831176784|47644 / 142091|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 2831176784 (0.000000%)|0 / 142091 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|142091|0|0|2831176784|0 / 2831176784|0 / 142091|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999996|20161 / 2831176784 (0.000712%)|0 / 142091 (0.000000%)|0.898061|0.933756|0.935808|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.875743|1.000000|0.999993|0.999996|0.999993|0.000007|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.898061|0.933756|0.972405|0.875743|0.935811|0.935808|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|142091|20161|0|2831156623|20161 / 2831176784|0 / 142091|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|7104|11482|8679|4204|5.586637%|9|79733|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `NB_NO`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/norwegian-nynorsk.md

@@ -0,0 +1,396 @@
+# Norwegian Nynorsk Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Norwegian Nynorsk. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `NN_NO` | 4,688 | 19,651 | 6,089 | 13,562 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **19,651**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 224 | 1.140% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 1,505 | 7.659% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 11,017 | 56.063% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 6,427 | 32.706% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 478 | 2.432% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 93.089% | 91.395% | 96.863% | Full Radixor dictionary patch-command stemmer. |
+| Official Snowball direct | 60.974% | 60.212% | 62.670% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+| Lucene SnowballFilter | 60.918% | 60.146% | 62.638% | Lucene TokenFilter integration path around the Snowball algorithm. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `radixor[NORWEGIAN_NYNORSK]` | 0.571 | 0.012 | 42.1 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Official Snowball direct | `snowballDirect[NORWEGIAN_NYNORSK]` | 0.919 | 0.038 | 67.7 | 1.609 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[NORWEGIAN_NYNORSK]` | 1.309 | 0.015 | 96.5 | 2.292 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `NN_NO` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/nn_no/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.935777** among 3 deterministic stemmers. The runner-up is `SNOWBALL NORWEGIAN NYNORSK DIRECT` at 0.858908, a difference of 0.076869. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.935853** among 3 deterministic stemmers. The runner-up is `SNOWBALL NORWEGIAN NYNORSK DIRECT` at 0.859037, a difference of 0.076816. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **5 result rows**, **3 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.935777|6230 / 166491473 (0.003742%)|3936 / 30652 (12.840924%)|0.822355|0.840152|0.840669|
+|2|SNOWBALL NORWEGIAN NYNORSK DIRECT|PRIMARY_OUTPUT|0.858908|8274 / 166491473 (0.004970%)|8648 / 30652 (28.213493%)|0.724941|0.722271|0.722234|
+|3|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|PRIMARY_OUTPUT|0.858484|8295 / 166491473 (0.004982%)|8674 / 30652 (28.298317%)|0.724180|0.721477|0.721440|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.810903|0.871591|0.999963|0.935777|0.999939|0.000061|
+|2|SNOWBALL NORWEGIAN NYNORSK DIRECT|PRIMARY_OUTPUT|0.726732|0.717865|0.999950|0.858908|0.999898|0.000102|
+|3|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|PRIMARY_OUTPUT|0.725993|0.717017|0.999950|0.858484|0.999898|0.000102|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.822355|0.840152|0.858737|0.724364|0.840699|0.840669|
+|2|SNOWBALL NORWEGIAN NYNORSK DIRECT|PRIMARY_OUTPUT|0.724941|0.722271|0.719621|0.565278|0.722285|0.722234|
+|3|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|PRIMARY_OUTPUT|0.724180|0.721477|0.718794|0.564305|0.721491|0.721440|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.840122|0.983845|0.986802|0.985321|0.985321|
+|2|SNOWBALL NORWEGIAN NYNORSK DIRECT|PRIMARY_OUTPUT|0.722221|0.980542|0.964998|0.972708|0.972708|
+|3|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|PRIMARY_OUTPUT|0.721426|0.980461|0.964862|0.972599|0.972599|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|26716|6230|3936|166485243|6230 / 166491473|3936 / 30652|
+|2|SNOWBALL NORWEGIAN NYNORSK DIRECT|PRIMARY_OUTPUT|22004|8274|8648|166483199|8274 / 166491473|8648 / 30652|
+|3|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|PRIMARY_OUTPUT|21978|8295|8674|166483178|8295 / 166491473|8674 / 30652|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 166491473 (0.000000%)|0 / 30652 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|30652|0|0|166491473|0 / 166491473|0 / 30652|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999960|13214 / 166491473 (0.007937%)|0 / 30652 (0.000000%)|0.743562|0.822674|0.835888|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.698764|1.000000|0.999921|0.999960|0.999921|0.000079|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.743562|0.822674|0.920624|0.698764|0.835921|0.835888|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|30652|13214|0|166478259|13214 / 166491473|0 / 30652|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|3936|6230|6984|2404|13.172603%|5|21513|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **5 result rows**, **3 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.935853|6230 / 165926276 (0.003755%)|3924 / 30595 (12.825625%)|0.822169|0.840084|0.840609|
+|2|SNOWBALL NORWEGIAN NYNORSK DIRECT|PRIMARY_OUTPUT|0.859037|8274 / 165926276 (0.004987%)|8624 / 30595 (28.187612%)|0.724757|0.722255|0.722216|
+|3|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|PRIMARY_OUTPUT|0.858661|8274 / 165926276 (0.004987%)|8647 / 30595 (28.262788%)|0.724438|0.721772|0.721734|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.810644|0.871744|0.999962|0.935853|0.999939|0.000061|
+|2|SNOWBALL NORWEGIAN NYNORSK DIRECT|PRIMARY_OUTPUT|0.726434|0.718124|0.999950|0.859037|0.999898|0.000102|
+|3|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|PRIMARY_OUTPUT|0.726226|0.717372|0.999950|0.858661|0.999898|0.000102|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.822169|0.840084|0.858798|0.724263|0.840639|0.840609|
+|2|SNOWBALL NORWEGIAN NYNORSK DIRECT|PRIMARY_OUTPUT|0.724757|0.722255|0.719771|0.565258|0.722267|0.722216|
+|3|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|PRIMARY_OUTPUT|0.724438|0.721772|0.719126|0.564666|0.721785|0.721734|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.840054|0.983815|0.986842|0.985326|0.985326|
+|2|SNOWBALL NORWEGIAN NYNORSK DIRECT|PRIMARY_OUTPUT|0.722204|0.980506|0.965065|0.972724|0.972724|
+|3|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|PRIMARY_OUTPUT|0.721721|0.980506|0.964945|0.972663|0.972663|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|26671|6230|3924|165920046|6230 / 165926276|3924 / 30595|
+|2|SNOWBALL NORWEGIAN NYNORSK DIRECT|PRIMARY_OUTPUT|21971|8274|8624|165918002|8274 / 165926276|8624 / 30595|
+|3|SNOWBALL NORWEGIAN NYNORSK LUCENE FILTER|PRIMARY_OUTPUT|21948|8274|8647|165918002|8274 / 165926276|8647 / 30595|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 165926276 (0.000000%)|0 / 30595 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|30595|0|0|165926276|0 / 165926276|0 / 30595|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999960|13214 / 165926276 (0.007964%)|0 / 30595 (0.000000%)|0.743207|0.822402|0.835654|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.698372|1.000000|0.999920|0.999960|0.999920|0.000080|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.743207|0.822402|0.920488|0.698372|0.835687|0.835654|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|30595|13214|0|165913062|13214 / 165926276|0 / 30595|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|3924|6230|6984|2399|13.167572%|5|21477|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `NN_NO`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/persian.md

@@ -0,0 +1,382 @@
+# Persian Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Persian. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `FA_IR` | 69 | 3,770 | 138 | 3,632 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **3,770**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `DeletePrefixCommand` | Deletes one or more leading characters from the word form in forward traversal. | 65 | 1.724% |
+| `ForwardCompoundCommand` | Applies a multi-step forward patch made from skip, delete, insert, and replace operations. | 3,567 | 94.615% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 138 | 3.660% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 95.836% | 95.677% | 100.000% | Full Radixor dictionary patch-command stemmer. |
+| Lucene PersianStemFilter | 1.485% | 0.000% | 40.580% | Lucene Persian suffix stemmer with required normalization in the measured path. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `persianRadixor` | 0.245 | 0.025 | 49.0 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene PersianStemFilter | `persianLucenePersianStemFilter` | 0.466 | 0.015 | 93.3 | 1.902 | Persian suffix stemmer with Lucene normalization in the measured path. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `FA_IR` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/fa_ir/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.974922** among 2 deterministic stemmers. The runner-up is `PERSIAN LUCENE PERSIAN STEM FILTER` at 0.502171, a difference of 0.472751. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.974922** among 2 deterministic stemmers. The runner-up is `PERSIAN LUCENE PERSIAN STEM FILTER` at 0.502171, a difference of 0.472751. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **4 result rows**, **2 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.974922|8621 / 6748402 (0.127749%)|4812 / 98448 (4.887860%)|0.922566|0.933071|0.932249|
+|2|PERSIAN LUCENE PERSIAN STEM FILTER|PRIMARY_OUTPUT|0.502171|179 / 6748402 (0.002652%)|98018 / 98448 (99.563221%)|0.021312|0.008682|0.054801|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.915693|0.951121|0.998723|0.974922|0.998038|0.001962|
+|2|PERSIAN LUCENE PERSIAN STEM FILTER|PRIMARY_OUTPUT|0.706076|0.004368|0.999973|0.502171|0.985658|0.014342|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.922566|0.933071|0.943818|0.874539|0.933239|0.932249|
+|2|PERSIAN LUCENE PERSIAN STEM FILTER|PRIMARY_OUTPUT|0.021312|0.008682|0.005451|0.004360|0.055534|0.054801|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.932076|0.980343|0.984586|0.982460|0.982460|
+|2|PERSIAN LUCENE PERSIAN STEM FILTER|PRIMARY_OUTPUT|0.008507|0.985686|0.520347|0.681125|0.681125|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|93636|8621|4812|6739781|8621 / 6748402|4812 / 98448|
+|2|PERSIAN LUCENE PERSIAN STEM FILTER|PRIMARY_OUTPUT|430|179|98018|6748223|179 / 6748402|98018 / 98448|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 6748402 (0.000000%)|0 / 98448 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|98448|0|0|6748402|0 / 6748402|0 / 98448|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999005|13433 / 6748402 (0.199055%)|0 / 98448 (0.000000%)|0.901585|0.936133|0.937114|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.879935|1.000000|0.998009|0.999005|0.998038|0.001962|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.901585|0.936133|0.973435|0.879935|0.938048|0.937114|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|98448|13433|0|6734969|13433 / 6748402|0 / 98448|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|4812|8621|4812|314|8.484193%|2|4015|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **4 result rows**, **2 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.974922|8621 / 6748402 (0.127749%)|4812 / 98448 (4.887860%)|0.922566|0.933071|0.932249|
+|2|PERSIAN LUCENE PERSIAN STEM FILTER|PRIMARY_OUTPUT|0.502171|179 / 6748402 (0.002652%)|98018 / 98448 (99.563221%)|0.021312|0.008682|0.054801|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.915693|0.951121|0.998723|0.974922|0.998038|0.001962|
+|2|PERSIAN LUCENE PERSIAN STEM FILTER|PRIMARY_OUTPUT|0.706076|0.004368|0.999973|0.502171|0.985658|0.014342|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.922566|0.933071|0.943818|0.874539|0.933239|0.932249|
+|2|PERSIAN LUCENE PERSIAN STEM FILTER|PRIMARY_OUTPUT|0.021312|0.008682|0.005451|0.004360|0.055534|0.054801|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.932076|0.980343|0.984586|0.982460|0.982460|
+|2|PERSIAN LUCENE PERSIAN STEM FILTER|PRIMARY_OUTPUT|0.008507|0.985686|0.520347|0.681125|0.681125|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|93636|8621|4812|6739781|8621 / 6748402|4812 / 98448|
+|2|PERSIAN LUCENE PERSIAN STEM FILTER|PRIMARY_OUTPUT|430|179|98018|6748223|179 / 6748402|98018 / 98448|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 6748402 (0.000000%)|0 / 98448 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|98448|0|0|6748402|0 / 6748402|0 / 98448|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999005|13433 / 6748402 (0.199055%)|0 / 98448 (0.000000%)|0.901585|0.936133|0.937114|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.879935|1.000000|0.998009|0.999005|0.998038|0.001962|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.901585|0.936133|0.973435|0.879935|0.938048|0.937114|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|98448|13433|0|6734969|13433 / 6748402|0 / 98448|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|4812|8621|4812|314|8.484193%|2|4015|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `FA_IR`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/polish.md

@@ -0,0 +1,464 @@
+# Polish Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Polish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `PL_PL` | 9,990 | 132,308 | 19,957 | 112,351 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **132,308**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 1,719 | 1.299% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 53,303 | 40.287% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 37,051 | 28.004% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 20,415 | 15.430% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 19,820 | 14.980% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 98.837% | 98.744% | 99.359% | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | 89.545% | 88.272% | 96.713% | Benchmark-only Polish Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene MorfologikFilter | 87.729% | 86.606% | 94.047% | Dictionary-based path; Morfologik can emit multiple terms. |
+| Lucene StempelFilter | 70.009% | 69.262% | 74.220% | Lucene TokenFilter integration path for table-driven Polish Stempel. |
+| Lucene StempelStemmer direct | 70.009% | 69.262% | 74.220% | Direct table-driven Polish Stempel stemmer API. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `polishRadixor` | 9.049 | 0.485 | 80.5 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 483.316 | 11.455 | 4301.8 | 53.408 | Benchmark-only Polish Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene StempelStemmer direct | `polishLuceneStempelStemmerDirect` | 41.932 | 1.916 | 373.2 | 4.634 | Direct table-driven Polish Stempel stemmer API. |
+| Lucene StempelFilter | `polishLuceneStempelFilter` | 45.277 | 13.693 | 403.0 | 5.003 | Lucene TokenFilter integration path for table-driven Polish Stempel. |
+| Lucene MorfologikFilter | `polishLuceneMorfologikFilter` | 135.763 | 31.634 | 1208.4 | 15.002 | Dictionary-based Morfologik TokenFilter; may emit multiple terms. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `PL_PL` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/pl_pl/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.990388** among 5 deterministic stemmers. The runner-up is `POLISH LUCENE MORFOLOGIK FILTER` at 0.948154, a difference of 0.042234. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.990579** among 5 deterministic stemmers. The runner-up is `POLISH LUCENE MORFOLOGIK FILTER` at 0.948177, a difference of 0.042402. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **11 result rows**, **5 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.990388|13669 / 7482478003 (0.000183%)|21547 / 1120967 (1.922180%)|0.986324|0.984237|0.984241|
+|2|POLISH LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.948154|99228 / 7482478003 (0.001326%)|116220 / 1120967 (10.367834%)|0.907324|0.903167|0.903179|
+|3|HUNSPELL POLISH LUCENE FILTER|PRIMARY_OUTPUT|0.933222|52652 / 7482478003 (0.000704%)|149705 / 1120967 (13.354987%)|0.930930|0.905656|0.906571|
+|4|POLISH LUCENE STEMPEL DIRECT|PRIMARY_OUTPUT|0.855748|66669 / 7482478003 (0.000891%)|323394 / 1120967 (28.849556%)|0.871106|0.803515|0.810296|
+|5|POLISH LUCENE STEMPEL FILTER|PRIMARY_OUTPUT|0.855748|66669 / 7482478003 (0.000891%)|323394 / 1120967 (28.849556%)|0.871106|0.803515|0.810296|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987720|0.980778|0.999998|0.990388|0.999995|0.000005|
+|2|POLISH LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.910118|0.896322|0.999987|0.948154|0.999971|0.000029|
+|3|HUNSPELL POLISH LUCENE FILTER|PRIMARY_OUTPUT|0.948578|0.866450|0.999993|0.933222|0.999973|0.000027|
+|4|POLISH LUCENE STEMPEL DIRECT|PRIMARY_OUTPUT|0.922858|0.711504|0.999991|0.855748|0.999948|0.000052|
+|5|POLISH LUCENE STEMPEL FILTER|PRIMARY_OUTPUT|0.922858|0.711504|0.999991|0.855748|0.999948|0.000052|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.986324|0.984237|0.982159|0.968963|0.984243|0.984241|
+|2|POLISH LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.907324|0.903167|0.899047|0.823432|0.903193|0.903179|
+|3|HUNSPELL POLISH LUCENE FILTER|PRIMARY_OUTPUT|0.930930|0.905656|0.881718|0.827579|0.906584|0.906571|
+|4|POLISH LUCENE STEMPEL DIRECT|PRIMARY_OUTPUT|0.871106|0.803515|0.745659|0.671564|0.810320|0.810296|
+|5|POLISH LUCENE STEMPEL FILTER|PRIMARY_OUTPUT|0.871106|0.803515|0.745659|0.671564|0.810320|0.810296|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.984234|0.996967|0.996469|0.996718|0.996718|
+|2|POLISH LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.903153|0.990022|0.977054|0.983495|0.983495|
+|3|HUNSPELL POLISH LUCENE FILTER|PRIMARY_OUTPUT|0.905642|0.994546|0.970520|0.982386|0.982386|
+|4|POLISH LUCENE STEMPEL DIRECT|PRIMARY_OUTPUT|0.803490|0.991767|0.931069|0.960460|0.960460|
+|5|POLISH LUCENE STEMPEL FILTER|PRIMARY_OUTPUT|0.803490|0.991767|0.931069|0.960460|0.960460|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|1099420|13669|21547|7482464334|13669 / 7482478003|21547 / 1120967|
+|2|POLISH LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|1004747|99228|116220|7482378775|99228 / 7482478003|116220 / 1120967|
+|3|HUNSPELL POLISH LUCENE FILTER|PRIMARY_OUTPUT|971262|52652|149705|7482425351|52652 / 7482478003|149705 / 1120967|
+|4|POLISH LUCENE STEMPEL DIRECT|PRIMARY_OUTPUT|797573|66669|323394|7482411334|66669 / 7482478003|323394 / 1120967|
+|5|POLISH LUCENE STEMPEL FILTER|PRIMARY_OUTPUT|797573|66669|323394|7482411334|66669 / 7482478003|323394 / 1120967|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 7482478003 (0.000000%)|0 / 1120967 (0.000000%)|1.000000|1.000000|1.000000|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.987570|85532 / 7482478003 (0.001143%)|27855 / 1120967 (2.484908%)|0.936598|0.950693|0.950985|
+|3|HUNSPELL POLISH LUCENE FILTER|ANY_CANDIDATE|0.963982|42213 / 7482478003 (0.000564%)|80743 / 1120967 (7.202977%)|0.954209|0.944197|0.944333|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.927432|0.975151|0.999989|0.987570|0.999985|0.000015|
+|3|HUNSPELL POLISH LUCENE FILTER|ANY_CANDIDATE|0.961002|0.927970|0.999994|0.963982|0.999984|0.000016|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.936598|0.950693|0.965218|0.906020|0.950992|0.950985|
+|3|HUNSPELL POLISH LUCENE FILTER|ANY_CANDIDATE|0.954209|0.944197|0.934394|0.894293|0.944342|0.944333|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|3|HUNSPELL POLISH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1120967|0|0|7482478003|0 / 7482478003|0 / 1120967|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|1093112|85532|27855|7482392471|85532 / 7482478003|27855 / 1120967|
+|3|HUNSPELL POLISH LUCENE FILTER|ANY_CANDIDATE|1040224|42213|80743|7482435790|42213 / 7482478003|80743 / 1120967|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999997|38073 / 7482478003 (0.000509%)|0 / 1120967 (0.000000%)|0.973547|0.983301|0.983436|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.987566|143096 / 7482478003 (0.001912%)|27855 / 1120967 (2.484908%)|0.901045|0.927476|0.928576|
+|3|HUNSPELL POLISH LUCENE FILTER|ALL_CANDIDATES|0.963980|82745 / 7482478003 (0.001106%)|80743 / 1120967 (7.202977%)|0.926646|0.927142|0.927132|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.967151|1.000000|0.999995|0.999997|0.999995|0.000005|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.884246|0.975151|0.999981|0.987566|0.999977|0.000023|
+|3|HUNSPELL POLISH LUCENE FILTER|ALL_CANDIDATES|0.926316|0.927970|0.999989|0.963980|0.999978|0.000022|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.973547|0.983301|0.993253|0.967151|0.983438|0.983436|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.901045|0.927476|0.955505|0.864761|0.928587|0.928576|
+|3|HUNSPELL POLISH LUCENE FILTER|ALL_CANDIDATES|0.926646|0.927142|0.927639|0.864180|0.927143|0.927132|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|3|HUNSPELL POLISH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|1120967|38073|0|7482439930|38073 / 7482478003|0 / 1120967|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|1093112|143096|27855|7482334907|143096 / 7482478003|27855 / 1120967|
+|3|HUNSPELL POLISH LUCENE FILTER|ALL_CANDIDATES|1040224|82745|80743|7482395258|82745 / 7482478003|80743 / 1120967|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|HUNSPELL POLISH LUCENE FILTER|68962|10439|30093|11447|9.356634%|6|135231|
+|POLISH LUCENE MORFOLOGIK FILTER|88365|13696|43868|12873|10.522229%|5|136636|
+|Radixor|21547|13669|24404|2866|2.342632%|4|125778|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **11 result rows**, **5 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.990579|13669 / 7310252699 (0.000187%)|21000 / 1114651 (1.883998%)|0.986350|0.984397|0.984400|
+|2|POLISH LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.948177|99224 / 7310252699 (0.001357%)|115513 / 1114651 (10.363154%)|0.906972|0.902966|0.902976|
+|3|HUNSPELL POLISH LUCENE FILTER|PRIMARY_OUTPUT|0.933309|51950 / 7310252699 (0.000711%)|148667 / 1114651 (13.337538%)|0.931269|0.905928|0.906847|
+|4|POLISH LUCENE STEMPEL DIRECT|PRIMARY_OUTPUT|0.856382|66274 / 7310252699 (0.000907%)|320158 / 1114651 (28.722712%)|0.871591|0.804380|0.811082|
+|5|POLISH LUCENE STEMPEL FILTER|PRIMARY_OUTPUT|0.856382|66274 / 7310252699 (0.000907%)|320158 / 1114651 (28.722712%)|0.871591|0.804380|0.811082|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987656|0.981160|0.999998|0.990579|0.999995|0.000005|
+|2|POLISH LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.909662|0.896368|0.999986|0.948177|0.999971|0.000029|
+|3|HUNSPELL POLISH LUCENE FILTER|PRIMARY_OUTPUT|0.948965|0.866625|0.999993|0.933309|0.999973|0.000027|
+|4|POLISH LUCENE STEMPEL DIRECT|PRIMARY_OUTPUT|0.923006|0.712773|0.999991|0.856382|0.999947|0.000053|
+|5|POLISH LUCENE STEMPEL FILTER|PRIMARY_OUTPUT|0.923006|0.712773|0.999991|0.856382|0.999947|0.000053|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.986350|0.984397|0.982452|0.969274|0.984403|0.984400|
+|2|POLISH LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.906972|0.902966|0.898996|0.823098|0.902991|0.902976|
+|3|HUNSPELL POLISH LUCENE FILTER|PRIMARY_OUTPUT|0.931269|0.905928|0.881929|0.828033|0.906861|0.906847|
+|4|POLISH LUCENE STEMPEL DIRECT|PRIMARY_OUTPUT|0.871591|0.804380|0.746792|0.672772|0.811106|0.811082|
+|5|POLISH LUCENE STEMPEL FILTER|PRIMARY_OUTPUT|0.871591|0.804380|0.746792|0.672772|0.811106|0.811082|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.984395|0.996926|0.996647|0.996786|0.996786|
+|2|POLISH LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.902952|0.989889|0.977012|0.983408|0.983408|
+|3|HUNSPELL POLISH LUCENE FILTER|PRIMARY_OUTPUT|0.905914|0.994584|0.970514|0.982402|0.982402|
+|4|POLISH LUCENE STEMPEL DIRECT|PRIMARY_OUTPUT|0.804354|0.991711|0.931318|0.960566|0.960566|
+|5|POLISH LUCENE STEMPEL FILTER|PRIMARY_OUTPUT|0.804354|0.991711|0.931318|0.960566|0.960566|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|1093651|13669|21000|7310239030|13669 / 7310252699|21000 / 1114651|
+|2|POLISH LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|999138|99224|115513|7310153475|99224 / 7310252699|115513 / 1114651|
+|3|HUNSPELL POLISH LUCENE FILTER|PRIMARY_OUTPUT|965984|51950|148667|7310200749|51950 / 7310252699|148667 / 1114651|
+|4|POLISH LUCENE STEMPEL DIRECT|PRIMARY_OUTPUT|794493|66274|320158|7310186425|66274 / 7310252699|320158 / 1114651|
+|5|POLISH LUCENE STEMPEL FILTER|PRIMARY_OUTPUT|794493|66274|320158|7310186425|66274 / 7310252699|320158 / 1114651|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 7310252699 (0.000000%)|0 / 1114651 (0.000000%)|1.000000|1.000000|1.000000|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.987661|85532 / 7310252699 (0.001170%)|27494 / 1114651 (2.466602%)|0.936331|0.950586|0.950885|
+|3|HUNSPELL POLISH LUCENE FILTER|ANY_CANDIDATE|0.963946|41671 / 7310252699 (0.000570%)|80368 / 1114651 (7.210149%)|0.954406|0.944290|0.944429|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.927063|0.975334|0.999988|0.987661|0.999985|0.000015|
+|3|HUNSPELL POLISH LUCENE FILTER|ANY_CANDIDATE|0.961271|0.927899|0.999994|0.963946|0.999983|0.000017|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.936331|0.950586|0.965282|0.905826|0.950892|0.950885|
+|3|HUNSPELL POLISH LUCENE FILTER|ANY_CANDIDATE|0.954406|0.944290|0.934386|0.894459|0.944437|0.944429|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|3|HUNSPELL POLISH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1114651|0|0|7310252699|0 / 7310252699|0 / 1114651|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|1087157|85532|27494|7310167167|85532 / 7310252699|27494 / 1114651|
+|3|HUNSPELL POLISH LUCENE FILTER|ANY_CANDIDATE|1034283|41671|80368|7310211028|41671 / 7310252699|80368 / 1114651|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999997|38073 / 7310252699 (0.000521%)|0 / 1114651 (0.000000%)|0.973401|0.983208|0.983344|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.987657|143085 / 7310252699 (0.001957%)|27494 / 1114651 (2.466602%)|0.900618|0.927255|0.928372|
+|3|HUNSPELL POLISH LUCENE FILTER|ALL_CANDIDATES|0.963944|81865 / 7310252699 (0.001120%)|80368 / 1114651 (7.210149%)|0.926903|0.927276|0.927265|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.966971|1.000000|0.999995|0.999997|0.999995|0.000005|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.883694|0.975334|0.999980|0.987657|0.999977|0.000023|
+|3|HUNSPELL POLISH LUCENE FILTER|ALL_CANDIDATES|0.926654|0.927899|0.999989|0.963944|0.999978|0.000022|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.973401|0.983208|0.993215|0.966971|0.983347|0.983344|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.900618|0.927255|0.955516|0.864376|0.928384|0.928372|
+|3|HUNSPELL POLISH LUCENE FILTER|ALL_CANDIDATES|0.926903|0.927276|0.927649|0.864412|0.927276|0.927265|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|3|HUNSPELL POLISH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|1114651|38073|0|7310214626|38073 / 7310252699|0 / 1114651|
+|2|POLISH LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|1087157|143085|27494|7310109614|143085 / 7310252699|27494 / 1114651|
+|3|HUNSPELL POLISH LUCENE FILTER|ALL_CANDIDATES|1034283|81865|80368|7310170834|81865 / 7310252699|80368 / 1114651|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|HUNSPELL POLISH LUCENE FILTER|68299|10279|29915|11265|9.315692%|6|133595|
+|POLISH LUCENE MORFOLOGIK FILTER|88019|13692|43861|12763|10.554476%|5|135105|
+|Radixor|21000|13669|24404|2780|2.298946%|4|124274|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `PL_PL`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/portuguese.md

@@ -0,0 +1,432 @@
+# Portuguese Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Portuguese. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `PT_PT` | 4,001 | 215,490 | 8,002 | 207,488 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **215,490**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 3,806 | 1.766% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 120,691 | 56.008% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 71,284 | 33.080% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 8,003 | 3.714% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 11,706 | 5.432% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 99.815% | 99.808% | 100.000% | Full Radixor dictionary patch-command stemmer. |
+| Lucene PortugueseLightStemFilter | 8.966% | 5.558% | 97.326% | Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer. |
+| Lucene PortugueseMinimalStemFilter | 5.539% | 1.896% | 100.000% | Minimal suffix reducer; narrow baseline, not a full stemmer. |
+| Lucene SnowballFilter | 0.625% | 0.558% | 2.374% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Official Snowball direct | 0.625% | 0.558% | 2.374% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+| Lucene PortugueseStemFilter | 0.312% | 0.308% | 0.425% | Portuguese RSLP-style Lucene TokenFilter stemmer. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `portugueseRadixor` | 12.109 | 0.698 | 58.4 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene PortugueseLightStemFilter | `portugueseLucenePortugueseLightStemFilter` | 11.172 | 1.870 | 53.8 | 0.923 | Light Portuguese suffix stemmer. |
+| Lucene PortugueseMinimalStemFilter | `portugueseLucenePortugueseMinimalStemFilter` | 16.038 | 1.752 | 77.3 | 1.325 | Minimal Portuguese suffix reducer. |
+| Official Snowball direct | `snowballDirect[PORTUGUESE]` | 53.725 | 5.356 | 258.9 | 4.437 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[PORTUGUESE]` | 57.457 | 1.182 | 276.9 | 4.745 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Lucene PortugueseStemFilter | `portugueseLucenePortugueseStemFilter` | 165.447 | 40.334 | 797.4 | 13.663 | Portuguese RSLP-style Lucene TokenFilter. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `PT_PT` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/pt_pt/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.998502** among 6 deterministic stemmers. The runner-up is `SNOWBALL PORTUGUESE DIRECT` at 0.938800, a difference of 0.059702. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.998502** among 6 deterministic stemmers. The runner-up is `SNOWBALL PORTUGUESE DIRECT` at 0.938800, a difference of 0.059702. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **8 result rows**, **6 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.998502|20678 / 22358203756 (0.000092%)|16444 / 5489060 (0.299578%)|0.996389|0.996620|0.996619|
+|2|SNOWBALL PORTUGUESE DIRECT|PRIMARY_OUTPUT|0.938800|167230 / 22358203756 (0.000748%)|671821 / 5489060 (12.239272%)|0.947271|0.919888|0.920940|
+|3|SNOWBALL PORTUGUESE LUCENE FILTER|PRIMARY_OUTPUT|0.938800|167230 / 22358203756 (0.000748%)|671821 / 5489060 (12.239272%)|0.947271|0.919888|0.920940|
+|4|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|PRIMARY_OUTPUT|0.846459|99075 / 22358203756 (0.000443%)|1685572 / 5489060 (30.707844%)|0.901330|0.809975|0.821750|
+|5|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|PRIMARY_OUTPUT|0.513648|2511 / 22358203756 (0.000011%)|5339230 / 5489060 (97.270389%)|0.122843|0.053118|0.163828|
+|6|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.503954|598 / 22358203756 (0.000003%)|5445654 / 5489060 (99.209227%)|0.038310|0.015690|0.088308|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996236|0.997004|0.999999|0.998502|0.999998|0.000002|
+|2|SNOWBALL PORTUGUESE DIRECT|PRIMARY_OUTPUT|0.966450|0.877607|0.999993|0.938800|0.999962|0.000038|
+|3|SNOWBALL PORTUGUESE LUCENE FILTER|PRIMARY_OUTPUT|0.966450|0.877607|0.999993|0.938800|0.999962|0.000038|
+|4|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|PRIMARY_OUTPUT|0.974613|0.692922|0.999996|0.846459|0.999920|0.000080|
+|5|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|PRIMARY_OUTPUT|0.983517|0.027296|1.000000|0.513648|0.999761|0.000239|
+|6|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.986410|0.007908|1.000000|0.503954|0.999756|0.000244|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996389|0.996620|0.996850|0.993262|0.996620|0.996619|
+|2|SNOWBALL PORTUGUESE DIRECT|PRIMARY_OUTPUT|0.947271|0.919888|0.894045|0.851661|0.920958|0.920940|
+|3|SNOWBALL PORTUGUESE LUCENE FILTER|PRIMARY_OUTPUT|0.947271|0.919888|0.894045|0.851661|0.920958|0.920940|
+|4|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|PRIMARY_OUTPUT|0.901330|0.809975|0.735434|0.680636|0.821785|0.821750|
+|5|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|PRIMARY_OUTPUT|0.122843|0.053118|0.033885|0.027284|0.163848|0.163828|
+|6|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.038310|0.015690|0.009865|0.007907|0.088319|0.088308|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996619|0.999299|0.999347|0.999323|0.999323|
+|2|SNOWBALL PORTUGUESE DIRECT|PRIMARY_OUTPUT|0.919870|0.996663|0.967924|0.982083|0.982083|
+|3|SNOWBALL PORTUGUESE LUCENE FILTER|PRIMARY_OUTPUT|0.919870|0.996663|0.967924|0.982083|0.982083|
+|4|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|PRIMARY_OUTPUT|0.809936|0.996729|0.918475|0.956003|0.956003|
+|5|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|PRIMARY_OUTPUT|0.053105|0.999226|0.720580|0.837330|0.837330|
+|6|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.015686|0.999664|0.692383|0.818122|0.818122|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|5472616|20678|16444|22358183078|20678 / 22358203756|16444 / 5489060|
+|2|SNOWBALL PORTUGUESE DIRECT|PRIMARY_OUTPUT|4817239|167230|671821|22358036526|167230 / 22358203756|671821 / 5489060|
+|3|SNOWBALL PORTUGUESE LUCENE FILTER|PRIMARY_OUTPUT|4817239|167230|671821|22358036526|167230 / 22358203756|671821 / 5489060|
+|4|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|PRIMARY_OUTPUT|3803488|99075|1685572|22358104681|99075 / 22358203756|1685572 / 5489060|
+|5|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|PRIMARY_OUTPUT|149830|2511|5339230|22358201245|2511 / 22358203756|5339230 / 5489060|
+|6|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|PRIMARY_OUTPUT|43406|598|5445654|22358203158|598 / 22358203756|5445654 / 5489060|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 22358203756 (0.000000%)|0 / 5489060 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|5489060|0|0|22358203756|0 / 22358203756|0 / 5489060|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999999|38310 / 22358203756 (0.000171%)|0 / 5489060 (0.000000%)|0.994448|0.996522|0.996528|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.993069|1.000000|0.999998|0.999999|0.999998|0.000002|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.994448|0.996522|0.998606|0.993069|0.996528|0.996528|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|5489060|38310|0|22358165446|38310 / 22358203756|0 / 5489060|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|16444|20678|17632|790|0.373542%|3|212297|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **8 result rows**, **6 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.998502|20678 / 22358203756 (0.000092%)|16444 / 5489060 (0.299578%)|0.996389|0.996620|0.996619|
+|2|SNOWBALL PORTUGUESE DIRECT|PRIMARY_OUTPUT|0.938800|167230 / 22358203756 (0.000748%)|671821 / 5489060 (12.239272%)|0.947271|0.919888|0.920940|
+|3|SNOWBALL PORTUGUESE LUCENE FILTER|PRIMARY_OUTPUT|0.938800|167230 / 22358203756 (0.000748%)|671821 / 5489060 (12.239272%)|0.947271|0.919888|0.920940|
+|4|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|PRIMARY_OUTPUT|0.846459|99075 / 22358203756 (0.000443%)|1685572 / 5489060 (30.707844%)|0.901330|0.809975|0.821750|
+|5|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|PRIMARY_OUTPUT|0.513648|2511 / 22358203756 (0.000011%)|5339230 / 5489060 (97.270389%)|0.122843|0.053118|0.163828|
+|6|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.503954|598 / 22358203756 (0.000003%)|5445654 / 5489060 (99.209227%)|0.038310|0.015690|0.088308|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996236|0.997004|0.999999|0.998502|0.999998|0.000002|
+|2|SNOWBALL PORTUGUESE DIRECT|PRIMARY_OUTPUT|0.966450|0.877607|0.999993|0.938800|0.999962|0.000038|
+|3|SNOWBALL PORTUGUESE LUCENE FILTER|PRIMARY_OUTPUT|0.966450|0.877607|0.999993|0.938800|0.999962|0.000038|
+|4|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|PRIMARY_OUTPUT|0.974613|0.692922|0.999996|0.846459|0.999920|0.000080|
+|5|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|PRIMARY_OUTPUT|0.983517|0.027296|1.000000|0.513648|0.999761|0.000239|
+|6|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.986410|0.007908|1.000000|0.503954|0.999756|0.000244|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996389|0.996620|0.996850|0.993262|0.996620|0.996619|
+|2|SNOWBALL PORTUGUESE DIRECT|PRIMARY_OUTPUT|0.947271|0.919888|0.894045|0.851661|0.920958|0.920940|
+|3|SNOWBALL PORTUGUESE LUCENE FILTER|PRIMARY_OUTPUT|0.947271|0.919888|0.894045|0.851661|0.920958|0.920940|
+|4|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|PRIMARY_OUTPUT|0.901330|0.809975|0.735434|0.680636|0.821785|0.821750|
+|5|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|PRIMARY_OUTPUT|0.122843|0.053118|0.033885|0.027284|0.163848|0.163828|
+|6|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.038310|0.015690|0.009865|0.007907|0.088319|0.088308|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.996619|0.999299|0.999347|0.999323|0.999323|
+|2|SNOWBALL PORTUGUESE DIRECT|PRIMARY_OUTPUT|0.919870|0.996663|0.967924|0.982083|0.982083|
+|3|SNOWBALL PORTUGUESE LUCENE FILTER|PRIMARY_OUTPUT|0.919870|0.996663|0.967924|0.982083|0.982083|
+|4|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|PRIMARY_OUTPUT|0.809936|0.996729|0.918475|0.956003|0.956003|
+|5|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|PRIMARY_OUTPUT|0.053105|0.999226|0.720580|0.837330|0.837330|
+|6|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.015686|0.999664|0.692383|0.818122|0.818122|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|5472616|20678|16444|22358183078|20678 / 22358203756|16444 / 5489060|
+|2|SNOWBALL PORTUGUESE DIRECT|PRIMARY_OUTPUT|4817239|167230|671821|22358036526|167230 / 22358203756|671821 / 5489060|
+|3|SNOWBALL PORTUGUESE LUCENE FILTER|PRIMARY_OUTPUT|4817239|167230|671821|22358036526|167230 / 22358203756|671821 / 5489060|
+|4|PORTUGUESE LUCENE PORTUGUESE STEM FILTER|PRIMARY_OUTPUT|3803488|99075|1685572|22358104681|99075 / 22358203756|1685572 / 5489060|
+|5|PORTUGUESE LUCENE PORTUGUESE LIGHT STEM FILTER|PRIMARY_OUTPUT|149830|2511|5339230|22358201245|2511 / 22358203756|5339230 / 5489060|
+|6|PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|PRIMARY_OUTPUT|43406|598|5445654|22358203158|598 / 22358203756|5445654 / 5489060|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 22358203756 (0.000000%)|0 / 5489060 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|5489060|0|0|22358203756|0 / 22358203756|0 / 5489060|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999999|38310 / 22358203756 (0.000171%)|0 / 5489060 (0.000000%)|0.994448|0.996522|0.996528|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.993069|1.000000|0.999998|0.999999|0.999998|0.000002|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.994448|0.996522|0.998606|0.993069|0.996528|0.996528|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|5489060|38310|0|22358165446|38310 / 22358203756|0 / 5489060|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|16444|20678|17632|790|0.373542%|3|212297|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `PT_PT`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/russian.md

@@ -0,0 +1,408 @@
+# Russian Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Russian. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `RU_RU` | 37,410 | 806,279 | 74,808 | 731,471 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **806,279**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 9,260 | 1.148% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 584,785 | 72.529% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 82,864 | 10.277% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 75,646 | 9.382% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 53,724 | 6.663% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 98.807% | 98.696% | 99.896% | Full Radixor dictionary patch-command stemmer. |
+| Lucene RussianLightStemFilter | 9.658% | 8.452% | 21.447% | Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer. |
+| Lucene SnowballFilter | 9.162% | 8.162% | 18.936% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Official Snowball direct | 9.162% | 8.162% | 18.936% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `russianRadixor` | 89.671 | 3.886 | 122.6 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene RussianLightStemFilter | `russianLuceneRussianLightStemFilter` | 60.522 | 5.310 | 82.7 | 0.675 | Light Russian suffix stemmer. |
+| Official Snowball direct | `snowballDirect[RUSSIAN]` | 106.031 | 9.287 | 145.0 | 1.182 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[RUSSIAN]` | 137.512 | 10.801 | 188.0 | 1.534 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `RU_RU` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/ru_ru/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.989827** among 4 deterministic stemmers. The runner-up is `SNOWBALL RUSSIAN LUCENE FILTER` at 0.834876, a difference of 0.154951. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.989852** among 4 deterministic stemmers. The runner-up is `SNOWBALL RUSSIAN DIRECT` at 0.834854, a difference of 0.154998. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **6 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.989827|155850 / 295576291016 (0.000053%)|266302 / 13089505 (2.034470%)|0.986313|0.983806|0.983814|
+|2|SNOWBALL RUSSIAN LUCENE FILTER|PRIMARY_OUTPUT|0.834876|3785790 / 295576291016 (0.001281%)|4322616 / 13089505 (33.023525%)|0.692485|0.683786|0.683923|
+|3|SNOWBALL RUSSIAN DIRECT|PRIMARY_OUTPUT|0.834867|3782908 / 295576291016 (0.001280%)|4322849 / 13089505 (33.025305%)|0.692603|0.683851|0.683989|
+|4|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.617692|321183 / 295576291016 (0.000109%)|10008438 / 13089505 (76.461547%)|0.577011|0.373649|0.461687|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987992|0.979655|0.999999|0.989827|0.999999|0.000001|
+|2|SNOWBALL RUSSIAN LUCENE FILTER|PRIMARY_OUTPUT|0.698408|0.669765|0.999987|0.834876|0.999973|0.000027|
+|3|SNOWBALL RUSSIAN DIRECT|PRIMARY_OUTPUT|0.698563|0.669747|0.999987|0.834867|0.999973|0.000027|
+|4|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.905597|0.235385|0.999999|0.617692|0.999965|0.000035|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.986313|0.983806|0.981311|0.968128|0.983815|0.983814|
+|2|SNOWBALL RUSSIAN LUCENE FILTER|PRIMARY_OUTPUT|0.692485|0.683786|0.675304|0.519510|0.683936|0.683923|
+|3|SNOWBALL RUSSIAN DIRECT|PRIMARY_OUTPUT|0.692603|0.683851|0.675318|0.519585|0.684003|0.683989|
+|4|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.577011|0.373649|0.276278|0.229747|0.461696|0.461687|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.983805|0.997699|0.997274|0.997487|0.997487|
+|2|SNOWBALL RUSSIAN LUCENE FILTER|PRIMARY_OUTPUT|0.683773|0.974131|0.953674|0.963794|0.963794|
+|3|SNOWBALL RUSSIAN DIRECT|PRIMARY_OUTPUT|0.683838|0.974180|0.953661|0.963811|0.963811|
+|4|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.373638|0.994311|0.870888|0.928516|0.928516|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|12823203|155850|266302|295576135166|155850 / 295576291016|266302 / 13089505|
+|2|SNOWBALL RUSSIAN LUCENE FILTER|PRIMARY_OUTPUT|8766889|3785790|4322616|295572505226|3785790 / 295576291016|4322616 / 13089505|
+|3|SNOWBALL RUSSIAN DIRECT|PRIMARY_OUTPUT|8766656|3782908|4322849|295572508108|3782908 / 295576291016|4322849 / 13089505|
+|4|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|3081067|321183|10008438|295575969833|321183 / 295576291016|10008438 / 13089505|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 295576291016 (0.000000%)|13 / 13089505 (0.000099%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0.999999|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|0.999999|0.999999|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|13089492|0|13|295576291016|0 / 295576291016|13 / 13089505|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999999|434710 / 295576291016 (0.000147%)|13 / 13089505 (0.000099%)|0.974119|0.983665|0.983796|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.967857|0.999999|0.999999|0.999999|0.999999|0.000001|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.974119|0.983665|0.993401|0.967856|0.983797|0.983796|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|13089492|434710|13|295575856306|434710 / 295576291016|13 / 13089505|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|266289|155850|278860|19162|2.492190%|4|788492|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **6 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.989852|155850 / 295000681652 (0.000053%)|265613 / 13087126 (2.029575%)|0.986322|0.983830|0.983838|
+|2|SNOWBALL RUSSIAN DIRECT|PRIMARY_OUTPUT|0.834854|3782908 / 295000681652 (0.001282%)|4322407 / 13087126 (33.027931%)|0.692561|0.683815|0.683953|
+|3|SNOWBALL RUSSIAN LUCENE FILTER|PRIMARY_OUTPUT|0.834854|3782908 / 295000681652 (0.001282%)|4322407 / 13087126 (33.027931%)|0.692561|0.683815|0.683953|
+|4|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.617630|318921 / 295000681652 (0.000108%)|10008238 / 13087126 (76.473918%)|0.577038|0.373540|0.461703|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987991|0.979704|0.999999|0.989852|0.999999|0.000001|
+|2|SNOWBALL RUSSIAN DIRECT|PRIMARY_OUTPUT|0.698516|0.669721|0.999987|0.834854|0.999973|0.000027|
+|3|SNOWBALL RUSSIAN LUCENE FILTER|PRIMARY_OUTPUT|0.698516|0.669721|0.999987|0.834854|0.999973|0.000027|
+|4|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.906139|0.235261|0.999999|0.617630|0.999965|0.000035|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.986322|0.983830|0.981350|0.968175|0.983839|0.983838|
+|2|SNOWBALL RUSSIAN DIRECT|PRIMARY_OUTPUT|0.692561|0.683815|0.675288|0.519544|0.683967|0.683953|
+|3|SNOWBALL RUSSIAN LUCENE FILTER|PRIMARY_OUTPUT|0.692561|0.683815|0.675288|0.519544|0.683967|0.683953|
+|4|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.577038|0.373540|0.276152|0.229664|0.461713|0.461703|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.983829|0.997697|0.997321|0.997509|0.997509|
+|2|SNOWBALL RUSSIAN DIRECT|PRIMARY_OUTPUT|0.683802|0.974149|0.953634|0.963782|0.963782|
+|3|SNOWBALL RUSSIAN LUCENE FILTER|PRIMARY_OUTPUT|0.683802|0.974149|0.953634|0.963782|0.963782|
+|4|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|0.373528|0.994350|0.870767|0.928464|0.928464|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|12821513|155850|265613|295000525802|155850 / 295000681652|265613 / 13087126|
+|2|SNOWBALL RUSSIAN DIRECT|PRIMARY_OUTPUT|8764719|3782908|4322407|294996898744|3782908 / 295000681652|4322407 / 13087126|
+|3|SNOWBALL RUSSIAN LUCENE FILTER|PRIMARY_OUTPUT|8764719|3782908|4322407|294996898744|3782908 / 295000681652|4322407 / 13087126|
+|4|RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|PRIMARY_OUTPUT|3078888|318921|10008238|295000362731|318921 / 295000681652|10008238 / 13087126|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 295000681652 (0.000000%)|0 / 13087126 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|13087126|0|0|295000681652|0 / 295000681652|0 / 13087126|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999999|434710 / 295000681652 (0.000147%)|0 / 13087126 (0.000000%)|0.974115|0.983663|0.983794|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.967851|1.000000|0.999999|0.999999|0.999999|0.000001|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.974115|0.983663|0.993401|0.967851|0.983794|0.983794|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|13087126|434710|0|295000246942|434710 / 295000681652|0 / 13087126|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|265613|155850|278860|18991|2.472358%|4|787549|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `RU_RU`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/spanish.md

@@ -0,0 +1,466 @@
+# Spanish Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Spanish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `ES_ES` | 65,059 | 926,393 | 120,121 | 806,272 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **926,393**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 5,367 | 0.579% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 524,682 | 56.637% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 240,872 | 26.001% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 130,089 | 14.043% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 25,383 | 2.740% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 97.459% | 97.544% | 96.891% | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | 49.074% | 42.656% | 92.154% | Benchmark-only Spanish Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene SpanishMinimalStemFilter | 17.284% | 5.347% | 97.403% | Minimal suffix reducer; narrow baseline, not a full stemmer. |
+| Lucene SpanishPluralStemFilter | 15.140% | 5.802% | 77.820% | Plural-focused suffix reducer; narrow baseline. |
+| Lucene SpanishLightStemFilter | 9.577% | 7.088% | 26.279% | Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer. |
+| Lucene SnowballFilter | 4.889% | 4.287% | 8.932% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Official Snowball direct | 4.889% | 4.287% | 8.930% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `spanishRadixor` | 78.919 | 7.253 | 97.9 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 2079.041 | 193.548 | 2578.6 | 26.344 | Benchmark-only Spanish Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene SpanishMinimalStemFilter | `spanishLuceneSpanishMinimalStemFilter` | 45.596 | 4.639 | 56.6 | 0.578 | Minimal Spanish suffix reducer; narrow baseline. |
+| Lucene SpanishLightStemFilter | `spanishLuceneSpanishLightStemFilter` | 42.003 | 1.683 | 52.1 | 0.532 | Light Spanish suffix stemmer. |
+| Lucene SpanishPluralStemFilter | `spanishLuceneSpanishPluralStemFilter` | 93.734 | 6.247 | 116.3 | 1.188 | Plural-oriented Spanish suffix reducer. |
+| Official Snowball direct | `snowballDirect[SPANISH]` | 171.995 | 11.035 | 213.3 | 2.179 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[SPANISH]` | 211.138 | 17.940 | 261.9 | 2.675 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `ES_ES` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/es_es/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.989295** among 7 deterministic stemmers. The runner-up is `SNOWBALL SPANISH LUCENE FILTER` at 0.652614, a difference of 0.336680. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.989429** among 7 deterministic stemmers. The runner-up is `SNOWBALL SPANISH DIRECT` at 0.652720, a difference of 0.336709. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **11 result rows**, **7 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.989295|288483 / 379567318110 (0.000076%)|898652 / 41973336 (2.141007%)|0.990105|0.985755|0.985780|
+|2|SNOWBALL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.652614|2230481 / 379567318110 (0.000588%)|29161643 / 41973336 (69.476591%)|0.627151|0.449411|0.509848|
+|3|SNOWBALL SPANISH DIRECT|PRIMARY_OUTPUT|0.652614|2228819 / 379567318110 (0.000587%)|29161649 / 41973336 (69.476605%)|0.627192|0.449424|0.509876|
+|4|HUNSPELL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.615102|536192 / 379567318110 (0.000141%)|32310860 / 41973336 (76.979490%)|0.583708|0.370408|0.466992|
+|5|SPANISH LUCENE SPANISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.514823|147956 / 379567318110 (0.000039%)|40729019 / 41973336 (97.035458%)|0.130864|0.057387|0.162762|
+|6|SPANISH LUCENE SPANISH PLURAL STEM FILTER|PRIMARY_OUTPUT|0.503874|58578 / 379567318110 (0.000015%)|41648091 / 41973336 (99.225115%)|0.037377|0.015357|0.081026|
+|7|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.501768|47859 / 379567318110 (0.000013%)|41824873 / 41973336 (99.646292%)|0.017361|0.007041|0.051714|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.993026|0.978590|0.999999|0.989295|0.999997|0.000003|
+|2|SNOWBALL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.851718|0.305234|0.999994|0.652614|0.999917|0.000083|
+|3|SNOWBALL SPANISH DIRECT|PRIMARY_OUTPUT|0.851812|0.305234|0.999994|0.652614|0.999917|0.000083|
+|4|HUNSPELL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.947425|0.230205|0.999999|0.615102|0.999913|0.000087|
+|5|SPANISH LUCENE SPANISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.893731|0.029645|1.000000|0.514823|0.999892|0.000108|
+|6|SPANISH LUCENE SPANISH PLURAL STEM FILTER|PRIMARY_OUTPUT|0.847383|0.007749|1.000000|0.503874|0.999890|0.000110|
+|7|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.756222|0.003537|1.000000|0.501768|0.999890|0.000110|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.990105|0.985755|0.981443|0.971910|0.985781|0.985780|
+|2|SNOWBALL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.627151|0.449411|0.350170|0.289832|0.509876|0.509848|
+|3|SNOWBALL SPANISH DIRECT|PRIMARY_OUTPUT|0.627192|0.449424|0.350173|0.289843|0.509904|0.509876|
+|4|HUNSPELL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.583708|0.370408|0.271278|0.227301|0.467014|0.466992|
+|5|SPANISH LUCENE SPANISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.130864|0.057387|0.036752|0.029541|0.162773|0.162762|
+|6|SPANISH LUCENE SPANISH PLURAL STEM FILTER|PRIMARY_OUTPUT|0.037377|0.015357|0.009664|0.007738|0.081032|0.081026|
+|7|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.017361|0.007041|0.004416|0.003533|0.051719|0.051714|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.985753|0.995418|0.993266|0.994341|0.994341|
+|2|SNOWBALL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.449379|0.981386|0.852461|0.912391|0.912391|
+|3|SNOWBALL SPANISH DIRECT|PRIMARY_OUTPUT|0.449392|0.981406|0.852463|0.912401|0.912401|
+|4|HUNSPELL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.370381|0.993314|0.790558|0.880414|0.880414|
+|5|SPANISH LUCENE SPANISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.057381|0.993824|0.756690|0.859195|0.859195|
+|6|SPANISH LUCENE SPANISH PLURAL STEM FILTER|PRIMARY_OUTPUT|0.015355|0.995442|0.723731|0.838115|0.838115|
+|7|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.007040|0.995635|0.710610|0.829316|0.829316|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|41074684|288483|898652|379567029627|288483 / 379567318110|898652 / 41973336|
+|2|SNOWBALL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|12811693|2230481|29161643|379565087629|2230481 / 379567318110|29161643 / 41973336|
+|3|SNOWBALL SPANISH DIRECT|PRIMARY_OUTPUT|12811687|2228819|29161649|379565089291|2228819 / 379567318110|29161649 / 41973336|
+|4|HUNSPELL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|9662476|536192|32310860|379566781918|536192 / 379567318110|32310860 / 41973336|
+|5|SPANISH LUCENE SPANISH LIGHT STEM FILTER|PRIMARY_OUTPUT|1244317|147956|40729019|379567170154|147956 / 379567318110|40729019 / 41973336|
+|6|SPANISH LUCENE SPANISH PLURAL STEM FILTER|PRIMARY_OUTPUT|325245|58578|41648091|379567259532|58578 / 379567318110|41648091 / 41973336|
+|7|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|148463|47859|41824873|379567270251|47859 / 379567318110|41824873 / 41973336|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.999993|2 / 379567318110 (0.000000%)|626 / 41973336 (0.001491%)|0.999997|0.999993|0.999993|
+|2|HUNSPELL SPANISH LUCENE FILTER|ANY_CANDIDATE|0.620065|416345 / 379567318110 (0.000110%)|31894218 / 41973336 (75.986855%)|0.600268|0.384195|0.480192|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0.999985|1.000000|0.999993|1.000000|0.000000|
+|2|HUNSPELL SPANISH LUCENE FILTER|ANY_CANDIDATE|0.960331|0.240131|0.999999|0.620065|0.999915|0.000085|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|0.999997|0.999993|0.999988|0.999985|0.999993|0.999993|
+|2|HUNSPELL SPANISH LUCENE FILTER|ANY_CANDIDATE|0.600268|0.384195|0.282504|0.237773|0.480214|0.480192|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL SPANISH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|41972710|2|626|379567318108|2 / 379567318110|626 / 41973336|
+|2|HUNSPELL SPANISH LUCENE FILTER|ANY_CANDIDATE|10079118|416345|31894218|379566901765|416345 / 379567318110|31894218 / 41973336|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999991|1349800 / 379567318110 (0.000356%)|626 / 41973336 (0.001491%)|0.974915|0.984168|0.984289|
+|2|HUNSPELL SPANISH LUCENE FILTER|ALL_CANDIDATES|0.620065|888077 / 379567318110 (0.000234%)|31894218 / 41973336 (75.986855%)|0.587073|0.380771|0.469749|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.968843|0.999985|0.999996|0.999991|0.999996|0.000004|
+|2|HUNSPELL SPANISH LUCENE FILTER|ALL_CANDIDATES|0.919024|0.240131|0.999998|0.620065|0.999914|0.000086|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.974915|0.984168|0.993598|0.968829|0.984291|0.984289|
+|2|HUNSPELL SPANISH LUCENE FILTER|ALL_CANDIDATES|0.587073|0.380771|0.281759|0.235156|0.469773|0.469749|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL SPANISH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|41972710|1349800|626|379565968310|1349800 / 379567318110|626 / 41973336|
+|2|HUNSPELL SPANISH LUCENE FILTER|ALL_CANDIDATES|10079118|888077|31894218|379566430033|888077 / 379567318110|31894218 / 41973336|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|HUNSPELL SPANISH LUCENE FILTER|416642|119847|351885|17877|2.051686%|5|890999|
+|Radixor|898026|288481|1061317|42637|4.893313%|21|916797|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **11 result rows**, **7 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.989429|276044 / 377860669765 (0.000073%)|885033 / 41863370 (2.114099%)|0.990385|0.986031|0.986056|
+|2|SNOWBALL SPANISH DIRECT|PRIMARY_OUTPUT|0.652720|2201196 / 377860669765 (0.000583%)|29076352 / 41863370 (69.455354%)|0.627946|0.449839|0.510450|
+|3|SNOWBALL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.652720|2201196 / 377860669765 (0.000583%)|29076352 / 41863370 (69.455354%)|0.627946|0.449839|0.510450|
+|4|HUNSPELL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.614999|531181 / 377860669765 (0.000141%)|32234855 / 41863370 (77.000144%)|0.583531|0.370163|0.466854|
+|5|SPANISH LUCENE SPANISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.514832|146613 / 377860669765 (0.000039%)|40621522 / 41863370 (97.033569%)|0.130949|0.057424|0.162875|
+|6|SPANISH LUCENE SPANISH PLURAL STEM FILTER|PRIMARY_OUTPUT|0.503877|57716 / 377860669765 (0.000015%)|41538714 / 41863370 (99.224487%)|0.037409|0.015370|0.081139|
+|7|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.501770|47148 / 377860669765 (0.000012%)|41715144 / 41863370 (99.645929%)|0.017379|0.007049|0.051824|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.993309|0.978859|0.999999|0.989429|0.999997|0.000003|
+|2|SNOWBALL SPANISH DIRECT|PRIMARY_OUTPUT|0.853138|0.305446|0.999994|0.652720|0.999917|0.000083|
+|3|SNOWBALL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.853138|0.305446|0.999994|0.652720|0.999917|0.000083|
+|4|HUNSPELL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.947717|0.229999|0.999999|0.614999|0.999913|0.000087|
+|5|SPANISH LUCENE SPANISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.894406|0.029664|1.000000|0.514832|0.999892|0.000108|
+|6|SPANISH LUCENE SPANISH PLURAL STEM FILTER|PRIMARY_OUTPUT|0.849058|0.007755|1.000000|0.503877|0.999890|0.000110|
+|7|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.758678|0.003541|1.000000|0.501770|0.999889|0.000111|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.990385|0.986031|0.981715|0.972447|0.986057|0.986056|
+|2|SNOWBALL SPANISH DIRECT|PRIMARY_OUTPUT|0.627946|0.449839|0.350441|0.290188|0.510478|0.510450|
+|3|SNOWBALL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.627946|0.449839|0.350441|0.290188|0.510478|0.510450|
+|4|HUNSPELL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.583531|0.370163|0.271053|0.227117|0.466876|0.466854|
+|5|SPANISH LUCENE SPANISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.130949|0.057424|0.036775|0.029561|0.162886|0.162875|
+|6|SPANISH LUCENE SPANISH PLURAL STEM FILTER|PRIMARY_OUTPUT|0.037409|0.015370|0.009672|0.007744|0.081145|0.081139|
+|7|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.017379|0.007049|0.004421|0.003537|0.051829|0.051824|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.986029|0.995464|0.993323|0.994392|0.994392|
+|2|SNOWBALL SPANISH DIRECT|PRIMARY_OUTPUT|0.449806|0.981469|0.852556|0.912482|0.912482|
+|3|SNOWBALL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.449806|0.981469|0.852556|0.912482|0.912482|
+|4|HUNSPELL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|0.370136|0.993362|0.790500|0.880396|0.880396|
+|5|SPANISH LUCENE SPANISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.057417|0.993866|0.756725|0.859234|0.859234|
+|6|SPANISH LUCENE SPANISH PLURAL STEM FILTER|PRIMARY_OUTPUT|0.015368|0.995484|0.723753|0.838145|0.838145|
+|7|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.007048|0.995676|0.710626|0.829341|0.829341|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|40978337|276044|885033|377860393721|276044 / 377860669765|885033 / 41863370|
+|2|SNOWBALL SPANISH DIRECT|PRIMARY_OUTPUT|12787018|2201196|29076352|377858468569|2201196 / 377860669765|29076352 / 41863370|
+|3|SNOWBALL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|12787018|2201196|29076352|377858468569|2201196 / 377860669765|29076352 / 41863370|
+|4|HUNSPELL SPANISH LUCENE FILTER|PRIMARY_OUTPUT|9628515|531181|32234855|377860138584|531181 / 377860669765|32234855 / 41863370|
+|5|SPANISH LUCENE SPANISH LIGHT STEM FILTER|PRIMARY_OUTPUT|1241848|146613|40621522|377860523152|146613 / 377860669765|40621522 / 41863370|
+|6|SPANISH LUCENE SPANISH PLURAL STEM FILTER|PRIMARY_OUTPUT|324656|57716|41538714|377860612049|57716 / 377860669765|41538714 / 41863370|
+|7|SPANISH LUCENE SPANISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|148226|47148|41715144|377860622617|47148 / 377860669765|41715144 / 41863370|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 377860669765 (0.000000%)|0 / 41863370 (0.000000%)|1.000000|1.000000|1.000000|
+|2|HUNSPELL SPANISH LUCENE FILTER|ANY_CANDIDATE|0.619928|412198 / 377860669765 (0.000109%)|31822108 / 41863370 (76.014205%)|0.600000|0.383864|0.479978|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|HUNSPELL SPANISH LUCENE FILTER|ANY_CANDIDATE|0.960568|0.239858|0.999999|0.619928|0.999915|0.000085|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|HUNSPELL SPANISH LUCENE FILTER|ANY_CANDIDATE|0.600000|0.383864|0.282205|0.237519|0.480000|0.479978|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL SPANISH LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|41863370|0|0|377860669765|0 / 377860669765|0 / 41863370|
+|2|HUNSPELL SPANISH LUCENE FILTER|ANY_CANDIDATE|10041262|412198|31822108|377860257567|412198 / 377860669765|31822108 / 41863370|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999998|1255381 / 377860669765 (0.000332%)|0 / 41863370 (0.000000%)|0.976572|0.985228|0.985334|
+|2|HUNSPELL SPANISH LUCENE FILTER|ALL_CANDIDATES|0.619928|878949 / 377860669765 (0.000233%)|31822108 / 41863370 (76.014205%)|0.586905|0.380469|0.469606|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.970885|1.000000|0.999997|0.999998|0.999997|0.000003|
+|2|HUNSPELL SPANISH LUCENE FILTER|ALL_CANDIDATES|0.919512|0.239858|0.999998|0.619928|0.999913|0.000087|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.976572|0.985228|0.994038|0.970885|0.985335|0.985334|
+|2|HUNSPELL SPANISH LUCENE FILTER|ALL_CANDIDATES|0.586905|0.380469|0.281467|0.234926|0.469630|0.469606|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|HUNSPELL SPANISH LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|41863370|1255381|0|377859414384|1255381 / 377860669765|0 / 41863370|
+|2|HUNSPELL SPANISH LUCENE FILTER|ALL_CANDIDATES|10041262|878949|31822108|377859790816|878949 / 377860669765|31822108 / 41863370|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|HUNSPELL SPANISH LUCENE FILTER|412747|118983|347768|17807|2.048262%|5|888962|
+|Radixor|885033|276044|979337|42403|4.877434%|21|914127|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `ES_ES`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/swedish.md

@@ -0,0 +1,420 @@
+# Swedish Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Swedish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `SV_SE` | 12,371 | 110,468 | 24,731 | 85,737 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **110,468**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 502 | 0.454% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 14,268 | 12.916% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 66,796 | 60.466% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 25,745 | 23.305% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 3,157 | 2.858% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 96.713% | 97.407% | 94.307% | Full Radixor dictionary patch-command stemmer. |
+| Lucene SwedishMinimalStemFilter | 49.532% | 49.186% | 50.730% | Minimal suffix reducer; narrow baseline, not a full stemmer. |
+| Lucene SwedishLightStemFilter | 45.672% | 46.383% | 43.209% | Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer. |
+| Official Snowball direct | 40.068% | 37.512% | 48.926% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+| Lucene SnowballFilter | 38.785% | 35.839% | 48.999% | Lucene TokenFilter integration path around the Snowball algorithm. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `swedishRadixor` | 5.489 | 0.355 | 64.0 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene SwedishMinimalStemFilter | `swedishLuceneSwedishMinimalStemFilter` | 4.630 | 0.130 | 54.0 | 0.843 | Minimal Swedish suffix reducer. |
+| Lucene SwedishLightStemFilter | `swedishLuceneSwedishLightStemFilter` | 4.876 | 0.328 | 56.9 | 0.888 | Light Swedish suffix stemmer. |
+| Official Snowball direct | `snowballDirect[SWEDISH]` | 7.517 | 0.072 | 87.7 | 1.370 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[SWEDISH]` | 9.793 | 0.338 | 114.2 | 1.784 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `SV_SE` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/sv_se/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.974636** among 5 deterministic stemmers. The runner-up is `SNOWBALL SWEDISH DIRECT` at 0.807534, a difference of 0.167101. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.974584** among 5 deterministic stemmers. The runner-up is `SNOWBALL SWEDISH DIRECT` at 0.807599, a difference of 0.166985. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **7 result rows**, **5 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.974636|24473 / 4812155436 (0.000509%)|19546 / 385342 (5.072377%)|0.939665|0.943246|0.943260|
+|2|SNOWBALL SWEDISH DIRECT|PRIMARY_OUTPUT|0.807534|67105 / 4812155436 (0.001394%)|148325 / 385342 (38.491781%)|0.739832|0.687540|0.692339|
+|3|SNOWBALL SWEDISH LUCENE FILTER|PRIMARY_OUTPUT|0.799307|64262 / 4812155436 (0.001335%)|154666 / 385342 (40.137333%)|0.736940|0.678180|0.684227|
+|4|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.796072|40227 / 4812155436 (0.000836%)|157161 / 385342 (40.784809%)|0.781991|0.698068|0.709491|
+|5|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.783685|45941 / 4812155436 (0.000955%)|166707 / 385342 (43.262089%)|0.757232|0.672808|0.684713|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.937292|0.949276|0.999995|0.974636|0.999991|0.000009|
+|2|SNOWBALL SWEDISH DIRECT|PRIMARY_OUTPUT|0.779348|0.615082|0.999986|0.807534|0.999955|0.000045|
+|3|SNOWBALL SWEDISH LUCENE FILTER|PRIMARY_OUTPUT|0.782117|0.598627|0.999987|0.799307|0.999955|0.000045|
+|4|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.850127|0.592152|0.999992|0.796072|0.999959|0.000041|
+|5|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.826360|0.567379|0.999990|0.783685|0.999956|0.000044|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.939665|0.943246|0.946855|0.892588|0.943265|0.943260|
+|2|SNOWBALL SWEDISH DIRECT|PRIMARY_OUTPUT|0.739832|0.687540|0.642152|0.523856|0.692361|0.692339|
+|3|SNOWBALL SWEDISH LUCENE FILTER|PRIMARY_OUTPUT|0.736940|0.678180|0.628098|0.513065|0.684249|0.684227|
+|4|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.781991|0.698068|0.630412|0.536179|0.709510|0.709491|
+|5|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.757232|0.672808|0.605321|0.506941|0.684733|0.684713|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.943241|0.992631|0.993395|0.993013|0.993013|
+|2|SNOWBALL SWEDISH DIRECT|PRIMARY_OUTPUT|0.687518|0.984860|0.942685|0.963311|0.963311|
+|3|SNOWBALL SWEDISH LUCENE FILTER|PRIMARY_OUTPUT|0.678157|0.985207|0.939659|0.961894|0.961894|
+|4|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.698048|0.988493|0.944582|0.966038|0.966038|
+|5|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.672787|0.986795|0.942303|0.964036|0.964036|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|365796|24473|19546|4812130963|24473 / 4812155436|19546 / 385342|
+|2|SNOWBALL SWEDISH DIRECT|PRIMARY_OUTPUT|237017|67105|148325|4812088331|67105 / 4812155436|148325 / 385342|
+|3|SNOWBALL SWEDISH LUCENE FILTER|PRIMARY_OUTPUT|230676|64262|154666|4812091174|64262 / 4812155436|154666 / 385342|
+|4|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|228181|40227|157161|4812115209|40227 / 4812155436|157161 / 385342|
+|5|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|PRIMARY_OUTPUT|218635|45941|166707|4812109495|45941 / 4812155436|166707 / 385342|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 4812155436 (0.000000%)|0 / 385342 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|385342|0|0|4812155436|0 / 4812155436|0 / 385342|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999995|47848 / 4812155436 (0.000994%)|0 / 385342 (0.000000%)|0.909640|0.941544|0.943152|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.889545|1.000000|0.999990|0.999995|0.999990|0.000010|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.909640|0.941544|0.975768|0.889545|0.943157|0.943152|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|385342|47848|0|4812107588|47848 / 4812155436|0 / 385342|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|19546|24473|23375|5767|5.878216%|5|104148|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **7 result rows**, **5 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.974584|24473 / 4789911577 (0.000511%)|19546 / 384563 (5.082652%)|0.939544|0.943132|0.943146|
+|2|SNOWBALL SWEDISH DIRECT|PRIMARY_OUTPUT|0.807599|67105 / 4789911577 (0.001401%)|147975 / 384563 (38.478741%)|0.739645|0.687500|0.692274|
+|3|SNOWBALL SWEDISH LUCENE FILTER|PRIMARY_OUTPUT|0.799355|64262 / 4789911577 (0.001342%)|154316 / 384563 (40.127625%)|0.736744|0.678122|0.684143|
+|4|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.795947|40227 / 4789911577 (0.000840%)|156939 / 384563 (40.809698%)|0.781694|0.697790|0.709212|
+|5|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.783598|45941 / 4789911577 (0.000959%)|166437 / 384563 (43.279515%)|0.756945|0.672575|0.684469|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.937167|0.949173|0.999995|0.974584|0.999991|0.000009|
+|2|SNOWBALL SWEDISH DIRECT|PRIMARY_OUTPUT|0.779037|0.615213|0.999986|0.807599|0.999955|0.000045|
+|3|SNOWBALL SWEDISH LUCENE FILTER|PRIMARY_OUTPUT|0.781800|0.598724|0.999987|0.799355|0.999954|0.000046|
+|4|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.849816|0.591903|0.999992|0.795947|0.999959|0.000041|
+|5|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.826025|0.567205|0.999990|0.783598|0.999956|0.000044|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.939544|0.943132|0.946748|0.892384|0.943151|0.943146|
+|2|SNOWBALL SWEDISH DIRECT|PRIMARY_OUTPUT|0.739645|0.687500|0.642223|0.523810|0.692296|0.692274|
+|3|SNOWBALL SWEDISH LUCENE FILTER|PRIMARY_OUTPUT|0.736744|0.678122|0.628142|0.512999|0.684165|0.684143|
+|4|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.781694|0.697790|0.630152|0.535851|0.709231|0.709212|
+|5|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.756945|0.672575|0.605126|0.506676|0.684489|0.684469|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.943127|0.992612|0.993378|0.992995|0.992995|
+|2|SNOWBALL SWEDISH DIRECT|PRIMARY_OUTPUT|0.687478|0.984821|0.942695|0.963298|0.963298|
+|3|SNOWBALL SWEDISH LUCENE FILTER|PRIMARY_OUTPUT|0.678100|0.985169|0.939661|0.961877|0.961877|
+|4|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|0.697770|0.988463|0.944528|0.965996|0.965996|
+|5|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|PRIMARY_OUTPUT|0.672553|0.986761|0.942265|0.964000|0.964000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|365017|24473|19546|4789887104|24473 / 4789911577|19546 / 384563|
+|2|SNOWBALL SWEDISH DIRECT|PRIMARY_OUTPUT|236588|67105|147975|4789844472|67105 / 4789911577|147975 / 384563|
+|3|SNOWBALL SWEDISH LUCENE FILTER|PRIMARY_OUTPUT|230247|64262|154316|4789847315|64262 / 4789911577|154316 / 384563|
+|4|SWEDISH LUCENE SWEDISH MINIMAL STEM FILTER|PRIMARY_OUTPUT|227624|40227|156939|4789871350|40227 / 4789911577|156939 / 384563|
+|5|SWEDISH LUCENE SWEDISH LIGHT STEM FILTER|PRIMARY_OUTPUT|218126|45941|166437|4789865636|45941 / 4789911577|166437 / 384563|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 4789911577 (0.000000%)|0 / 384563 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|384563|0|0|4789911577|0 / 4789911577|0 / 384563|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999995|47848 / 4789911577 (0.000999%)|0 / 384563 (0.000000%)|0.909473|0.941433|0.943047|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.889346|1.000000|0.999990|0.999995|0.999990|0.000010|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.909473|0.941433|0.975720|0.889346|0.943051|0.943047|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|384563|47848|0|4789863729|47848 / 4789911577|0 / 384563|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|19546|24473|23375|5767|5.891848%|5|103921|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `SV_SE`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/ukrainian.md

@@ -0,0 +1,474 @@
+# Ukrainian Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Ukrainian. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `UK_UA` | 1,493 | 15,737 | 2,985 | 12,752 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **15,737**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `AppendCharacterCommand` | Appends one character to the end of the word form. | 249 | 1.582% |
+| `BackwardCompoundCommand` | Applies a multi-step backward patch made from skip, delete, insert, and replace operations. | 4,160 | 26.435% |
+| `DeleteSuffixCommand` | Deletes one or more trailing characters from the word form. | 5,859 | 37.231% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 3,004 | 19.089% |
+| `ReplaceLastCharacterCommand` | Replaces the final character of the word form. | 2,465 | 15.664% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 99.307% | 99.365% | 99.062% | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | 86.815% | 83.759% | 99.866% | Benchmark-only Ukrainian Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene MorfologikFilter | 92.362% | 90.637% | 99.732% | Dictionary-based path; Morfologik can emit multiple terms. |
+| Morfologik direct | 92.362% | 90.637% | 99.732% | Direct dictionary lookup; first returned stem is used for quality when no ranking weight is exposed. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `ukrainianRadixor` | 0.682 | 0.057 | 53.5 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 43.527 | 1.207 | 3413.3 | 63.799 | Benchmark-only Ukrainian Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Morfologik direct | `ukrainianMorfologikDirect` | 8.680 | 0.073 | 680.7 | 12.723 | Direct Morfologik dictionary lookup; first returned stem is used for quality. |
+| Lucene MorfologikFilter | `ukrainianLuceneMorfologikFilter` | 14.575 | 0.248 | 1143.0 | 21.364 | Dictionary-based Morfologik TokenFilter; may emit multiple terms. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `UK_UA` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/uk_ua/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.995343** among 4 deterministic stemmers. The runner-up is `UKRAINIAN LUCENE MORFOLOGIK FILTER` at 0.928768, a difference of 0.066575. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.995342** among 4 deterministic stemmers. The runner-up is `UKRAINIAN LUCENE MORFOLOGIK FILTER` at 0.928751, a difference of 0.066591. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **12 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.995343|880 / 101387550 (0.000868%)|608 / 65340 (0.930517%)|0.987406|0.988637|0.988632|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.928768|828 / 101387550 (0.000817%)|9308 / 65340 (14.245485%)|0.956896|0.917054|0.919223|
+|3|UKRAINIAN MORFOLOGIK DIRECT|PRIMARY_OUTPUT|0.928646|828 / 101387550 (0.000817%)|9324 / 65340 (14.269972%)|0.956832|0.916912|0.919090|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|PRIMARY_OUTPUT|0.885793|794 / 101387550 (0.000783%)|14924 / 65340 (22.840526%)|0.933008|0.865139|0.871499|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.986588|0.990695|0.999991|0.995343|0.999985|0.000015|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.985438|0.857545|0.999992|0.928768|0.999900|0.000100|
+|3|UKRAINIAN MORFOLOGIK DIRECT|PRIMARY_OUTPUT|0.985434|0.857300|0.999992|0.928646|0.999900|0.000100|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|PRIMARY_OUTPUT|0.984495|0.771595|0.999992|0.885793|0.999845|0.000155|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987406|0.988637|0.989871|0.977529|0.988639|0.988632|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.956896|0.917054|0.880397|0.846814|0.919270|0.919223|
+|3|UKRAINIAN MORFOLOGIK DIRECT|PRIMARY_OUTPUT|0.956832|0.916912|0.880190|0.846572|0.919137|0.919090|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|PRIMARY_OUTPUT|0.933008|0.865139|0.806475|0.762331|0.871568|0.871499|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988630|0.997994|0.998266|0.998130|0.998130|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.917004|0.997990|0.971000|0.984310|0.984310|
+|3|UKRAINIAN MORFOLOGIK DIRECT|PRIMARY_OUTPUT|0.916862|0.997990|0.970876|0.984246|0.984246|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|PRIMARY_OUTPUT|0.865063|0.998114|0.949804|0.973360|0.973360|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|64732|880|608|101386670|880 / 101387550|608 / 65340|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|56032|828|9308|101386722|828 / 101387550|9308 / 65340|
+|3|UKRAINIAN MORFOLOGIK DIRECT|PRIMARY_OUTPUT|56016|828|9324|101386722|828 / 101387550|9324 / 65340|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|PRIMARY_OUTPUT|50416|794|14924|101386756|794 / 101387550|14924 / 65340|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 101387550 (0.000000%)|0 / 65340 (0.000000%)|1.000000|1.000000|1.000000|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.962151|122 / 101387550 (0.000120%)|4946 / 65340 (7.569636%)|0.982323|0.959732|0.960413|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ANY_CANDIDATE|0.962029|122 / 101387550 (0.000120%)|4962 / 65340 (7.594123%)|0.982267|0.959599|0.960286|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ANY_CANDIDATE|0.927570|326 / 101387550 (0.000322%)|9465 / 65340 (14.485767%)|0.962884|0.919443|0.922008|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.997984|0.924304|0.999999|0.962151|0.999950|0.000050|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ANY_CANDIDATE|0.997983|0.924059|0.999999|0.962029|0.999950|0.000050|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ANY_CANDIDATE|0.994199|0.855142|0.999997|0.927570|0.999903|0.000097|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.982323|0.959732|0.938156|0.922581|0.960438|0.960413|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ANY_CANDIDATE|0.982267|0.959599|0.937954|0.922337|0.960310|0.960286|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ANY_CANDIDATE|0.962884|0.919443|0.879752|0.850897|0.922053|0.922008|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|65340|0|0|101387550|0 / 101387550|0 / 65340|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|60394|122|4946|101387428|122 / 101387550|4946 / 65340|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ANY_CANDIDATE|60378|122|4962|101387428|122 / 101387550|4962 / 65340|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ANY_CANDIDATE|55875|326|9465|101387224|326 / 101387550|9465 / 65340|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999993|1490 / 101387550 (0.001470%)|0 / 65340 (0.000000%)|0.982084|0.988727|0.988782|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.962145|1368 / 101387550 (0.001349%)|4946 / 65340 (7.569636%)|0.966650|0.950323|0.950669|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ALL_CANDIDATES|0.962023|1368 / 101387550 (0.001349%)|4962 / 65340 (7.594123%)|0.966592|0.950191|0.950541|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ALL_CANDIDATES|0.927565|1271 / 101387550 (0.001254%)|9465 / 65340 (14.485767%)|0.950501|0.912349|0.914347|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.977705|1.000000|0.999985|0.999993|0.999985|0.000015|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.977850|0.924304|0.999987|0.962145|0.999938|0.000062|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ALL_CANDIDATES|0.977845|0.924059|0.999987|0.962023|0.999938|0.000062|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ALL_CANDIDATES|0.977759|0.855142|0.999987|0.927565|0.999894|0.000106|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.982084|0.988727|0.995460|0.977705|0.988789|0.988782|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.966650|0.950323|0.934539|0.905349|0.950700|0.950669|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ALL_CANDIDATES|0.966592|0.950191|0.934337|0.905109|0.950571|0.950541|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ALL_CANDIDATES|0.950501|0.912349|0.877142|0.838825|0.914398|0.914347|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|65340|1490|0|101386060|1490 / 101387550|0 / 65340|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|60394|1368|4946|101386182|1368 / 101387550|4946 / 65340|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ALL_CANDIDATES|60378|1368|4962|101386182|1368 / 101387550|4962 / 65340|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ALL_CANDIDATES|55875|1271|9465|101386279|1271 / 101387550|9465 / 65340|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|HUNSPELL UKRAINIAN LUCENE FILTER|5459|468|477|1322|9.280449%|6|15740|
+|UKRAINIAN LUCENE MORFOLOGIK FILTER|4362|706|540|2207|15.493155%|6|16937|
+|UKRAINIAN MORFOLOGIK DIRECT|4362|706|540|2207|15.493155%|6|16937|
+|Radixor|608|880|610|190|1.333801%|2|14435|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **12 result rows**, **4 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.995342|880 / 101259406 (0.000869%)|608 / 65324 (0.930745%)|0.987403|0.988634|0.988629|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.928751|828 / 101259406 (0.000818%)|9308 / 65324 (14.248974%)|0.956884|0.917032|0.919202|
+|3|UKRAINIAN MORFOLOGIK DIRECT|PRIMARY_OUTPUT|0.928751|828 / 101259406 (0.000818%)|9308 / 65324 (14.248974%)|0.956884|0.917032|0.919202|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|PRIMARY_OUTPUT|0.885796|794 / 101259406 (0.000784%)|14920 / 65324 (22.839998%)|0.933007|0.865141|0.871500|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.986585|0.990693|0.999991|0.995342|0.999985|0.000015|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.985434|0.857510|0.999992|0.928751|0.999900|0.000100|
+|3|UKRAINIAN MORFOLOGIK DIRECT|PRIMARY_OUTPUT|0.985434|0.857510|0.999992|0.928751|0.999900|0.000100|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|PRIMARY_OUTPUT|0.984492|0.771600|0.999992|0.885796|0.999845|0.000155|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.987403|0.988634|0.989868|0.977524|0.988636|0.988629|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.956884|0.917032|0.880367|0.846777|0.919249|0.919202|
+|3|UKRAINIAN MORFOLOGIK DIRECT|PRIMARY_OUTPUT|0.956884|0.917032|0.880367|0.846777|0.919249|0.919202|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|PRIMARY_OUTPUT|0.933007|0.865141|0.806479|0.762334|0.871570|0.871500|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988627|0.997992|0.998264|0.998128|0.998128|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|0.916982|0.997988|0.970978|0.984298|0.984298|
+|3|UKRAINIAN MORFOLOGIK DIRECT|PRIMARY_OUTPUT|0.916982|0.997988|0.970978|0.984298|0.984298|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|PRIMARY_OUTPUT|0.865065|0.998113|0.949788|0.973351|0.973351|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|64716|880|608|101258526|880 / 101259406|608 / 65324|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|PRIMARY_OUTPUT|56016|828|9308|101258578|828 / 101259406|9308 / 65324|
+|3|UKRAINIAN MORFOLOGIK DIRECT|PRIMARY_OUTPUT|56016|828|9308|101258578|828 / 101259406|9308 / 65324|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|PRIMARY_OUTPUT|50404|794|14920|101258612|794 / 101259406|14920 / 65324|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 101259406 (0.000000%)|0 / 65324 (0.000000%)|1.000000|1.000000|1.000000|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.962142|122 / 101259406 (0.000120%)|4946 / 65324 (7.571490%)|0.982318|0.959722|0.960404|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ANY_CANDIDATE|0.962142|122 / 101259406 (0.000120%)|4946 / 65324 (7.571490%)|0.982318|0.959722|0.960404|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ANY_CANDIDATE|0.927552|326 / 101259406 (0.000322%)|9465 / 65324 (14.489315%)|0.962874|0.919422|0.921988|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.997983|0.924285|0.999999|0.962142|0.999950|0.000050|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ANY_CANDIDATE|0.997983|0.924285|0.999999|0.962142|0.999950|0.000050|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ANY_CANDIDATE|0.994198|0.855107|0.999997|0.927552|0.999903|0.000097|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|0.982318|0.959722|0.938141|0.922562|0.960428|0.960404|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ANY_CANDIDATE|0.982318|0.959722|0.938141|0.922562|0.960428|0.960404|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ANY_CANDIDATE|0.962874|0.919422|0.879722|0.850861|0.922033|0.921988|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|65324|0|0|101259406|0 / 101259406|0 / 65324|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ANY_CANDIDATE|60378|122|4946|101259284|122 / 101259406|4946 / 65324|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ANY_CANDIDATE|60378|122|4946|101259284|122 / 101259406|4946 / 65324|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ANY_CANDIDATE|55859|326|9465|101259080|326 / 101259406|9465 / 65324|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999993|1490 / 101259406 (0.001471%)|0 / 65324 (0.000000%)|0.982079|0.988724|0.988779|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.962136|1368 / 101259406 (0.001351%)|4946 / 65324 (7.571490%)|0.966642|0.950311|0.950657|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ALL_CANDIDATES|0.962136|1368 / 101259406 (0.001351%)|4946 / 65324 (7.571490%)|0.966642|0.950311|0.950657|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ALL_CANDIDATES|0.927547|1271 / 101259406 (0.001255%)|9465 / 65324 (14.489315%)|0.950487|0.912326|0.914325|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.977699|1.000000|0.999985|0.999993|0.999985|0.000015|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.977845|0.924285|0.999986|0.962136|0.999938|0.000062|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ALL_CANDIDATES|0.977845|0.924285|0.999986|0.962136|0.999938|0.000062|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ALL_CANDIDATES|0.977752|0.855107|0.999987|0.927547|0.999894|0.000106|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.982079|0.988724|0.995459|0.977699|0.988787|0.988779|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|0.966642|0.950311|0.934522|0.905326|0.950688|0.950657|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ALL_CANDIDATES|0.966642|0.950311|0.934522|0.905326|0.950688|0.950657|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ALL_CANDIDATES|0.950487|0.912326|0.877111|0.838787|0.914376|0.914325|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|65324|1490|0|101257916|1490 / 101259406|0 / 65324|
+|2|UKRAINIAN LUCENE MORFOLOGIK FILTER|ALL_CANDIDATES|60378|1368|4946|101258038|1368 / 101259406|4946 / 65324|
+|3|UKRAINIAN MORFOLOGIK DIRECT|ALL_CANDIDATES|60378|1368|4946|101258038|1368 / 101259406|4946 / 65324|
+|4|HUNSPELL UKRAINIAN LUCENE FILTER|ALL_CANDIDATES|55859|1271|9465|101258135|1271 / 101259406|9465 / 65324|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|HUNSPELL UKRAINIAN LUCENE FILTER|5455|468|477|1321|9.279292%|6|15730|
+|UKRAINIAN LUCENE MORFOLOGIK FILTER|4362|706|540|2207|15.502950%|6|16928|
+|UKRAINIAN MORFOLOGIK DIRECT|4362|706|540|2207|15.502950%|6|16928|
+|Radixor|608|880|610|190|1.334645%|2|14426|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `UK_UA`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/languages/yiddish.md

@@ -0,0 +1,395 @@
+# Yiddish Stemmer Benchmarks
+
+This page reports same-language stemming benchmarks for Yiddish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+
+Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in `All exact` and `Changed exact`. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The [EnglishRadixorDictionaryCoverageBenchmark](../reference/english-coverage.md) table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.
+
+## Dictionary Corpus
+
+| Resource | Dictionary rows | Complete quality tokens | Already-root tokens | Changed speed tokens |
+| --- | ---: | ---: | ---: | ---: |
+| `YI` | 802 | 4,300 | 1,524 | 2,776 |
+
+## Radixor Patch Command Distribution
+
+Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is **4,300**.
+
+| Command class | Meaning | Word forms | Share |
+| --- | --- | ---: | ---: |
+| `DeletePrefixCommand` | Deletes one or more leading characters from the word form in forward traversal. | 25 | 0.581% |
+| `ForwardCompoundCommand` | Applies a multi-step forward patch made from skip, delete, insert, and replace operations. | 2,721 | 63.279% |
+| `PreserveCommand` | Returns the word form unchanged because it already matches the preferred root. | 1,551 | 36.070% |
+| `ReplaceFirstCharacterCommand` | Replaces the first character of the word form in forward traversal. | 3 | 0.070% |
+
+## Accuracy
+
+Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor | 98.930% | 98.343% | 100.000% | Full Radixor dictionary patch-command stemmer. |
+| Lucene SnowballFilter | 2.837% | 2.558% | 3.346% | Lucene TokenFilter integration path around the Snowball algorithm. |
+| Official Snowball direct | 2.837% | 2.558% | 3.346% | Official Snowball generated Java stemmer; rule-based suffix algorithm. |
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor | `radixor[YIDDISH]` | 0.254 | 0.004 | 50.7 | 1.000 | Full Radixor dictionary patch-command stemmer. |
+| Official Snowball direct | `snowballDirect[YIDDISH]` | 1.537 | 0.220 | 307.3 | 6.058 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[YIDDISH]` | 1.714 | 0.120 | 342.8 | 6.756 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+
+## Interpretation Notes
+
+- Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
+- Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
+- Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+- Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
+- Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
+
+<!-- STEMMING-QUALITY:START -->
+
+## Stemming Quality
+
+Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `YI` using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.
+
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+
+### Evaluation Scope and Key Findings
+
+The dictionary resource is `src/main/resources/yi/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.
+
+- **ALL_WORDS:** `Radixor` ranks first by balanced accuracy at **0.988241** among 3 deterministic stemmers. The runner-up is `SNOWBALL YIDDISH DIRECT` at 0.890988, a difference of 0.097253. This rank does not imply leadership in throughput or every secondary metric.
+- **LOWERCASE_GROUPS_ONLY:** `Radixor` ranks first by balanced accuracy at **0.988241** among 3 deterministic stemmers. The runner-up is `SNOWBALL YIDDISH DIRECT` at 0.890988, a difference of 0.097253. This rank does not imply leadership in throughput or every secondary metric.
+### `ALL_WORDS`
+
+This mode contains **5 result rows**, **3 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988241|195 / 6392909 (0.003050%)|149 / 6344 (2.348676%)|0.970881|0.972986|0.972965|
+|2|SNOWBALL YIDDISH DIRECT|PRIMARY_OUTPUT|0.890988|1151 / 6392909 (0.018004%)|1382 / 6344 (21.784363%)|0.805624|0.796661|0.796600|
+|3|SNOWBALL YIDDISH LUCENE FILTER|PRIMARY_OUTPUT|0.890988|1151 / 6392909 (0.018004%)|1382 / 6344 (21.784363%)|0.805624|0.796661|0.796600|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.969484|0.976513|0.999969|0.988241|0.999946|0.000054|
+|2|SNOWBALL YIDDISH DIRECT|PRIMARY_OUTPUT|0.811713|0.782156|0.999820|0.890988|0.999604|0.000396|
+|3|SNOWBALL YIDDISH LUCENE FILTER|PRIMARY_OUTPUT|0.811713|0.782156|0.999820|0.890988|0.999604|0.000396|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.970881|0.972986|0.975099|0.947393|0.972992|0.972965|
+|2|SNOWBALL YIDDISH DIRECT|PRIMARY_OUTPUT|0.805624|0.796661|0.787894|0.662041|0.796798|0.796600|
+|3|SNOWBALL YIDDISH LUCENE FILTER|PRIMARY_OUTPUT|0.805624|0.796661|0.787894|0.662041|0.796798|0.796600|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.972959|0.995691|0.996142|0.995917|0.995917|
+|2|SNOWBALL YIDDISH DIRECT|PRIMARY_OUTPUT|0.796462|0.982919|0.962014|0.972354|0.972354|
+|3|SNOWBALL YIDDISH LUCENE FILTER|PRIMARY_OUTPUT|0.796462|0.982919|0.962014|0.972354|0.972354|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|6195|195|149|6392714|195 / 6392909|149 / 6344|
+|2|SNOWBALL YIDDISH DIRECT|PRIMARY_OUTPUT|4962|1151|1382|6391758|1151 / 6392909|1382 / 6344|
+|3|SNOWBALL YIDDISH LUCENE FILTER|PRIMARY_OUTPUT|4962|1151|1382|6391758|1151 / 6392909|1382 / 6344|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 6392909 (0.000000%)|0 / 6344 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|6344|0|0|6392909|0 / 6392909|0 / 6344|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999970|389 / 6392909 (0.006085%)|0 / 6344 (0.000000%)|0.953240|0.970253|0.970653|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.942225|1.000000|0.999939|0.999970|0.999939|0.000061|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.953240|0.970253|0.987885|0.942225|0.970683|0.970653|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|6344|389|0|6392520|389 / 6392909|0 / 6344|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|149|195|194|89|2.487423%|3|3676|
+
+### `LOWERCASE_GROUPS_ONLY`
+
+This mode contains **5 result rows**, **3 evaluated stemmers**, and **3 output policies**. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.
+
+#### `PRIMARY_OUTPUT` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.988241|195 / 6392909 (0.003050%)|149 / 6344 (2.348676%)|0.970881|0.972986|0.972965|
+|2|SNOWBALL YIDDISH DIRECT|PRIMARY_OUTPUT|0.890988|1151 / 6392909 (0.018004%)|1382 / 6344 (21.784363%)|0.805624|0.796661|0.796600|
+|3|SNOWBALL YIDDISH LUCENE FILTER|PRIMARY_OUTPUT|0.890988|1151 / 6392909 (0.018004%)|1382 / 6344 (21.784363%)|0.805624|0.796661|0.796600|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.969484|0.976513|0.999969|0.988241|0.999946|0.000054|
+|2|SNOWBALL YIDDISH DIRECT|PRIMARY_OUTPUT|0.811713|0.782156|0.999820|0.890988|0.999604|0.000396|
+|3|SNOWBALL YIDDISH LUCENE FILTER|PRIMARY_OUTPUT|0.811713|0.782156|0.999820|0.890988|0.999604|0.000396|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.970881|0.972986|0.975099|0.947393|0.972992|0.972965|
+|2|SNOWBALL YIDDISH DIRECT|PRIMARY_OUTPUT|0.805624|0.796661|0.787894|0.662041|0.796798|0.796600|
+|3|SNOWBALL YIDDISH LUCENE FILTER|PRIMARY_OUTPUT|0.805624|0.796661|0.787894|0.662041|0.796798|0.796600|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|0.972959|0.995691|0.996142|0.995917|0.995917|
+|2|SNOWBALL YIDDISH DIRECT|PRIMARY_OUTPUT|0.796462|0.982919|0.962014|0.972354|0.972354|
+|3|SNOWBALL YIDDISH LUCENE FILTER|PRIMARY_OUTPUT|0.796462|0.982919|0.962014|0.972354|0.972354|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|PRIMARY_OUTPUT|6195|195|149|6392714|195 / 6392909|149 / 6344|
+|2|SNOWBALL YIDDISH DIRECT|PRIMARY_OUTPUT|4962|1151|1382|6391758|1151 / 6392909|1382 / 6344|
+|3|SNOWBALL YIDDISH LUCENE FILTER|PRIMARY_OUTPUT|4962|1151|1382|6391758|1151 / 6392909|1382 / 6344|
+
+</details>
+
+#### `ANY_CANDIDATE` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|0 / 6392909 (0.000000%)|0 / 6344 (0.000000%)|1.000000|1.000000|1.000000|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|0.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|1.000000|1.000000|1.000000|1.000000|1.000000|1.000000|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ANY_CANDIDATE|6344|0|0|6392909|0 / 6392909|0 / 6344|
+
+</details>
+
+#### `ALL_CANDIDATES` ranking
+
+<div class="quality-table quality-table--compact" role="region" aria-label="Compact stemming-quality ranking; scroll horizontally for additional columns" tabindex="0" markdown="1">
+
+| Rank | Stemmer | Output policy | Balanced accuracy | Over-stemming | Under-stemming | F0.5 | F1 | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.999970|389 / 6392909 (0.006085%)|0 / 6344 (0.000000%)|0.953240|0.970253|0.970653|
+
+</div>
+
+<details class="quality-details" markdown="1"><summary>Classification metrics</summary>
+
+| Rank | Stemmer | Output policy | Precision | Recall | Specificity | Balanced accuracy | Pairwise accuracy | Error rate |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.942225|1.000000|0.999939|0.999970|0.999939|0.000061|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Pair-relation metrics</summary>
+
+| Rank | Stemmer | Output policy | F0.5 | F1 | F2 | Jaccard | Fowlkes–Mallows | MCC |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|0.953240|0.970253|0.987885|0.942225|0.970683|0.970653|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Partition metrics (PRIMARY_OUTPUT only)</summary>
+
+| Rank | Stemmer | Output policy | Adjusted Rand Index | Homogeneity | Completeness | V-measure | Normalized mutual information |
+|---:|---|---|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|n/a|n/a|n/a|n/a|n/a|
+
+</details>
+
+<details class="quality-details" markdown="1"><summary>Raw pair counts</summary>
+
+| Rank | Stemmer | Output policy | TP | FP | FN | TN | Over error / possible | Under error / possible |
+|---:|---|---|---:|---:|---:|---:|---:|---:|
+|1|Radixor|ALL_CANDIDATES|6344|389|0|6392520|389 / 6392909|0 / 6344|
+
+</details>
+
+#### Multi-output analysis
+
+Alternative candidates are capability analyses, not replacements for the deterministic comparison.
+
+| Stemmer | Under pairs repaired | Best-case over pairs avoided | All-candidate collisions added | Multi-candidate forms | Multi-candidate share | Maximum candidates | Total candidate assignments |
+|---|---:|---:|---:|---:|---:|---:|---:|
+|Radixor|149|195|194|89|2.487423%|3|3676|
+
+### Output Policies and Metric Definitions
+
+`PRIMARY_OUTPUT` uses one deterministic stem per form and therefore defines a strict partition. `ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. `ALL_CANDIDATES` activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.
+
+For each row, `TP = underPossiblePairs - underErrorPairs`, `FN = underErrorPairs`, `FP = overErrorPairs`, and `TN = overPossiblePairs - overErrorPairs`. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as `n/a`.
+
+- Under-stemming rate: `FN / (TP + FN)`, the false-negative rate over same-group pairs.
+- Over-stemming rate: `FP / (TN + FP)`, the false-positive rate over different-group pairs.
+- Pairwise precision: `TP / (TP + FP)`, the fraction of predicted conflations that are gold-standard positive pairs.
+- Pairwise recall: `TP / (TP + FN)`, the fraction of gold-standard positive pairs successfully connected.
+- Pairwise specificity: `TN / (TN + FP)`, the fraction of different-group pairs correctly separated.
+- Balanced accuracy: `(recall + specificity) / 2`. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
+- Pairwise F-beta: `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`. F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
+- MCC: `(TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))`. It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
+- Jaccard index: `TP / (TP + FP + FN)`.
+- Fowlkes–Mallows index: `sqrt(precision * recall)`.
+- Pairwise accuracy: `(TP + TN) / (TP + TN + FP + FN)`. It can be dominated by true-negative cross-group pairs.
+- Pairwise error rate: `(FP + FN) / (TP + TN + FP + FN)`.
+
+Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is `1 - H(gold | predicted) / H(gold)`; completeness is `1 - H(predicted | gold) / H(predicted)`; V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization `MI / ((H(gold) + H(predicted)) / 2)`. These partition-only metrics apply to `PRIMARY_OUTPUT`; candidate-relation rows show `n/a`.
+
+### Provenance
+
+- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
+- Source SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Evaluation command: `./gradlew stemmingQuality`
+- Dictionary language: `YI`
+- Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
+- Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and `gradle.lockfile`
+- Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
+
+<!-- STEMMING-QUALITY:END -->

docs/benchmarks/reference/candidates.md

@@ -0,0 +1,31 @@
+# Benchmark Candidate Evaluation
+
+Implemented benchmark methods are documented on the per-language pages under [Language Benchmark Pages](../languages/index.md). This keeps the exact method names, implementation descriptions, accuracy tables, and speed tables close to the language where they are valid.
+
+## Included Candidate Families
+
+The current benchmark pages include Radixor baselines, Lucene language filters where the language matches a bundled Radixor resource, Lucene Stempel and Morfologik paths where applicable, official Snowball Java stemmers where same-language comparison is available, benchmark-only CISTEM German stemmer evaluation, benchmark-only Hunspell comparisons, and selected English-specific non-Lucene baselines such as OpenNLP Porter and Paice/Husk Lancaster.
+
+Benchmark-only Hunspell comparisons use bundled benchmark dictionaries and the Lucene HunspellStemFilter adapter over the selected language token streams.
+The CISTEM candidate is implemented in `src/jmh/java/org/egothor/stemmer/benchmark/Cistem.java` and follows the original MIT-licensed upstream implementation from Leonie Weissweiler's CISTEM project.
+CISTEM German gold-standard files are not vendored in this repository. The Gradle JMH resource preparation tasks download `goldstandard1.txt` and `goldstandard2.txt` from the upstream CISTEM repository into generated build resources.
+
+Direct stemmer APIs and Lucene TokenFilter paths are documented separately on language pages. TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
+
+For the benchmark refresh used in this documentation build:
+
+- Hunspell families are included in `HunspellStemmerComparisonBenchmark` (speed) and `HunspellStemmerComparisonBenchmarkQuality` (quality for all benchmark languages in this corpus). The legacy
+  `EnglishHunspellStemmerComparisonBenchmarkQuality` result is retained for continuity.
+- CISTEM quality is present in the published per-language results under `GERMAN_CISTEM`. CISTEM speed is present as `germanCistem` in `MultiLanguageStemmerComparisonBenchmark`.
+
+## Evaluated But Skipped Candidates
+
+| Candidate | Language | Link/source | Reason skipped |
+| --- | --- | --- | --- |
+| Lucene Arabic, Bulgarian, Bengali, Sorani, Greek, Galician, Hindi, Indonesian, Latvian, Telugu filters | Various | `lucene-analysis-common` | No bundled same-language Radixor resource in this repository snapshot. |
+| Lucene analyzer-only paths | Multiple | Lucene analyzers | Full analyzers mix tokenization, stop-word handling, and other behavior; direct filters are used where available. |
+| Lucene StemmerOverrideFilter | Multiple | `lucene-analysis-common` | Override map facility, not a stemmer algorithm. |
+| Additional Snowball Lovins | English | Official Snowball Java distribution | No Lovins Java stemmer was present in the selected Snowball Java distribution. |
+| Lemur Project Krovetz Stemmer | English | Lemur Project | Lucene KStem represents the Krovetz-style path without adding separate dependency and license risk. |
+| Smile Lancaster / Paice-Husk | English | Smile NLP | Smile is large for one stemmer; Paice/Husk is included through a smaller benchmark-only generated path. |
+| `stemmerEval` reference repository | Multiple | `https://github.com/endredy/stemmerEval` | Used only as a candidate reference; no code or data copied. |

docs/benchmarks/reference/corpora.md

@@ -0,0 +1,28 @@
+# Benchmark Corpora
+
+The table below describes the Radixor resources used to build speed and quality corpora. `Total tokens` is the complete dictionary token count used by quality benchmarks. `Already-root tokens` counts fields where the token is already equal to the line root. `Changed tokens` is the speed workload before the minimum-size repeat rule.
+
+| Language resource | Dictionary rows | Total tokens | Already-root tokens | Changed tokens | Speed timing tokens |
+| --- | ---: | ---: | ---: | ---: | ---: |
+| `cs_cz` | 5,113 | 56,612 | 10,049 | 46,563 | 46,563 |
+| `da_dk` | 4,179 | 32,256 | 8,356 | 23,900 | 23,900 |
+| `de_de` | 39,315 | 213,440 | 73,799 | 139,641 | 139,641 |
+| `es_es` | 65,059 | 926,393 | 120,121 | 806,272 | 806,272 |
+| `fa_ir` | 69 | 3,770 | 138 | 3,632 | 5,000 |
+| `fi_fi` | 57,027 | 1,865,215 | 110,525 | 1,754,690 | 1,754,690 |
+| `fr_fr` | 59,240 | 474,110 | 108,141 | 365,969 | 365,969 |
+| `he_il` | 2,358 | 61,071 | 4,715 | 56,356 | 56,356 |
+| `hu_hu` | 19,406 | 935,713 | 38,775 | 896,938 | 896,938 |
+| `it_it` | 10,009 | 337,546 | 20,004 | 317,542 | 317,542 |
+| `nb_no` | 17,929 | 90,757 | 33,376 | 57,381 | 57,381 |
+| `nl_nl` | 4,992 | 31,466 | 9,981 | 21,485 | 21,485 |
+| `nn_no` | 4,688 | 19,651 | 6,089 | 13,562 | 13,562 |
+| `pl_pl` | 9,990 | 132,308 | 19,957 | 112,351 | 112,351 |
+| `pt_pt` | 4,001 | 215,490 | 8,002 | 207,488 | 207,488 |
+| `ru_ru` | 37,410 | 806,279 | 74,808 | 731,471 | 731,471 |
+| `sv_se` | 12,371 | 110,468 | 24,731 | 85,737 | 85,737 |
+| `uk_ua` | 1,493 | 15,737 | 2,985 | 12,752 | 12,752 |
+| `us_uk` | 396,939 | 1,004,374 | 793,874 | 210,500 | 210,500 |
+| `yi` | 802 | 4,300 | 1,524 | 2,776 | 5,000 |
+
+Speed benchmarks process the complete changed-token dictionary sequence for the language. Only resources with fewer than 5,000 changed tokens are repeated to reach the minimum timing size; larger resources are not sampled or truncated.

docs/benchmarks/reference/english-coverage.md

@@ -0,0 +1,43 @@
+# English Dictionary Coverage Benchmark
+
+`EnglishRadixorDictionaryCoverageBenchmark` builds Radixor from deterministic slices of the English dictionary rows and evaluates accuracy against the complete dictionary. The speed method then stems the full changed-token English timing corpus.
+
+This benchmark is the clearest demonstration of the Radixor quality/speed envelope after contracted-trie compilation. More dictionary knowledge still gives the strongest changed-form precision, but uniform-subtree contraction removes much of the historical lookup-depth penalty. The table should therefore be read as a measured operating curve rather than as a strictly monotonic function of dictionary size.
+
+| Used rows | Actual row ratio | All exact | Changed exact | Root preserved | Speed ms/op | Error ms | ns/token |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 100.000% | 97.478% | 97.197% | 97.552% | 28.578 | 7.571 | 135.8 |
+| 90% | 90.000% | 97.047% | 94.913% | 97.613% | 26.612 | 9.227 | 126.4 |
+| 80% | 80.000% | 96.635% | 92.768% | 97.661% | 23.331 | 8.106 | 110.8 |
+| 70% | 70.000% | 96.209% | 90.565% | 97.705% | 22.362 | 1.957 | 106.2 |
+| 60% | 60.000% | 95.750% | 88.384% | 97.703% | 16.497 | 2.026 | 78.4 |
+| 50% | 50.000% | 95.262% | 86.107% | 97.690% | 16.035 | 0.986 | 76.2 |
+| 40% | 40.000% | 94.753% | 83.855% | 97.643% | 16.459 | 0.664 | 78.2 |
+| 30% | 30.000% | 94.208% | 81.651% | 97.537% | 19.566 | 0.758 | 92.9 |
+| 20% | 20.000% | 93.633% | 79.366% | 97.416% | 14.616 | 0.487 | 69.4 |
+| 10% | 10.000% | 92.868% | 76.516% | 97.204% | 18.093 | 3.147 | 86.0 |
+
+## Column Meanings
+
+- `Used rows`: requested deterministic percentage of English dictionary rows used to build the trie.
+- `Actual row ratio`: selected rows divided by all parsed English dictionary rows.
+- `All exact`: exact agreement over the complete dictionary.
+- `Changed exact`: exact agreement over dictionary tokens where `token != expectedRoot`.
+- `Root preserved`: percentage of already-root dictionary tokens that are left unchanged.
+- `Speed ms/op`: JMH average time for one full changed-token English operation.
+- `Error ms`: JMH score error converted to milliseconds.
+- `ns/token`: `Speed ms/op` divided by 210,500 changed English tokens.
+
+For non-English languages, the same principle applies: dictionary-driven Radixor quality depends on the amount and consistency of the language resource, while contracted tries reduce the cost of uniform regions in the compiled lookup graph. The English table is the clearest because the English resource is large and the benchmark can show gradual deterministic reductions from 100% to 10%.
+
+## Why The Historical Porter Ratio Changed
+
+The historical English benchmark in `HEAD` used synthetic lexical families. Its `familyCount=5000` parameter generated roughly 70,000 artificial tokens rather than measuring the complete real English dictionary resource. That older workload was useful as a low-level stress test, but it was not a dictionary-quality comparison. Many synthetic tokens were not present in the Radixor dictionary, so Radixor often executed a fast miss path where lookup returned `null` and no patch command was applied.
+
+The current benchmark is intentionally based on real Radixor dictionary data. For English, the speed workload processes 210,500 changed token/root pairs where the dictionary token differs from the expected root, and the quality workload evaluates the complete 1,004,374-token dictionary. This is a hit-heavy workload that measures real lookup plus compiled patch-command application against known expected roots. It is therefore a different and more linguistically meaningful workload than the historical synthetic benchmark.
+
+The result must be interpreted in Radixor's favor through both speed and exact-root quality. Non-Radixor stemmers can look faster because many of them perform narrower or more aggressive transformations and do not attempt to match the dictionary root with the same precision. The English result table shows that this speed often comes with substantially lower `All exact` and `Changed exact` accuracy.
+
+Radixor uses the dictionary as training data for transformation rules. With the full English dictionary, it reaches much higher exact-root agreement than the Porter-family and other narrow baselines. Higher speed is still possible by reducing the amount or complexity of the input dictionary used to build the stemmer, but that is an explicit quality/speed trade-off rather than an accidental benchmark artifact.
+
+The coverage table shows that contracted tries substantially improve the operating point. Reducing dictionary knowledge still primarily damages changed-form exactness, while root preservation remains high. Even when Radixor is trained from only 10% of the English dictionary rows, the complete-dictionary `All exact` score remains above 92%. This is why Radixor performance should be discussed as a configurable quality/speed point, not as a single fixed ratio against Porter.

docs/benchmarks/reference/environment.md

@@ -0,0 +1,56 @@
+# Benchmark Environment And Reports
+
+The values below are environment-specific and must not be read as universal performance claims.
+
+| Item | Value |
+| --- | --- |
+| Benchmark date | 2026-07-06 (Europe/Prague) |
+| Focused comparison command family | `./gradlew jmh -Pjmh.includes='.*StemmerComparisonBenchmark.*' --no-daemon` |
+| English coverage command | `./gradlew jmh -Pjmh.includes='.*EnglishRadixorDictionaryCoverageBenchmark.*' --no-daemon` |
+| Speed result reports | `build/reports/jmh/stemmer-comparison-2026-07-06.csv`, `build/reports/jmh/stemmer-comparison-2026-07-06.txt`, `build/reports/jmh/english-coverage-2026-07-06.csv`, and `build/reports/jmh/english-coverage-2026-07-06.txt` |
+| Accuracy result reports | `build/reports/jmh/stemmer-comparison-2026-07-06.csv`, `build/reports/jmh/english-coverage-2026-07-06.csv`, and deterministic Radixor exact-root accounting over the same bundled language corpora |
+| Final comparison JMH scope | Stemmer comparison benchmarks only; internal `FrequencyTrie*` microbenchmarks were not run |
+| Coverage JMH scope | English Radixor dictionary coverage benchmark only |
+| JMH version | 1.37 |
+| Speed benchmark mode | Average time, `time/op` |
+| Score unit | `ns/op` |
+| Speed warmup | 3 iterations, 1 s each |
+| Speed measurement | 5 iterations, 1 s each |
+| Accuracy warmup | 3 JMH warmup iterations were applied by the Gradle invocation; timing scores from quality methods are not interpreted |
+| Accuracy measurement | 5 JMH measurement samples; documentation uses deterministic auxiliary counter ratios from the same report |
+| Fork count in generated report files | 1 |
+| Default fork policy for accuracy-only benchmark classes | `@Fork(0)` for future default runs because accuracy counters are deterministic and not interpreted as speed |
+| Thread count | 1 |
+| JVM reported by JMH | JDK 25.0.3, OpenJDK 64-Bit Server VM, 25.0.3+9 |
+| Java runtime | OpenJDK Runtime Environment, Red Hat build 25.0.3+9 |
+| JVM invoker | `/usr/lib/jvm/java-25-openjdk/bin/java` |
+| Operating system | Fedora Linux 44 (MATE-Compiz) |
+| Kernel | Linux 7.0.12-201.fc44.x86_64 |
+| Architecture | x86_64 |
+| CPU | AMD Ryzen 5 8600G w/ Radeon 760M Graphics |
+| Physical cores | 6 |
+| Logical CPUs | 12 |
+
+## Contracted Trie Baseline
+
+All Radixor rows in the refreshed benchmark tables use contracted compiled patch tries. During compilation, a subtree whose reachable entries all resolve to the same preferred patch command is represented as an accepting leaf. Runtime lookup can therefore stop as soon as that leaf is reached, which reduces depth in uniform regions while preserving the preferred result used by `get()`.
+
+## Report Files
+
+Generated local report files for this benchmark update:
+
+- `build/reports/jmh/stemmer-comparison-2026-07-06.csv`
+- `build/reports/jmh/stemmer-comparison-2026-07-06.txt`
+- `build/reports/jmh/english-coverage-2026-07-06.csv`
+- `build/reports/jmh/english-coverage-2026-07-06.txt`
+
+JMH TXT and CSV reports are still published as benchmark artifacts. They are not converted into a Porter speed badge.
+
+## Published Metrics
+
+The historical English Radixor versus Porter performance badge is no longer generated. `tools/generate-pages-badges.py` now produces only coverage and mutation badge endpoint JSON files:
+
+- `coverage-badge.json`
+- `pitest-badge.json`
+
+The README therefore no longer presents a single Porter speed ratio. Benchmark interpretation now uses both speed and quality, because a narrow or aggressive stemmer can be fast while disagreeing with the dictionary root much more often than Radixor.

docs/benchmarks/reference/linguistic-quality.md

@@ -0,0 +1,82 @@
+# Linguistic Quality Methodology
+
+This evaluation measures agreement between the relation predicted by a stemmer and the gold-standard relation represented by Radixor dictionary groups. It does not require a generated stem to equal one predetermined lemma string. Runtime performance and linguistic quality are separate measurements.
+
+## Scope and fair-comparison rules
+
+The authoritative Radixor language universe is the reconciled set of `stemmer.gz` resources under `src/main/resources` and `StemmerPatchTrieLoader.Language`. Radixor is evaluated for every reconciled language. A third-party adapter is evaluated only for languages supported by its tested implementation and having a compatible Radixor dictionary; unsupported combinations are absent rather than assigned zero quality.
+
+Within one language and dictionary mode, every adapter receives the same original included forms. Exact duplicates are removed only within one dictionary row. Identical surface forms in different rows remain distinct entries. Candidate strings use exact `String.equals`, with no evaluation-only lowercasing, normalization, accent removal, or gold-label-aware selection. Adapter preprocessing and lifecycle match the JMH comparison path.
+
+## Gold-standard pairs
+
+Every usable dictionary row is a gold-standard equivalence group. An unordered pair from the same row is positive; a pair from different rows is negative. For group size `n`, `C2(n) = n * (n - 1) / 2`.
+
+- `TP = underPossiblePairs - underErrorPairs`: same-group pairs correctly related.
+- `FN = underErrorPairs`: same-group pairs incorrectly separated.
+- `FP = overErrorPairs`: different-group pairs incorrectly related.
+- `TN = overPossiblePairs - overErrorPairs`: different-group pairs correctly separated.
+
+Under-stemming is the false-negative relation among same-group pairs. Over-stemming is the false-positive relation among different-group pairs. Their percentages use different denominators and must not be added or averaged without an explicitly defined composite.
+
+## Dictionary-processing modes
+
+- `ALL_WORDS` includes every valid group and preserves every original form.
+- `LOWERCASE_GROUPS_ONLY` excludes an entire group if any Unicode code point is uppercase or titlecase. Retained forms are not converted to lowercase. Digits, punctuation, combining marks, and characters without case distinctions do not exclude a group by themselves.
+
+## Output policies
+
+`PRIMARY_OUTPUT` uses the adapter's deterministic primary stem. It defines a strict predicted partition and is the principal direct comparison between implementations.
+
+`ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound. Same-group pairs succeed when candidate sets intersect. Different-group pairs avoid an error whenever a non-colliding candidate selection exists. Selection may differ between pairs, so this policy is not deterministic runtime behaviour and may not correspond to one globally realizable assignment.
+
+`ALL_CANDIDATES` treats every returned candidate as active. Two forms are related when their candidate sets intersect. Alternatives can recover same-group relationships while introducing cross-group collisions. This overlapping relation need not be transitive or form a partition.
+
+Candidate-aware policies are reported as capability analyses. They are not mixed into the principal `PRIMARY_OUTPUT` ranking.
+
+## Relation metrics
+
+Undefined denominators produce `n/a`, never zero, `NaN`, or infinity. Metrics are calculated from unrounded raw counts and displayed with six decimals.
+
+| Metric | Formula | Range and interpretation | Sensitivity and applicability |
+| --- | --- | --- | --- |
+| Under-stemming rate | `FN / (TP + FN)` | `[0, 1]`; lower is better. False-negative rate over same-group pairs. | Sensitive to splitting large gold groups. All policies. |
+| Over-stemming rate | `FP / (TN + FP)` | `[0, 1]`; lower is better. False-positive rate over different-group pairs. | The denominator is usually very large. All policies. |
+| Precision | `TP / (TP + FP)` | `[0, 1]`; higher is better. Fraction of predicted relations that are gold-positive. | Penalizes over-stemming. All policies, with oracle-assisted interpretation for `ANY_CANDIDATE`. |
+| Recall | `TP / (TP + FN)` | `[0, 1]`; higher is better. Fraction of gold-positive pairs recovered. | Equivalent to one minus the under-stemming rate. All policies. |
+| Specificity | `TN / (TN + FP)` | `[0, 1]`; higher is better. Fraction of negative pairs separated. | Sensitive to cross-group collisions. All policies. |
+| Balanced accuracy | `(recall + specificity) / 2` | `[0, 1]`; higher is better. Equal weight for positive and negative classes. | Primary navigation metric; less dominated by TN than ordinary accuracy, but not uniquely authoritative. |
+| Pairwise accuracy | `(TP + TN) / (TP + TN + FP + FN)` | `[0, 1]`; higher is better. | Can be dominated by the very large TN class and is not the default ranking metric. |
+| Pairwise error rate | `(FP + FN) / (TP + TN + FP + FN)` | `[0, 1]`; lower is better. | Also sensitive to the number of negative pairs. |
+| F0.5 | `1.25 TP / (1.25 TP + 0.25 FN + FP)` | `[0, 1]`; higher is better. | Gives greater weight to precision and over-stemming avoidance. |
+| F1 | `2 TP / (2 TP + FN + FP)` | `[0, 1]`; higher is better. | Equal precision/recall emphasis. |
+| F2 | `5 TP / (5 TP + 4 FN + FP)` | `[0, 1]`; higher is better. | Gives greater weight to recall and under-stemming avoidance. |
+| Jaccard | `TP / (TP + FP + FN)` | `[0, 1]`; higher is better. | Excludes TN. All policies. |
+| Fowlkes–Mallows | `sqrt(precision * recall)` | `[0, 1]`; higher is better. | Geometric balance of precision and recall. All policies. |
+| MCC | `(TP TN - FP FN) / sqrt((TP+FP)(TP+FN)(TN+FP)(TN+FN))` | `[-1, 1]`; higher is better. Uses all four counts. | Informative under imbalance; undefined for a zero product denominator. All policies with policy-specific interpretation. |
+
+The general F-beta formula is `((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP)`.
+
+## Partition-only metrics
+
+These metrics apply only to `PRIMARY_OUTPUT`. Candidate relations are not forced into artificial partitions.
+
+- Adjusted Rand Index is the Rand agreement corrected for agreement expected from the gold/predicted contingency-table marginals. Its usual range is `[-1, 1]`, with `1` indicating identical partitions.
+- Homogeneity is `1 - H(gold | predicted) / H(gold)`, in `[0, 1]`; each predicted cluster ideally contains one gold group.
+- Completeness is `1 - H(predicted | gold) / H(predicted)`, in `[0, 1]`; each gold group ideally maps to one predicted cluster.
+- V-measure is the harmonic mean of homogeneity and completeness, in `[0, 1]`.
+- Normalized mutual information uses arithmetic-mean entropy normalization: `MI / ((H(gold) + H(predicted)) / 2)`, in `[0, 1]` under this implementation.
+
+Entropy zero cases follow the evaluator's explicit perfect/undefined conventions. Language tables render inapplicable candidate-policy values as `n/a`.
+
+## Aggregation and ranking
+
+Macro metrics average defined per-language values, giving each language equal weight. Micro metrics sum TP, FP, FN, and TN before calculating a metric. Cross-stemmer aggregate comparisons require the exact common supported-language intersection; unsupported languages are not zero-filled.
+
+Language tables sort by unrounded balanced accuracy, then MCC, F1, over-stemming rate, over-stemming error count, under-stemming rate, stemmer name, and stable policy order. Display rounding never controls rank.
+
+Multiple metrics and Pearson/Spearman correlation datasets are published because metric suitability and correlation remain analytical questions. Strong correlation does not establish equivalence.
+
+## Limitations
+
+Dictionary groups encode the available annotation, not every linguistic distinction. Homographs may occur in different groups, singleton rows contribute no under-stemming pair, and group size affects pair counts. `ANY_CANDIDATE` is optimistic; `ALL_CANDIDATES` measures an overlapping graph; neither is a deterministic global assignment. Results characterize the tested versions, adapters, dictionaries, and preprocessing, not every deployment or domain.

docs/benchmarks/reference/methodology.md

@@ -0,0 +1,59 @@
+# Benchmark Methodology
+
+The stemmer comparison suite measures Radixor and Java stemmers on the same language and deterministic Radixor dictionary-derived data. Published Radixor rows in this refresh use contracted compiled patch tries, where uniform preferred-command subtrees are collapsed into accepting leaves before the trie is frozen for lookup. For each language, the bundled dictionary resource stores the expected root as the first tab-separated field on a line and its surface forms on the same line. Every single-token field on that line can therefore be paired with the same expected root.
+
+Published stemmer comparison results must come only from benchmark classes matching `.*StemmerComparisonBenchmark.*`. Internal `FrequencyTrie*` microbenchmarks are not part of those results.
+
+## Benchmark Passes
+
+There are two distinct benchmark passes:
+
+- Speed benchmarks process only changed dictionary pairs where `token != expectedRoot`. This removes already-root tokens from timing so a stemmer is measured on words that actually require a transformation. If a language has fewer than 5,000 changed pairs, the complete changed-pair sequence is repeated in stable order until the timing corpus has at least 5,000 tokens. Larger changed-pair corpora are not sampled or truncated.
+- Quality benchmarks process the complete dictionary for the language. They report exact agreement over all tokens, exact agreement over changed tokens only, and preservation of tokens that are already roots.
+
+Timing corpora are generated once per JMH JVM and kept in memory as shared `{token, expectedRoot}` arrays. Corpus construction, dictionary loading, trie loading, table loading, and analyzer construction are setup work and are not included in measured benchmark methods.
+
+Performance is interpreted as average time per input token:
+
+```text
+timePerChangedTokenNs = JMH score ns/op / changedTimingTokenCount
+```
+
+This is necessary because Radixor dictionaries have different token counts by language.
+
+## Exact-root quality and interpretation
+
+Runtime and exact-root agreement must be interpreted separately. Light, minimal, possessive, and aggressive rule-based implementations deliberately address different scopes and may achieve lower latency by performing fewer transformations. A throughput advantage does not establish higher linguistic quality, and higher dictionary agreement does not establish lower operational cost.
+
+The [English dictionary coverage benchmark](english-coverage.md) shows this operating curve explicitly: contracted tries reduce lookup cost in uniform regions, while reduced dictionary coverage still lowers changed-form precision.
+
+## Normalization Policy
+
+Radixor is measured over dictionary tokens from its own resources: lower-case with `Locale.ROOT`, diacritics preserved. The corpus is normalized during setup, so the Radixor benchmark path uses `FrequencyTrie.getNormalized(CharSequence)` and does not measure redundant lookup-time lowercasing or diacritic normalization.
+
+Lucene TokenFilter paths include required normalization in the measured pipeline. Examples include lower-case normalization for filters requiring lower-case input, German normalization before German light/minimal stemming, and Persian decimal, Arabic, and Persian normalization before Persian stemming. No ASCII folding is applied to Czech or Polish paths, because those Lucene stemmers are diacritic-aware or dictionary/table-backed for those languages. TokenFilter throughput methods materialize each emitted `CharTermAttribute` as a `String` before passing it to the JMH `Blackhole`, so output consumption is easier to inspect and closer to the direct stemmer methods.
+
+For right-to-left Radixor languages, patch application uses the traversal direction stored in trie metadata. This is required because static backward patch application is not correct for all bundled languages.
+
+## Quality Metric
+
+The quality pass reports exact-root agreement against the expected root from the Radixor dictionary line. It writes to the normal JMH report files:
+
+- `build/reports/jmh/jmh-results.csv`
+- `build/reports/jmh/jmh-results.txt`
+
+Accuracy is computed from standard JMH secondary rows:
+
+```text
+allExactPercent = correctMatches / evaluatedTokens * 100
+changedExactPercent = changedCorrectMatches / changedEvaluatedTokens * 100
+rootPreservedPercent = rootPreservedMatches / rootEvaluatedTokens * 100
+```
+
+`allExactPercent` uses the complete dictionary. `changedExactPercent` uses only tokens where `token != expectedRoot`. `rootPreservedPercent` measures whether a stemmer leaves already-root dictionary entries unchanged.
+
+Morfologik can emit multiple terms for one input token. The quality benchmark uses the first emitted term for exact-root accounting when no ranking weight is exposed. Throughput benchmarks for Morfologik TokenFilter paths consume all emitted terms.
+
+Quality reports use JMH auxiliary counter rows. Exact-root accounting is deterministic for a fixed corpus and stemmer, so repeated measurement samples duplicate the same counters; documentation uses the counter ratios and does not interpret quality benchmark timing scores.
+
+Pairwise over-stemming, under-stemming, candidate-aware policies, balanced accuracy, and partition comparison are a separate analytical evaluation. See [Linguistic Quality Methodology](linguistic-quality.md); exact-root accuracy must not be interpreted as the complement of pairwise under-stemming.

docs/benchmarks/reference/reproducibility.md

@@ -0,0 +1,61 @@
+# Reproducibility and Raw Data
+
+## Published quality snapshot
+
+- Machine-readable CSV: [stemming-quality.csv](../data/stemming-quality.csv)
+- SHA-256 record: [stemming-quality.sha256](../data/stemming-quality.sha256)
+- SHA-256: `5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28`
+- Complete scenarios: 308
+- Authoritative language universe: 20 languages
+- Language-page scenarios: 302 across 19 existing benchmark pages
+
+The six remaining scenarios are the three Radixor policies in two modes for `HE_IL`. Hebrew is present in the complete result snapshot but has no existing language benchmark page.
+
+The CSV contains raw TP, FP, FN, and TN counts; raw over/under numerators and denominators; candidate statistics; relation metrics; and partition-only metrics. Documentation is regenerated from this file rather than manually transcribed.
+
+## Commands
+
+```bash
+./gradlew stemmingQuality
+./gradlew publishStemmingQualityDocumentation
+./gradlew verifyStemmingQualityDocumentation
+./gradlew test
+mkdocs build --strict
+```
+
+`stemmingQuality` performs the expensive complete evaluation and is intentionally not attached to `test` or `check`. It prepares JMH third-party dependencies automatically and writes:
+
+- `build/reports/stemming-quality/stemming-quality.csv`
+- `build/reports/stemming-quality/stemming-quality.md`
+- `build/reports/stemming-quality/metric-correlations-pearson.csv`
+- `build/reports/stemming-quality/metric-correlations-spearman.csv`
+
+Audit mode is enabled with `-PstemmingQualityAudit=true`. Language, stemmer, dictionary-mode, output-policy, and ranking filters are documented on the central [stemming-quality page](../../stemming-quality.md). Filtered reports use separate filenames and cannot be accepted as publication sources.
+
+`publishStemmingQualityDocumentation` validates the complete build CSV, copies a versioned documentation snapshot, and replaces only marked generated sections. `verifyStemmingQualityDocumentation` re-renders from the checked-in snapshot and fails on changed values, ordering, missing pages, duplicate keys, arithmetic inconsistencies, policy violations, or stale sections.
+
+## Performance benchmark reproduction
+
+The JMH comparison command family is:
+
+```bash
+./gradlew jmh -Pjmh.includes='.*StemmerComparisonBenchmark.*' --no-daemon
+```
+
+The exact JMH configuration, hardware, operating system, and JDK captured for the published performance tables are listed in [Environment and reports](environment.md). Quality and performance reports are separate datasets and are not combined into an undocumented scalar.
+
+## Recorded and unavailable provenance
+
+The performance documentation records its 2026-07-06 environment, JDK 25.0.3, operating system, and hardware. The quality CSV records the evaluated identifiers and counts but does not embed the Radixor Git revision, generation date, JDK, operating system, dictionary content hash, or immutable upstream revisions for every downloaded source. These fields are explicitly unavailable for this snapshot and are not reconstructed from filesystem timestamps.
+
+Dependency versions that are reproducible from repository configuration include Apache Lucene 10.5.0, Morfologik 2.1.9, the Ukrainian dictionary artifact 4.9.1, and JMH 1.37. Other upstream branches or downloaded dictionary revisions should be pinned and embedded in a future result schema.
+
+## Correlation and audit data
+
+Pearson and Spearman files are generated from unrounded metric values in cohorts separated by dictionary mode and output policy. A missing coefficient means too few observations, undefined input, or zero variance. Correlation is descriptive and does not demonstrate that two metrics are scientifically interchangeable.
+
+Audit reports preserve original multilingual forms and identify high-contributing dictionary groups. They are build artifacts rather than checked-in publication data because of their size. No documentation value is manually altered after generation.
+
+## JMH badge compatibility
+
+The quality documentation generator does not invoke JMH, change JMH result formats, or modify badge tooling. Existing JMH result paths and historical badge-compatible inputs remain independent. The repository currently publishes coverage and mutation badge metadata and retains JMH TXT/CSV artifacts as documented in [Environment and reports](environment.md).

docs/benchmarks/reference/tested-stemmers.md

@@ -0,0 +1,29 @@
+# Tested Stemmer Inventory
+
+The JMH adapter registry is authoritative for evaluated implementations and language mappings. Names below describe the implementation actually invoked, not an abstract algorithm in every possible implementation. Unsupported language combinations are omitted rather than scored as failures.
+
+| Family or implementation | Upstream / attribution | Tested version or revision | Evaluated scope | Output capability and adapter behaviour | Interpretation notes |
+| --- | --- | --- | --- | --- | --- |
+| Radixor | Egothor / Radixor project | Current repository revision; exact revision was not embedded in the quality CSV | All 20 reconciled Radixor dictionary languages; 19 have benchmark pages | Deterministic preferred patch via `get`; ranked distinct alternatives via `getAll`; primary is always included | Dictionary-derived compiled patch trie. Quality depends on dictionary coverage and annotation. |
+| Apache Lucene language stem filters | Apache Lucene project | 10.5.0 | Adapter-declared language-specific subsets | TokenFilter lifecycle and language normalization match JMH; normally single-output | Light, minimal, possessive, and language stem filters deliberately implement different scopes. Narrow scope is not a defect. |
+| Apache Lucene SnowballFilter | Apache Lucene project using Snowball algorithms | Lucene 10.5.0 | Snowball-supported subset of Radixor languages | Single primary token emitted through the Lucene TokenFilter path | Includes TokenStream overhead and required normalization. |
+| Official Snowball Java | Snowball project | Repository preparation downloads the configured upstream Java distribution; an immutable revision was not recorded in the quality CSV | Same-language adapter subset | Direct generated Java API; single output | Rule-based suffix algorithms provide broad baselines rather than dictionary-root guarantees. |
+| Lucene Stempel | Apache Lucene / Polish stemming tables | Lucene 10.5.0 | Polish | Direct and TokenFilter paths where registered; single primary output | Table-driven Polish implementation. |
+| Morfologik | Morfologik project; Lucene integration by Apache Lucene | Morfologik 2.1.9, Lucene integration 10.5.0; Ukrainian dictionary artifact 4.9.1 | Registered Polish and Ukrainian paths | Deterministic first lemma for primary comparison; all distinct lemma strings for candidate policies | Several analyses may share a lemma and are deduplicated by exact string equality. |
+| Hunspell via Lucene | Hunspell dictionaries from the `wooorm/dictionaries` repository; adapter by Apache Lucene | Lucene 10.5.0; dictionary repository revision was not recorded | Configured German, English, Spanish, French, Dutch, Polish, and Ukrainian dictionaries | First emitted stem is primary; all distinct stems at the token position are candidates | Dictionary content and affix rules differ by language. |
+| CISTEM | Leonie Weissweiler, CISTEM project | Upstream `master` source path used by preparation; immutable commit not recorded | German | Single output | German stemming algorithm; benchmark-only implementation and gold-standard preparation remain under JMH infrastructure. |
+| OpenNLP Porter | Apache OpenNLP project | Version resolved by `gradle/opennlp-benchmarks.gradle` and `gradle.lockfile` | English | Direct single output | Porter-family English baseline. |
+| Lucene Porter source copy | Apache Lucene project | 10.5.0 source artifact | English | Package-isolated benchmark-only generated source; single output | Generated into the JMH build tree, never production code. |
+| Paice/Husk Lancaster | Upstream Java implementation from `Hopper262/paice-husk-stemmer` | Configured upstream branch/revision in `gradle/paicehusk-benchmarks.gradle`; immutable commit not recorded | English | Direct single output | Aggressive rule-based English baseline; benchmark-only generated source. |
+
+## Preprocessing and lifecycle
+
+The quality evaluator calls the same adapter matrix used by JMH. Each language mapping is explicit. Retained dictionary forms are not evaluation-lowercased or normalized. Where an implementation requires preprocessing, such as Lucene German or Persian normalization, that operation is part of its documented adapter path. Stateful TokenFilters are reset through the same sequential lifecycle used by the benchmark and are not invoked concurrently.
+
+Candidate sets are non-null, non-empty, contain the deterministic primary output, contain no null strings, and are deduplicated using exact Java string equality. Gold-standard group identity never selects, removes, or ranks a candidate.
+
+## Coverage fairness
+
+Radixor coverage is derived independently from its resources and language enumeration. Third-party coverage is the intersection of that universe with actual adapter support. Absence therefore means “not supported or not configured for this language,” not “zero quality.” Consult each language page for the exact evaluated rows.
+
+Project authors and organizations are named only where repository configuration or source notices establish attribution. No broader authorship or license claim is inferred when metadata was not captured.

docs/built-in-languages.md

@@ -1,252 +1,267 @@
 # Built-in Languages
 
-> ← Back to [README.md](../README.md)
-
-Radixor provides a set of **bundled stemmer dictionaries** that can be loaded directly without preparing custom data.
-
-These built-in resources are useful for:
-
-- quick integration
-- testing and evaluation
-- reference behavior
-- prototyping search pipelines
-
-
+Radixor ships with a curated set of bundled stemmer dictionaries that can be loaded directly from the library distribution. These resources are intended to provide an immediately usable baseline for evaluation, prototyping, integration, and general-purpose stemming workloads, while still fitting naturally into workflows where the bundled baseline is later refined, extended, or replaced with custom lexical data.
 
 ## Overview
 
 Bundled dictionaries are exposed through:
 
 ```java
-StemmerPatchTrieLoader.Language
+org.egothor.stemmer.StemmerPatchTrieLoader.Language
 ```
 
-They are packaged with the library and loaded from the classpath.
+Each bundled dictionary is packaged with the library as a compressed UTF-8 text resource. When loaded through the runtime API, the resource is parsed by `StemmerDictionaryParser`, transformed into patch-command mappings, and compiled into a read-only `FrequencyTrie<CompiledPatchCommand>` by `StemmerPatchTrieLoader`.
 
+The bundled language definition also carries a language-level right-to-left flag. That flag is used by the loader to derive the `WordTraversalDirection` used for both trie-key construction and patch-command generation. In practice, left-to-right bundled languages use historical backward Egothor traversal, while right-to-left bundled languages use forward traversal over the stored form.
 
+## Supported bundled languages
 
-## Supported languages
-
-The following language identifiers are currently available:
-
-| Language | Enum constant     | Description                  |
-|----------|------------------|------------------------------|
-| Danish   | `DA_DK`          | Danish                       |
-| German   | `DE_DE`          | German                       |
-| Spanish  | `ES_ES`          | Spanish                      |
-| French   | `FR_FR`          | French                       |
-| Italian  | `IT_IT`          | Italian                      |
-| Dutch    | `NL_NL`          | Dutch                        |
-| Norwegian| `NO_NO`          | Norwegian                    |
-| Portuguese| `PT_PT`         | Portuguese                   |
-| Russian  | `RU_RU`          | Russian                      |
-| Swedish  | `SV_SE`          | Swedish                      |
-| English  | `US_UK`          | Standard English             |
-| English  | `US_UK_PROFI`    | Extended English dictionary  |
-
+The following bundled language identifiers are currently available:
 
+| Language | Enum constant | Writing direction | Notes | Benchmark page |
+|---|---|---:|---|---|
+| Czech | `CS_CZ` | LTR | Bundled general-purpose dictionary | [Czech](benchmarks/languages/czech.md) |
+| Danish | `DA_DK` | LTR | Bundled general-purpose dictionary | [Danish](benchmarks/languages/danish.md) |
+| German | `DE_DE` | LTR | Bundled general-purpose dictionary | [German](benchmarks/languages/german.md) |
+| Spanish | `ES_ES` | LTR | Bundled general-purpose dictionary | [Spanish](benchmarks/languages/spanish.md) |
+| Persian | `FA_IR` | RTL | Bundled dictionary uses forward traversal over the stored form | [Persian](benchmarks/languages/persian.md) |
+| Finnish | `FI_FI` | LTR | Bundled general-purpose dictionary | [Finnish](benchmarks/languages/finnish.md) |
+| French | `FR_FR` | LTR | Bundled general-purpose dictionary | [French](benchmarks/languages/french.md) |
+| Hebrew | `HE_IL` | RTL | Bundled dictionary uses forward traversal over the stored form | No same-language external benchmark in this run |
+| Hungarian | `HU_HU` | LTR | Bundled general-purpose dictionary | [Hungarian](benchmarks/languages/hungarian.md) |
+| Italian | `IT_IT` | LTR | Bundled general-purpose dictionary | [Italian](benchmarks/languages/italian.md) |
+| Norwegian Bokmål | `NB_NO` | LTR | Bundled general-purpose dictionary | [Norwegian Bokmal](benchmarks/languages/norwegian-bokmal.md) |
+| Dutch | `NL_NL` | LTR | Bundled general-purpose dictionary | [Dutch](benchmarks/languages/dutch.md) |
+| Norwegian Nynorsk | `NN_NO` | LTR | Bundled general-purpose dictionary | [Norwegian Nynorsk](benchmarks/languages/norwegian-nynorsk.md) |
+| Polish | `PL_PL` | LTR | Bundled general-purpose dictionary | [Polish](benchmarks/languages/polish.md) |
+| Portuguese | `PT_PT` | LTR | Bundled general-purpose dictionary | [Portuguese](benchmarks/languages/portuguese.md) |
+| Russian | `RU_RU` | LTR | Bundled general-purpose dictionary | [Russian](benchmarks/languages/russian.md) |
+| Swedish | `SV_SE` | LTR | Bundled general-purpose dictionary | [Swedish](benchmarks/languages/swedish.md) |
+| Ukrainian | `UK_UA` | LTR | Bundled general-purpose dictionary | [Ukrainian](benchmarks/languages/ukrainian.md) |
+| English | `US_UK` | LTR | Bundled general-purpose dictionary | [English](benchmarks/languages/english.md) |
+| Yiddish | `YI` | RTL | Bundled dictionary uses forward traversal over the stored form | [Yiddish](benchmarks/languages/yiddish.md) |
 
 ## Basic usage
 
-Load a bundled stemmer:
+Load a bundled dictionary like this:
 
 ```java
 import java.io.IOException;
 
+import org.egothor.stemmer.CompiledPatchCommand;
 import org.egothor.stemmer.FrequencyTrie;
 import org.egothor.stemmer.ReductionMode;
 import org.egothor.stemmer.StemmerPatchTrieLoader;
 
 public final class BuiltInExample {
 
-    public static void main(String[] args) throws IOException {
-        FrequencyTrie<String> trie = StemmerPatchTrieLoader.load(
-                StemmerPatchTrieLoader.Language.US_UK_PROFI,
+    private BuiltInExample() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadCompiled(
+                StemmerPatchTrieLoader.Language.US_UK,
                 true,
-                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
-        );
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+
+        System.out.println(trie.traversalDirection());
     }
 }
 ```
 
+This call loads the bundled dictionary resource for the selected language, parses its lexical entries, derives patch-command mappings, and compiles the result into a read-only trie.
 
-
-## Example: stemming with `US_UK_PROFI`
+## Example: stemming with a bundled dictionary
 
 ```java
 import java.io.IOException;
 
-import org.egothor.stemmer.*;
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
 
 public final class EnglishExample {
 
-    public static void main(String[] args) throws IOException {
-        FrequencyTrie<String> trie = StemmerPatchTrieLoader.load(
-                StemmerPatchTrieLoader.Language.US_UK_PROFI,
-                true,
-                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
-        );
+    private EnglishExample() {
+        throw new AssertionError("No instances.");
+    }
 
-        String word = "running";
-        String patch = trie.get(word);
-        String stem = PatchCommandEncoder.apply(word, patch);
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadCompiled(
+                StemmerPatchTrieLoader.Language.US_UK,
+                true,
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+
+        final String word = "running";
+        final CompiledPatchCommand patch = trie.get(word);
+        final String stem = patch == null ? word : patch.apply(word);
 
         System.out.println(word + " -> " + stem);
     }
 }
 ```
 
+`CompiledPatchCommand` values are compiled with the traversal direction used when the trie and its patch commands were produced.
 
+## Traversal behavior and right-to-left languages
 
-## `US_UK` vs `US_UK_PROFI`
+Bundled dictionaries are not all processed identically.
 
-### `US_UK`
+For traditional left-to-right suffix-oriented resources, Radixor preserves historical Egothor behavior and traverses logical word characters backward. That means trie paths are constructed from the logical end of the stored word toward its beginning, and patch commands are interpreted with the same backward traversal model.
 
-* smaller dictionary
-* faster load time
-* suitable for lightweight use cases
+For bundled right-to-left languages such as Persian, Hebrew, and Yiddish, Radixor uses forward traversal over the stored form. In those cases:
 
-### `US_UK_PROFI`
+- trie keys are traversed from the logical beginning of the stored form,
+- patch commands are generated in that same forward direction,
+- compiled patch-command application uses `WordTraversalDirection.FORWARD`, which is naturally captured when `loadCompiled(...)` creates `CompiledPatchCommand` values.
 
-* larger and more complete dataset
-* better coverage of word forms
-* improved stemming quality
-* slightly larger memory footprint
+This design keeps the traversal policy explicit and consistent across dictionary loading, trie lookup, binary persistence, builder reconstruction, and patch application.
 
-### Recommendation
+## Reduction behavior
 
-Use:
+Bundled dictionaries can be compiled using any supported `ReductionMode`. The reduction configuration controls how semantically equivalent subtrees are merged during trie compilation, while preserving the contract of the selected mode.
 
-```
-US_UK_PROFI
-```
+Typical entry points are:
 
-for most applications unless memory constraints are strict.
+- `StemmerPatchTrieLoader.loadCompiled(language, storeOriginal, reductionMode)`
+- `StemmerPatchTrieLoader.loadCompiled(language, storeOriginal, reductionSettings)`
 
+For most users, `ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS` is the most conservative general-purpose choice because it preserves ranked `getAll(...)` behavior.
 
+Compiled bundled dictionaries also use internal uniform-subtree contraction. If a whole subtree
+would return the same preferred patch command, Radixor stores that subtree as an accepting leaf and
+removes the deeper branches. This is the contracted trie representation used by the published
+benchmark tables and is independent of the public reduction mode selected by the caller.
 
-## How bundled dictionaries are loaded
+## Intended role of bundled dictionaries
 
-Internally:
+Bundled dictionaries should be understood as practical default resources.
 
-- dictionaries are stored as text resources
-- parsed using `StemmerDictionaryParser`
-- compiled into a trie at load time
+They are a good fit when:
 
-This means:
+- a supported language is already available,
+- immediate usability matters,
+- a reasonable baseline is sufficient,
+- the goal is evaluation, prototyping, or straightforward integration.
 
-- first load includes parsing + compilation cost
-- subsequent usage is fast
+They are also well suited to staged refinement workflows in which a bundled base is loaded first, then extended with domain-specific vocabulary, and finally persisted as a custom binary artifact.
 
+## Character representation
 
+Bundled dictionaries are ordinary UTF-8 lexical resources. The parser reads them as text, the trie stores standard Java strings, and the patch-command model operates on general character sequences.
 
-## When to use bundled languages
+This is important for two reasons:
 
-Bundled dictionaries are suitable when:
+1. the built-in resources are not limited to ASCII-only processing,
+2. the traversal model is orthogonal to character encoding and script choice.
 
-- you need quick results without preparing custom data
-- you are prototyping or experimenting
-- your language requirements match the provided datasets
+In other words, right-to-left handling in the loader is about logical traversal strategy, not about introducing a separate character model.
 
+## When to prefer custom dictionaries
 
+A custom dictionary is usually the better choice when:
 
-## When to use custom dictionaries
-
-You should prefer custom dictionaries when:
-
-- domain-specific vocabulary is important
-- accuracy requirements are high
-- you need full control over stemming behavior
-
-Typical examples:
-
-- technical terminology
-- product catalogs
-- biomedical text
-- legal or financial language
+- domain-specific vocabulary materially affects stemming quality,
+- lexical coverage must be controlled more precisely,
+- a stronger lexical resource is available than the bundled baseline,
+- operational requirements demand an explicitly curated, versioned artifact.
 
+Typical examples include:
 
+- technical terminology,
+- biomedical language,
+- legal or financial vocabulary,
+- organization-specific product and process names,
+- dictionaries maintained with project-specific validation rules.
 
 ## Production recommendation
 
-For production systems:
+For production systems, the most robust workflow is usually:
 
-1. Load a bundled dictionary
-2. Extend it with domain-specific terms (optional)
-3. Compile it into a binary `.radixor.gz` file
-4. Deploy the compiled artifact
-5. Load it using `loadBinary(...)`
+1. start from a bundled dictionary when it is suitable,
+2. extend it with domain-specific forms if needed,
+3. rebuild it into a binary artifact,
+4. deploy that compiled binary artifact,
+5. load it at runtime through `loadBinaryCompiled(...)`.
 
-This avoids:
+This avoids repeated startup parsing and makes the deployed stemming behavior explicit, reproducible, and versionable.
 
-- runtime parsing overhead
-- repeated compilation
-- startup latency
-
-
-
-## Example workflow
+## Example refinement workflow
 
 ```java
-// 1. Load bundled dictionary
-FrequencyTrie<String> base = StemmerPatchTrieLoader.load(
-        StemmerPatchTrieLoader.Language.US_UK_PROFI,
-        true,
-        ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
-);
+import java.io.IOException;
+import java.nio.file.Path;
 
-// 2. Modify (optional)
-FrequencyTrie.Builder<String> builder =
-        FrequencyTrieBuilders.copyOf(
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.FrequencyTrieBuilders;
+import org.egothor.stemmer.PatchCommandEncoder;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.ReductionSettings;
+import org.egothor.stemmer.StemmerPatchTrieBinaryIO;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+public final class BundledRefinementExample {
+
+    private BundledRefinementExample() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<String> base = StemmerPatchTrieLoader.load(
+                StemmerPatchTrieLoader.Language.US_UK,
+                true,
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+
+        final FrequencyTrie.Builder<String> builder = FrequencyTrieBuilders.copyOf(
                 base,
                 String[]::new,
                 ReductionSettings.withDefaults(
-                        ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
-                )
-        );
+                        ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS));
 
-builder.put("microservices", PatchCommandEncoder.NOOP_PATCH);
+        final PatchCommandEncoder encoder = PatchCommandEncoder.builder()
+                .traversalDirection(base.traversalDirection())
+                .build();
 
-// 3. Compile
-FrequencyTrie<String> compiled = builder.build();
+        builder.put("microservices", encoder.encode("microservices", "microservice"));
+
+        final FrequencyTrie<String> compiled = builder.build();
 
-// 4. Save
         StemmerPatchTrieBinaryIO.write(compiled, Path.of("english-custom.radixor.gz"));
+    }
+}
 ```
 
+The reconstructed builder preserves the traversal direction of the source trie, so refinements remain semantically aligned with the original bundled dictionary.
 
+## Extending language support
 
-## Limitations
+The built-in set is intentionally a practical baseline rather than a closed catalog. Additional languages, stronger lexical coverage, and improved dictionaries for currently supported languages are all natural extension paths.
 
-* bundled dictionaries are **general-purpose**
-* they may not reflect:
+What matters most is not only the number of entries, but the quality, consistency, maintainability, and operational usefulness of the lexical resource being added.
 
-  * domain-specific usage
-  * rare or specialized vocabulary
-  * organization-specific terminology
+## Related API surface
 
+The following types are typically involved when working with bundled dictionaries:
 
+- `StemmerPatchTrieLoader`
+- `StemmerPatchTrieLoader.Language`
+- `FrequencyTrie`
+- `PatchCommandEncoder`
+- `WordTraversalDirection`
+- `ReductionMode`
+- `ReductionSettings`
+- `StemmerPatchTrieBinaryIO`
+- `FrequencyTrieBuilders`
 
 ## Next steps
 
-* [Quick start](quick-start.md)
-* [Dictionary format](dictionary-format.md)
-* [CLI compilation](cli-compilation.md)
-* [Programmatic usage](programmatic-usage.md)
-
-
+- [Quick start](quick-start.md)
+- [Dictionary format](dictionary-format.md)
+- [CLI compilation](cli-compilation.md)
+- [Programmatic usage](programmatic-usage.md)
 
 ## Summary
 
-Radixor’s built-in language support provides:
-
-* immediate usability
-* reference datasets
-* a starting point for customization
-
-For production systems, they are best used as:
-
-* a baseline
-* a seed for further extension
-* a source for compiled deployment artifacts
-
+Radixor’s built-in language support provides immediate usability, a professionally defined baseline API, and a practical starting point for custom refinement. The bundled set now includes both left-to-right and right-to-left languages, and the library models that distinction explicitly through `WordTraversalDirection` so that trie construction, lookup, and patch application remain consistent.

docs/cli-compilation.md

@@ -1,305 +1,284 @@
 # CLI Compilation
 
-> ← Back to [README.md](../README.md)
+Radixor provides a command-line compiler for turning line-oriented dictionary files into compact binary stemmer artifacts.
 
-Radixor provides a command-line tool for compiling dictionary files into compact, production-ready binary stemmer tables.
+This is the preferred preparation workflow when stemming should run against an already compiled artifact rather than against raw dictionary input. The CLI reads the dictionary, derives patch commands, builds a mutable trie, applies the selected subtree reduction strategy, and writes the final compiled trie in the project binary format under GZip compression. The result is a deployment-ready `.radixor.gz` file that can be loaded directly by application code.
 
-This is the recommended workflow for deployment environments, as it separates:
+## What the CLI does
 
-- dictionary preparation (offline)
-- stemming execution (runtime)
-
-
-
-## Overview
-
-The `Compile` tool:
-
-1. reads a line-oriented dictionary file
-2. converts word–stem pairs into patch commands
-3. builds a trie structure
-4. applies subtree reduction
-5. writes a compressed binary artifact
-
-The output is a `.radixor.gz` file suitable for fast runtime loading.
+The `Compile` tool performs the following steps:
 
+1. reads the input dictionary in the standard Radixor stemmer format, accepting either plain UTF-8 text or GZip-compressed UTF-8 text,
+2. parses each line into a canonical stem column and its known variant columns,
+3. converts variants into patch commands,
+4. builds a mutable trie of patch-command values,
+5. applies the configured reduction mode,
+6. writes the compiled trie as a GZip-compressed binary artifact.
 
+This workflow is intentionally aligned with the same dictionary semantics used elsewhere in the library. Remarks introduced by `#` or `//` are supported through the shared dictionary parser.
 
 ## Basic usage
 
 ```bash
 java org.egothor.stemmer.Compile \
-  --input ./data/stemmer.txt \
+    --input ./data/stemmer.tsv \
     --output ./build/english.radixor.gz \
     --reduction-mode MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS \
+    --case-processing-mode LOWERCASE_WITH_LOCALE_ROOT \
     --store-original \
     --overwrite
 ```
 
+## Supported arguments
 
+The CLI supports the following arguments:
 
-## Required arguments
+```text
+--input <file>
+--output <file>
+--reduction-mode <mode>
+[--store-original]
+[--right-to-left]
+[--case-processing-mode <mode>]
+[--dominant-winner-min-percent <1..100>]
+[--dominant-winner-over-second-ratio <1..n>]
+[--overwrite]
+[--help]
+```
 
-### `--input`
+### `--input <file>`
 
 Path to the source dictionary file.
 
-* must be in the [dictionary format](dictionary-format.md)
-* must be readable
-* UTF-8 encoding is expected
-
-```
---input ./data/stemmer.txt
-```
-
-### `--output`
-
-Path to the output binary file.
-
-* parent directories are created automatically
-* output is written as **GZip-compressed binary**
-
-```
---output ./build/english.radixor.gz
-```
-
-
-
-## Optional arguments
-
-### `--reduction-mode`
-
-Controls how aggressively the trie is reduced during compilation.
-
-Available values:
-
-* `MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS`
-* `MERGE_SUBTREES_WITH_EQUIVALENT_UNORDERED_GET_ALL_RESULTS`
-* `MERGE_SUBTREES_WITH_EQUIVALENT_DOMINANT_GET_RESULTS`
+The file must use the standard line-oriented tab-separated values dictionary format, meaning that columns are separated by the tab character. Each non-empty logical line starts with the canonical stem column and may contain zero or more variant columns. The input may be plain UTF-8 text or GZip-compressed UTF-8 text; compression is detected from the stream header rather than the file extension. The parser processes case according to `CaseProcessingMode` (default: `LOWERCASE_WITH_LOCALE_ROOT`), ignores trailing remarks introduced by `#` or `//`, and currently ignores dictionary items containing embedded whitespace while reporting them through warning-level log entries.
 
 Example:
 
+```text
+--input ./data/stemmer.tsv
 ```
+
+### `--output <file>`
+
+Path to the output binary artifact.
+
+The output file is written as a GZip-compressed binary trie. Parent directories are created automatically when needed.
+
+Example:
+
+```text
+--output ./build/english.radixor.gz
+```
+
+### `--reduction-mode <mode>`
+
+Selects the subtree reduction strategy used during compilation.
+
+Supported values are:
+
+- `MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS`
+- `MERGE_SUBTREES_WITH_EQUIVALENT_UNORDERED_GET_ALL_RESULTS`
+- `MERGE_SUBTREES_WITH_EQUIVALENT_DOMINANT_GET_RESULTS`
+
+Example:
+
+```text
 --reduction-mode MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
 ```
 
-#### Recommendation
-
-Use:
-
-```
-MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
-```
-
-This provides:
-
-* safe behavior
-* deterministic ordering
-* good compression
-
-
+This argument is required.
 
 ### `--store-original`
 
-Stores the stem itself as a no-op mapping.
+When this flag is present, the canonical stem itself is inserted using the no-op patch command.
 
-```
+```text
 --store-original
 ```
 
-Effect:
+This is usually a sensible default for real dictionaries because it ensures that canonical forms are directly representable in the compiled trie rather than relying only on their variants.
 
-* ensures that canonical forms are always resolvable
-* improves robustness in real-world inputs
+### `--right-to-left`
 
-Recommended for most use cases.
+When present, compilation uses forward traversal (`WordTraversalDirection.FORWARD`) so stored forms are processed from their logical beginning.
 
+```text
+--right-to-left
+```
 
+This option is intended for right-to-left languages where affix behavior should operate on the written form without externally reversing words.
+
+### `--case-processing-mode <mode>`
+
+Controls dictionary key normalization during compilation and lookup. The setting is stored in persisted trie metadata and is therefore available to runtime lookup after binary loading.
+
+Supported values are:
+
+- `LOWERCASE_WITH_LOCALE_ROOT` (default)
+- `AS_IS`
+
+Example:
+
+```text
+--case-processing-mode AS_IS
+```
+
+### `--dominant-winner-min-percent <1..100>`
+
+Sets the minimum winner percentage used by dominant-result reduction settings.
+
+Example:
+
+```text
+--dominant-winner-min-percent 75
+```
+
+This option matters primarily when `--reduction-mode` is `MERGE_SUBTREES_WITH_EQUIVALENT_DOMINANT_GET_RESULTS`. The default value is `75`.
+
+### `--dominant-winner-over-second-ratio <1..n>`
+
+Sets the minimum winner-over-second ratio used by dominant-result reduction settings.
+
+Example:
+
+```text
+--dominant-winner-over-second-ratio 3
+```
+
+This option also matters primarily for `MERGE_SUBTREES_WITH_EQUIVALENT_DOMINANT_GET_RESULTS`. The default value is `3`.
 
 ### `--overwrite`
 
-Allows overwriting an existing output file.
+Allows the CLI to replace an already existing output file.
 
-```
+```text
 --overwrite
 ```
 
-Without this flag:
+Without this flag, compilation fails when the output path already exists.
 
-* compilation fails if the output file already exists
+### `--help`
 
+Prints usage help and exits successfully.
 
+```text
+--help
+```
 
-## Reduction strategy explained
+The short form `-h` is also supported.
 
-Reduction merges semantically equivalent subtrees to reduce memory and file size.
+## Reduction modes in practice
 
-Trade-offs:
+Reduction mode is not only a storage decision. It also influences what semantics are preserved when the mutable trie is compiled into its canonical read-only form.
 
-| Mode      | Compression | Behavioral fidelity |
-| --------- | ----------- | ------------------- |
-| Ranked    | Medium      | High                |
-| Unordered | High        | Medium              |
-| Dominant  | Highest     | Lower (heuristic)   |
+Before the selected public reduction mode is applied, compilation performs uniform-subtree
+contraction. If all reachable entries below a subtree select the same preferred patch command, the
+compiler stores that subtree as an accepting leaf and removes the deeper branches. This reduces
+runtime lookup depth without changing the preferred result returned by the standard stemming path.
 
-### Ranked (recommended)
+### Ranked `getAll()` equivalence
 
-* preserves full `getAll()` ordering
-* safest and most predictable
+`MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS` merges subtrees whose `getAll()` results remain equivalent for every reachable key suffix and whose local result ordering is the same.
 
-### Unordered
+This is the best general-purpose choice when result ordering and ambiguity handling matter. It preserves ranked multi-result semantics while still achieving useful structural reduction.
 
-* ignores ordering differences
-* higher compression, but less precise semantics
+This is the recommended default for most users.
 
-### Dominant
+### Unordered `getAll()` equivalence
 
-* focuses on the most frequent result
-* useful when only `get()` is relevant
-* may lose secondary candidates
+`MERGE_SUBTREES_WITH_EQUIVALENT_UNORDERED_GET_ALL_RESULTS` also uses `getAll()`-level equivalence, but it ignores local ordering differences in addition to absolute frequencies.
 
+This can yield stronger reduction, but it also weakens the precision of ordered multi-result semantics.
 
+Choose this mode only when the application does not depend on the ordering of alternative results.
 
-## Output format
+### Dominant `get()` equivalence
 
-The compiled file:
+`MERGE_SUBTREES_WITH_EQUIVALENT_DOMINANT_GET_RESULTS` focuses on preserving preferred-result semantics for `get()`, subject to dominance thresholds.
 
-* is a binary representation of the trie
-* uses **GZip compression**
-* is optimized for:
+If a node does not satisfy the configured dominance constraints, compilation falls back to ranked `getAll()` semantics for that node to avoid unsafe over-reduction.
 
-  * fast loading
-  * minimal memory footprint
+This mode is most suitable when the application primarily consumes the preferred result and does not rely on preserving richer ambiguity information.
 
-Typical properties:
+## Recommended usage patterns
 
-* small file size
-* fast deserialization
-* no runtime preprocessing required
+### Use offline preparation
 
+The CLI is best used as a preparation step during packaging, deployment, or controlled artifact generation. This keeps compilation outside the runtime startup path and allows services to load only the finished binary trie.
 
+### Treat compiled files as versioned assets
+
+A `.radixor.gz` file should be handled as a versioned output artifact. It represents a specific dictionary state, a specific reduction mode, whether uniform-subtree contraction was used, and, where relevant, specific dominant-result thresholds.
+
+Compiled tries also persist a human-readable metadata block (`key=value` lines) that includes format version, traversal direction, RTL indicator, reduction mode, contraction flag, dominant thresholds, diacritic-processing mode, and case-processing mode. After decompression, you can inspect this block directly to identify what dictionary/trie configuration the artifact contains. The current CLI uses `DiacriticProcessingMode.AS_IS`; custom diacritic stripping is available through the programmatic builder and loader APIs rather than through a CLI flag.
+
+### Choose reduction mode deliberately
+
+The ranked `getAll()` mode is the safest default. The unordered and dominant modes should be chosen only when their trade-offs are acceptable for the consuming application.
+
+### Expect memory pressure during preparation, not runtime
+
+Compilation is usually a one-time step and is generally fast. The more important operational consideration is memory usage during preparation, because the dictionary-derived mutable structure exists before reduction compacts it into the final read-only trie. This is especially relevant for very large source dictionaries.
 
 ## Example workflow
 
-### 1. Prepare dictionary
+### 1. Prepare a dictionary
 
-```
+```text
 run	running	runs	ran
 connect	connected	connecting
 ```
 
-### 2. Compile
+### 2. Compile it
 
 ```bash
 java org.egothor.stemmer.Compile \
-  --input ./data/stemmer.txt \
+    --input ./data/stemmer.tsv \
     --output ./build/english.radixor.gz \
     --reduction-mode MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS \
     --store-original
 ```
 
-### 3. Use in application
+### 3. Load it in an application
 
 ```java
-FrequencyTrie<String> trie =
-    StemmerPatchTrieLoader.loadBinary("english.radixor.gz");
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+final FrequencyTrie<CompiledPatchCommand> trie =
+        StemmerPatchTrieLoader.loadBinaryCompiled("english.radixor.gz");
 ```
 
+## Exit codes and error handling
 
+The CLI uses three exit outcomes:
 
-## Error handling
+- `0` for success,
+- `1` for processing failures such as I/O or compilation errors,
+- `2` for invalid command-line usage.
 
-The CLI reports:
+When argument parsing fails, the CLI prints the error message, prints the usage summary, and exits with usage error status.
 
-* missing input file
-* invalid arguments
-* I/O failures
-* parsing errors
+When compilation fails during processing, the CLI prints a `Compilation failed: ...` message to standard error and exits with processing error status.
 
-Typical exit codes:
+Examples of failure conditions include:
 
-* `0` – success
-* non-zero – failure
-
-Error details are printed to standard error.
-
-
-
-## Performance considerations
-
-### Compilation
-
-* typically CPU-bound
-* depends on dictionary size and reduction mode
-
-### Output size
-
-* depends on:
-
-  * dictionary completeness
-  * reduction strategy
-* can vary significantly between modes
-
-### Runtime impact
-
-* compiled tries are optimized for:
-
-  * fast lookup
-  * low allocation
-  * predictable latency
-
-
-
-## Best practices
-
-### Use offline compilation
-
-* compile dictionaries during build or deployment
-* do not compile on application startup
-
-### Version your artifacts
-
-* treat `.radixor.gz` files as versioned assets
-* store them alongside application releases
-
-### Choose reduction mode deliberately
-
-* use **ranked** for correctness
-* use **dominant** only if you fully understand the trade-offs
-
-### Keep dictionaries clean
-
-* better input → better compiled output
-* avoid noise and inconsistencies
-
-
-
-## Integration tips
-
-* store compiled files under `resources/` or a dedicated directory
-* load them once and reuse the trie instance
-* avoid repeated loading in frequently executed code paths (for example, per-request processing)
+- missing required arguments,
+- unknown arguments,
+- invalid integer values for dominant thresholds,
+- missing input files,
+- unreadable input,
+- existing output file without `--overwrite`,
+- general I/O failures during reading or writing.
 
+## Relation to programmatic usage
 
+The CLI and the programmatic API implement the same conceptual preparation step. The CLI is the operationally convenient choice when you want a ready-made binary artifact. The programmatic API is the better fit when compilation must be integrated directly into custom Java workflows.
 
 ## Next steps
 
-* [Dictionary format](dictionary-format.md)
-* [Programmatic usage](programmatic-usage.md)
-* [Quick start](quick-start.md)
-
-
-
-## Summary
-
-The `Compile` CLI is the bridge between:
-
-* human-readable dictionary data
-* optimized runtime stemmer tables
-
-It enables a clean separation between:
-
-* data preparation
-* runtime execution
-
-and is the preferred way to prepare Radixor for production use.
+- [Dictionary format](dictionary-format.md)
+- [Quick start](quick-start.md)
+- [Programmatic usage](programmatic-usage.md)
+- [Architecture and reduction](architecture-and-reduction.md)

docs/compatibility-and-guarantees.md

@@ -0,0 +1,202 @@
+# Compatibility and Guarantees
+
+This document explains what Radixor treats as stable public behavior, what should be regarded as internal implementation detail, and how to think about compatibility across versions.
+
+Its purpose is to make adoption safer. Users should be able to understand which parts of the project are intended as supported API, which parts may evolve more freely, and which kinds of change are expected to remain compatible in future releases.
+
+## Compatibility philosophy
+
+Radixor is designed to be used as a real library, not only as a code drop. That means compatibility matters.
+
+At the same time, the project distinguishes clearly between:
+
+- **public API and behavior** that users are expected to build against,
+- **internal implementation layers** that may change more freely when needed for correctness, performance, or maintainability.
+
+The practical goal is straightforward:
+
+- keep the main user-facing API in `org.egothor.stemmer` stable and supportable,
+- allow more freedom of evolution in internal trie-focused implementation layers,
+- extend the project conservatively without creating unnecessary behavioral ambiguity.
+
+## Public API posture
+
+As a general rule, the `org.egothor.stemmer` package should be treated as the primary supported API surface.
+
+That includes the main user-facing types involved in:
+
+- dictionary loading,
+- binary loading and persistence,
+- patch-command application,
+- compiled trie querying,
+- reconstruction workflows,
+- reduction configuration,
+- CLI use.
+
+This API is expected to remain supportable across future versions. The preferred compatibility model is additive evolution: improving documentation, clarifying behavior, and adding capabilities without unnecessary disruption of existing usage patterns.
+
+Examples of likely additive evolution include:
+
+- additional bundled language resources,
+- fuller support for diacritics or native-script language resources,
+- expanded documentation and operational tooling,
+- new convenience methods that do not break existing code.
+
+## Internal API posture
+
+The `org.egothor.stemmer.trie` package should be treated as internal or at least significantly less stable implementation API.
+
+It represents the structural machinery behind mutable nodes, reduced nodes, compiled nodes, reduction context, signatures, and related internal compilation details. These types may evolve more aggressively when needed to improve implementation quality, correctness, reduction behavior, internal representations, or performance characteristics.
+
+Users should therefore avoid building long-term integrations against `org.egothor.stemmer.trie` unless they are intentionally accepting that tighter coupling.
+
+In practical terms:
+
+- `org.egothor.stemmer` is the supported integration layer,
+- `org.egothor.stemmer.trie` is the implementation layer.
+
+## Behavioral guarantees
+
+Several project properties are intended as core behavioral guarantees.
+
+### Deterministic dictionary loading and compilation
+
+Given the same textual dictionary input and the same reduction settings, Radixor is intended to produce the same compiled stemming semantics in a reproducible way.
+
+This includes deterministic local result ordering and deterministic observable lookup behavior.
+
+### Stable meaning of `get()` and `getAll()`
+
+The distinction between preferred-result lookup and multi-result lookup is part of the supported behavior model.
+
+- `get()` returns the locally preferred stored value,
+- `getAll()` returns all locally stored values in deterministic ranked order,
+- `getEntries()` returns aligned values with counts.
+
+That model is part of how the public API should be understood.
+
+Visitor lookup methods such as `getAllNormalized(..., EntrySink, maxResults)` are additive hot-path APIs. They expose the same local ordering and count semantics without allocating result containers, but they do not replace `get()`, `getAll()`, or `getEntries()`.
+
+Compiled `FrequencyTrie` instances are immutable and thread-safe for concurrent reads. Visitor sinks are caller-owned and are not retained by the trie. Stored values passed to sinks are the model-owned trie values; for `FrequencyTrie<String>` patch tries, those patch strings are immutable stored strings rather than fresh per-result strings.
+
+### Stable patch application behavior
+
+Serialized patch-command strings remain the stable stored representation used by textual dictionaries and binary artifacts. Runtime stemming should use `CompiledPatchCommand` values produced by `StemmerPatchTrieLoader.loadCompiled(...)`, `StemmerPatchTrieLoader.loadBinaryCompiled(...)`, or `PatchCommandEncoder.compile(...)`.
+
+The historical `PatchCommandEncoder.apply(...)` and String-based `applyTo(...)` overloads remain compatibility APIs during the 2.x transition, but they are deprecated because they reparse the patch-command string on each application. See [Migration and Backward Compatibility](migration-and-backward-compatibility.md) for old and new code examples.
+
+Compiled buffer-oriented `CompiledPatchCommand.applyTo(...)` overloads use caller-owned output storage. They do not retain output arrays and report insufficient capacity with `CompiledPatchCommand.APPLY_INSUFFICIENT_CAPACITY`.
+
+### Stable reduction-mode intent
+
+Each public `ReductionMode` constant carries a semantic contract that should remain meaningful across versions.
+
+In other words, the implementation may evolve, but the intended meaning of modes such as ranked `getAll()` equivalence, unordered `getAll()` equivalence, and dominant `get()` equivalence should not drift casually.
+
+Internal pre-reduction optimizations may still change the physical compiled trie shape when they
+preserve the documented lookup contract. Uniform-subtree contraction is one such optimization: it
+can replace a subtree with an accepting leaf when all reachable entries choose the same preferred
+patch command.
+
+### Stable binary artifact purpose
+
+Compiled `.radixor.gz` artifacts are a first-class project output. Loading and persisting compiled stemmer artifacts is part of the intended usage model, not an incidental implementation side effect.
+
+## What is allowed to evolve
+
+Compatibility does not mean the project is frozen.
+
+The following kinds of change are generally compatible with the project’s direction:
+
+- improved internal data structures,
+- changes inside `org.egothor.stemmer.trie`,
+- expanded bundled dictionaries,
+- additional supported languages,
+- improved native-script handling,
+- better benchmarks, tests, and reports,
+- additive public API growth that does not invalidate existing usage.
+
+The project should be able to improve substantially while keeping the main user-facing integration model intact.
+
+## What may change more cautiously
+
+Some areas should be treated as stable in intent but still approached carefully when changed.
+
+### Bundled dictionary contents
+
+Bundled resources are versioned project data, not immutable language standards. Their contents may improve over time.
+
+That means stemming outcomes can legitimately change when bundled dictionaries are refined or expanded. Such changes are compatible with the project’s direction, but they should still be understood as behavior changes at the lexical-resource level.
+
+### Binary format evolution
+
+Compiled binary artifacts are an intended project output, but binary-format evolution may still be needed in future versions.
+
+If the format changes, that should be handled deliberately and documented clearly. Users should not assume that every historical persisted artifact will remain readable forever without versioning considerations. What should remain stable is the project’s support for compiled artifact workflows, not necessarily perpetual cross-version binary interchange without explicit format evolution rules.
+
+### Performance characteristics
+
+Radixor places strong emphasis on performance, but no benchmark number should be treated as a formal compatibility guarantee.
+
+What is more meaningful than any single raw number is the architectural performance posture: the library is intended to remain a compact compiled stemmer with very strong runtime throughput characteristics.
+
+## What users should rely on
+
+Long-term users should rely primarily on the following:
+
+- the main integration path in `org.egothor.stemmer`,
+- the documented meaning of `get()`, `getAll()`, and reduction modes,
+- the offline-compilation plus runtime-loading workflow,
+- the availability of compiled artifact support,
+- the project’s preference for deterministic and auditable behavior.
+
+These are the parts of the project that are intended to remain the most stable and supportable.
+
+## What users should not rely on casually
+
+Users should avoid depending on:
+
+- internal trie package details,
+- undocumented internal classes or intermediate representations,
+- incidental internal ordering outside documented lookup semantics,
+- assumptions that bundled dictionary contents will never evolve,
+- assumptions that internal binary-format details are frozen forever.
+
+If a behavior is important to your integration, it should ideally be documented at the public API or project-documentation level rather than inferred from internal implementation details.
+
+## Source compatibility and behavioral compatibility
+
+It is useful to distinguish two different notions of compatibility.
+
+### Source compatibility
+
+Whether existing Java code using the supported public API still compiles and integrates cleanly after an upgrade.
+
+### Behavioral compatibility
+
+Whether the upgraded system still behaves the same way for the same dictionary data, compiled artifacts, and runtime calls.
+
+Radixor aims to preserve both where reasonably possible, but behavioral compatibility can still be influenced by intentional improvements such as dictionary refinement or bug fixes. For that reason, upgrades should be evaluated not only as code upgrades but also as stemming-behavior upgrades.
+
+## Recommended upgrade discipline
+
+When upgrading Radixor in a production environment, it is good practice to:
+
+1. review release notes and documentation changes,
+2. rebuild compiled artifacts if the upgrade affects dictionary or artifact handling,
+3. rerun representative stemming validation tests,
+4. compare benchmark outputs where performance matters,
+5. inspect whether bundled-dictionary changes affect expected canonical results.
+
+This is especially important for deployments that treat stemming behavior as part of search relevance or normalization policy.
+
+## Summary
+
+Radixor’s compatibility model is intentionally layered.
+
+- `org.egothor.stemmer` should be treated as the supported public integration API,
+- `org.egothor.stemmer.trie` should be treated as an internal implementation layer,
+- deterministic public behavior and compiled-artifact workflows are core project commitments,
+- internal structure and lexical-resource quality can continue to evolve.
+
+This model gives the project room to improve while still providing a reliable surface for long-term use.

docs/contributing-dictionaries.md

@@ -0,0 +1,228 @@
+# Contributing Dictionaries
+
+High-quality dictionaries are one of the most valuable ways to improve **Radixor**.
+
+The project already includes practical bundled dictionaries for common use, but the long-term quality and language reach of the stemmer depend heavily on the quality of its lexical resources. Contributions are therefore welcome not only in the form of code changes, but also in the form of well-prepared dictionary data for existing or additional languages.
+
+This document explains what makes a dictionary contribution useful, how to structure it, and how to prepare it so that it integrates cleanly with the project.
+
+## What a good dictionary contribution looks like
+
+A good dictionary contribution is not defined only by the number of entries.
+
+The most useful contributions are dictionaries that are:
+
+- linguistically consistent,
+- operationally clean,
+- easy to review,
+- easy to reproduce,
+- appropriate for actual stemming use rather than raw lexical accumulation.
+
+In practice, dictionary quality matters more than dictionary size. A smaller but coherent and carefully normalized dictionary is often more valuable than a larger resource that mixes conventions, contains noisy forms, or introduces accidental ambiguity.
+
+## Preferred dictionary shape
+
+Radixor uses a simple line-oriented tab-separated values format, meaning that columns are separated by the tab character:
+
+```text
+<stem>	<variant1>	<variant2>	<variant3> ...
+```
+
+The first column on a line is the canonical stem. All following tab-separated columns on that line are known variants that should reduce to that stem.
+
+Example:
+
+```text
+run	running	runs	ran
+connect	connected	connecting	connection
+```
+
+The parser:
+
+- reads UTF-8 text,
+- interprets each line as tab-separated values,
+- applies configurable case processing through `CaseProcessingMode` (default: `LOWERCASE_WITH_LOCALE_ROOT`),
+- ignores empty lines,
+- supports remarks introduced by `#` or `//`,
+- currently ignores dictionary items containing embedded whitespace and reports them through warning-level log entries.
+
+For full format details, see [Dictionary format](dictionary-format.md).
+
+## Contribution priorities
+
+The most useful dictionary contributions generally fall into one of four categories.
+
+### 1. Stronger dictionaries for already bundled languages
+
+Improving lexical quality for already supported languages is often more valuable than merely expanding the language list. Better coverage, cleaner canonicalization, and improved consistency directly improve practical stemming outcomes.
+
+### 2. Additional languages
+
+New language support is welcome when the submitted resource is strong enough to be useful as a maintainable bundled baseline rather than as an incomplete demonstration artifact.
+
+### 3. Native-script language resources
+
+The current bundled resources follow a pragmatic normalization convention and may use transliterated or otherwise normalized forms. This is especially visible for languages such as Russian.
+
+That convention belongs to the supplied dictionaries, not to the underlying algorithm. The parser, trie, and patch-command model are not fundamentally restricted to plain ASCII. Contributions of high-quality native-script dictionaries in full UTF-8 text are therefore particularly valuable, because they would enable more direct language support without transliteration-based workflows.
+
+### 4. Domain-quality refinements
+
+Some contributions may be more appropriate as curated domain extensions than as replacements for a general-purpose bundled dictionary. These are still useful when they are clearly scoped and operationally coherent.
+
+## Normalization guidance
+
+A dictionary should follow one normalization convention consistently.
+
+For current general-purpose bundled resources, the safest convention remains normalized plain-ASCII lexical input where that is already the established project style. For languages where a stronger native-script resource exists, a coherent UTF-8 dictionary may be preferable, provided that the contribution is deliberate, well-structured, and consistently normalized.
+
+The important point is not to mix incompatible conventions casually.
+
+Avoid contributions that combine, without clear design intent:
+
+- native-script and transliterated forms,
+- multiple incompatible stem conventions,
+- inconsistent use of diacritics,
+- ad hoc spelling normalization,
+- noisy typo-like forms presented as ordinary lexical variants.
+
+## Choosing canonical stems
+
+A dictionary line should reflect a stable canonical target form.
+
+That means:
+
+- choose one canonical representation and use it consistently,
+- avoid mixing alternative stem conventions without a clear lexical reason,
+- keep variants grouped under the form that the project should actually return as the canonical result.
+
+For example, the following is coherent:
+
+```text
+analyze analyzing analyzed analyzes
+```
+
+The following is less useful if the project has not intentionally chosen mixed conventions:
+
+```text
+analyse analyzing analyzed analyzes
+```
+
+The contribution should make the intended canonical policy easy to understand.
+
+## Ambiguity handling
+
+Ambiguity is allowed, but it should be intentional.
+
+If the same surface form appears under multiple stems, the compiled trie may later expose multiple candidate patch commands. This can be correct and desirable when the lexical reality genuinely requires it. However, accidental ambiguity caused by inconsistent source preparation makes the resource harder to trust and harder to review.
+
+Before contributing a dictionary, check whether repeated surface forms across lines are:
+
+- linguistically intentional,
+- consistent with the chosen canonical policy,
+- useful for runtime stemming behavior.
+
+## What to avoid
+
+Dictionary contributions are much easier to review and accept when they avoid common quality problems.
+
+Avoid:
+
+- mechanically aggregated word lists without review,
+- inconsistent canonical forms,
+- mixed orthographic conventions without explanation,
+- accidental duplicates caused by source merging,
+- noisy or non-lexical tokens,
+- comments or formatting that make the source hard to audit.
+
+A dictionary should read like a curated lexical resource, not like an unfiltered export.
+
+## Practical preparation workflow
+
+A disciplined dictionary contribution should typically follow this path:
+
+1. prepare or normalize the lexical source,
+2. convert it into Radixor dictionary format,
+3. review canonical stem choices,
+4. check for accidental duplicates and unintended ambiguity,
+5. compile the dictionary,
+6. test representative lookups,
+7. inspect `get()` and `getAll()` behavior for important edge cases,
+8. include a concise explanation of source provenance and normalization choices.
+
+## What to test before submitting
+
+At minimum, a proposed dictionary should be checked for:
+
+- successful parsing,
+- successful compilation,
+- expected stemming behavior on representative examples,
+- acceptable ambiguity behavior,
+- stable canonical policy,
+- absence of obvious malformed lines or accidental source contamination.
+
+For important resources, it is also useful to test:
+
+- whether representative forms survive reduction as expected,
+- whether dominant-result behavior remains sensible if alternate reduction modes are used,
+- whether the resulting artifact has a practical size for the intended use case.
+
+## Contribution notes that help maintainers
+
+A dictionary contribution becomes much easier to review when it includes a short maintainer-facing note describing:
+
+- the language or domain covered,
+- the provenance of the lexical data,
+- the normalization convention used,
+- whether the dictionary is ASCII-normalized or native-script UTF-8,
+- the intended canonical stem policy,
+- any known limitations,
+- why the contribution improves the project in practical terms.
+
+This note does not need to be long. It simply needs to make the resource intelligible.
+
+## Bundled-resource expectations
+
+Not every useful dictionary must automatically become a bundled language resource.
+
+To be suitable for bundling, a dictionary should generally be:
+
+- broadly useful,
+- maintainable,
+- legally safe to include,
+- coherent enough to serve as a project baseline,
+- strong enough that users can rely on it as more than a demonstration resource.
+
+Some dictionaries are better treated as examples, experiments, or domain-specific artifacts rather than as general built-in resources.
+
+## Native scripts and future language support
+
+One of the most meaningful future directions for the project is stronger support for languages in their native writing systems.
+
+The architecture does not need to change fundamentally for that to happen. What matters is the availability of strong lexical resources and the willingness to define clear conventions for how those resources should be bundled and maintained.
+
+Contributions in this area are therefore especially valuable when they are:
+
+- internally consistent,
+- encoded as proper UTF-8 text,
+- accompanied by a clear explanation of normalization assumptions,
+- strong enough to support practical use rather than only demonstration.
+
+## Related documentation
+
+- [Built-in languages](built-in-languages.md)
+- [Dictionary format](dictionary-format.md)
+- [CLI compilation](cli-compilation.md)
+- [Programmatic usage](programmatic-usage.md)
+
+## Summary
+
+The best dictionary contributions improve Radixor not merely by adding more entries, but by improving the linguistic quality, consistency, and practical usefulness of the lexical resources the project can compile and ship.
+
+A strong contribution is therefore one that is:
+
+- coherent,
+- reviewable,
+- operationally clean,
+- well explained,
+- and valuable for real stemming workloads.

docs/dictionary-format.md

@@ -1,255 +1,237 @@
 # Dictionary Format
 
-> ← Back to [README.md](../README.md)
+Radixor uses a simple line-oriented dictionary format designed for practical stemming workflows. The textual source format is tab-separated values, meaning that columns are separated by the tab character.
 
-Radixor uses a simple, line-oriented dictionary format to define mappings between **word forms** and their **canonical stems**.
+Each logical line describes one canonical stem and zero or more known word variants that should reduce to that stem. The format is intentionally lightweight, easy to maintain in source control, and directly consumable both by the programmatic loader and by the CLI compiler.
 
-This format is intentionally minimal, language-agnostic, and easy to generate from existing linguistic resources or corpora.
+## Core structure
 
-## Overview
+Each non-empty logical line has the following shape:
 
-Each logical line defines:
-
-- one **canonical stem**
-- zero or more **word variants** belonging to that stem
-
-```
-stem variant1 variant2 variant3 ...
+```text
+<stem>	<variant1>	<variant2>	<variant3> ...
 ```
 
-At compile time:
+The first column is interpreted as the **canonical stem**. Every following token on the same line is interpreted as a **known variant** belonging to that stem.
 
-- each variant is converted into a **patch command** transforming the variant into the stem
-- the stem itself may optionally be stored as a **no-op mapping**
+Example:
 
-## Basic example
-
-```
+```text
 run	running	runs	ran
 connect	connected	connecting	connection
-analyze analyzing analysed analyses
 ```
 
-This defines:
+In this example:
 
-| Stem     | Variants                              |
-|----------|----------------------------------------|
-| run      | running, runs, ran                     |
-| connect  | connected, connecting, connection      |
-| analyze  | analyzing, analysed, analyses          |
+- `run` is the canonical stem for `running`, `runs`, and `ran`,
+- `connect` is the canonical stem for `connected`, `connecting`, and `connection`.
 
-## Syntax rules
+## How the loader interprets a line
 
-### 1. Tokenization
+When a dictionary is loaded through `StemmerPatchTrieLoader`, the loader processes each parsed line as follows:
 
-- Tokens are separated by **whitespace**
-- Multiple spaces and tabs are treated as a single separator
-- Leading and trailing whitespace is ignored
+1. the first column becomes the canonical stem,
+2. every following token is treated as a variant,
+3. each variant is converted into a patch command that transforms the variant into the stem,
+4. if `storeOriginal` is enabled, the stem itself is also inserted using the canonical no-op patch command.
 
-### 2. First token is the stem
+This means the textual dictionary is not used directly at runtime. Instead, it is transformed into patch-command data and compiled into a reduced read-only trie.
 
-- The **first token** on each line is always the canonical stem
-- All following tokens are treated as variants of that stem
+## Minimal valid lines
 
-### 3. Case normalization
+A line may consist of the stem only:
 
-- All input is normalized to **lowercase using `Locale.ROOT`**
-- Dictionaries should ideally already be lowercase to avoid ambiguity
+```text
+run
+```
 
-### 4. Empty lines
+This is syntactically valid. It defines a stem entry with no explicit variants on that line.
 
-- Empty lines are ignored
+Whether such a line is operationally useful depends on how the dictionary is loaded:
 
-### 5. Duplicate variants
+- if `storeOriginal` is enabled, the stem itself is inserted as a no-op mapping,
+- if `storeOriginal` is disabled, the line contributes no explicit variant mappings.
 
-- Duplicate variants are allowed but have no additional effect
-- Frequency is determined by occurrence across the entire dataset
+## Column and whitespace rules
 
-## Remarks (comments)
+Columns are separated by the tab character. Leading and trailing whitespace around each column is ignored.
+
+This is the canonical form:
+
+```text
+run	running	runs	ran
+```
+
+This is also accepted because the surrounding padding is removed before the item is processed:
+
+```text
+  run	 running 	 runs	 ran  
+```
+
+Embedded whitespace inside one dictionary item is currently not supported. A stem or variant such as `new york` therefore cannot yet be represented as one usable dictionary item in the textual source format. Such items are ignored during parsing and reported through a warning-level log entry together with the physical line number, the stem, and the ignored items from that line.
+
+## Empty lines
+
+Empty lines are ignored.
+
+Example:
+
+```text
+run	running	runs	ran
+
+connect	connected	connecting
+```
+
+The blank line between entries has no effect.
+
+## Remarks and comments
 
 The parser supports both full-line and trailing remarks.
 
-### Supported remark markers
+Two remark markers are recognized:
 
 - `#`
 - `//`
 
-### Examples
+The earliest occurrence of either marker terminates the logical content of the line, and the remainder of that line is ignored.
 
-```
+Examples:
+
+```text
 run	running	runs	ran # English verb forms
-connect connected connecting  // basic forms
+connect	connected	connecting // Common derived forms
 ```
 
-Everything after the first occurrence of a remark marker is ignored.
+This is also valid:
 
-### Important note
-
-Remark markers are not escaped. If `#` or `//` appear in a token, they will terminate the line.
-
-## Storing the original form
-
-When compiling, you may enable:
-
-```
---store-original
+```text
+# This line is ignored completely
+// This line is also ignored completely
 ```
 
-This causes the stem itself to be stored using a **no-op patch command**.
+## Case normalization
+
+Input-line case normalization is controlled by `CaseProcessingMode`; by default the parser uses `LOWERCASE_WITH_LOCALE_ROOT` before tab-separated columns are processed into dictionary entries.
+
+That means dictionary authors should treat the format as **case-insensitive at load time**. If a file contains uppercase or mixed-case tokens, they will be normalized during parsing.
 
 Example:
 
-```
-run running runs
+```text
+Run	Running	Runs	Ran
 ```
 
-With `--store-original`, this implicitly includes:
+is processed the same way as:
 
-```
-run -> run
+```text
+run	running	runs	ran
 ```
 
-This is useful when:
+## Character set, compression, and normalization
 
-- the input may already be normalized
-- you want stable identity mappings
-- you want to avoid missing entries for canonical forms
+Dictionary files are read as UTF-8 text. Files loaded through `StemmerPatchTrieLoader.load(Path, ...)` may be either plain UTF-8 text or GZip-compressed UTF-8 text; the loader detects GZip input from the stream header instead of relying on the file extension. Bundled dictionaries are stored as GZip resources and are decoded as UTF-8 after decompression.
 
-## Frequency and ordering
+The parser and trie are not restricted to ASCII. Dictionary items are ordinary Java `String` values, and trie traversal works over Java `char` sequences. This supports Latin-script data with diacritics, Cyrillic data, Hebrew, Persian, Yiddish, and other scripts represented in UTF-8, subject to the normal Java `String` model and the project’s traversal configuration.
 
-Radixor tracks **local frequencies** of values.
+Case normalization is controlled by `CaseProcessingMode`. The default `LOWERCASE_WITH_LOCALE_ROOT` mode lowercases the line before columns are split into dictionary items. `AS_IS` preserves the original casing.
 
-Frequency is determined by:
+Diacritic normalization is controlled at trie-build and lookup time by `DiacriticProcessingMode`:
 
-- how many times a mapping appears during construction
-- merging behavior during reduction
+- `AS_IS` preserves dictionary and lookup keys exactly after case handling,
+- `REMOVE` strips supported diacritics and common Latin ligatures on both insertion and lookup paths,
+- `AS_IS_AND_STRIPPED_FALLBACK` is declared in the public model but is not implemented yet and raises `UnsupportedOperationException`.
 
-When multiple stems exist for a word:
+For reliable production behavior, choose one normalization policy deliberately and apply it consistently. Normalized ASCII dictionaries remain a practical convention for some legacy stemming data, but they are not a format requirement.
 
-- results are ordered by **descending frequency**
-- ties are resolved deterministically:
-  1. shorter textual representation wins
-  2. lexicographically smaller value wins
-  3. earlier insertion order wins
+## Distinct stem and variant semantics
 
-This guarantees **stable and reproducible results**.
+The format expresses a one-line grouping of forms under a canonical stem. It does not encode linguistic metadata, part-of-speech information, weights, or explicit ambiguity markers.
 
-## Ambiguity and multiple stems
+For example:
 
-A word may legitimately map to more than one stem:
-
-```
-axes ax axe
+```text
+axis	axes
+axe	axes
 ```
 
-This allows Radixor to represent ambiguity explicitly.
+These are simply two independent lines. If both contribute mappings for the same surface form, the compiled trie may later expose one or more candidate patch commands depending on the accumulated local counts and the selected reduction mode.
 
-At runtime:
+In other words, the dictionary format itself is deliberately simple. Richer behavior such as preferred-result ranking or multiple candidate results emerges during trie construction and reduction rather than through extra syntax in the dictionary file.
 
-- `get(word)` returns the **preferred result**
-- `getAll(word)` returns **all candidates**
+## Duplicate forms and repeated entries
 
-## Design guidelines
+The format does not reserve any special syntax for duplicates. If the same mapping is inserted multiple times through repeated dictionary content, the builder accumulates local counts for the stored value at the addressed key.
 
-### Keep stems consistent
+This matters because compiled tries preserve local value frequencies and use them to determine preferred ordering for `get(...)`, `getAll(...)`, and `getEntries(...)`.
 
-Use a single canonical form:
+As a result, repeating the same mapping is not just redundant text. It can influence the ranking behavior of the compiled trie.
 
-- `run` instead of mixing `run` / `running`
-- `analyze` vs `analyse` — pick one convention
+## Practical examples
 
-### Avoid noise
+### Simple English example
 
-Do not include:
-
-- typos
-- extremely rare forms (unless required)
-- inconsistent normalization
-
-### Prefer completeness over clever rules
-
-Radixor is data-driven:
-
-- more complete dictionaries → better results
-- no hidden rule system compensates for missing entries
-
-### Handle domain-specific vocabulary
-
-You can extend dictionaries with:
-
-- product names
-- technical terms
-- organization-specific terminology
-
-## Example: minimal dictionary
-
-```
-go goes going went
-be is are was were being
-have has having had
+```text
+run	running	runs	ran
+connect	connected	connecting	connection
+build	building	builds	built
 ```
 
-## Example: domain-specific extension
+### Dictionary with remarks
 
-```
-microservice microservices
-container containers containerized
-kubernetes kubernetes
+```text
+run	running	runs	ran # canonical verb family
+connect	connected	connecting // derived forms
+build	building	builds	built
 ```
 
-## Common pitfalls
+### Stem-only entries
 
-### Mixing cases
-
-```
-Run running Runs   ❌
+```text
+run
+connect	connected	connecting
+build
 ```
 
-→ normalized to lowercase, but inconsistent input is error-prone
+### Mixed case input
 
-### Multiple stems on one line
-
-```
-run running connect   ❌
+```text
+Run	Running	Runs	Ran
+CONNECT	Connected	Connecting
 ```
 
-→ `connect` becomes a variant of `run`, which is incorrect
+This is accepted. Under the default `LOWERCASE_WITH_LOCALE_ROOT` mode it is normalized to lower case during parsing; under `AS_IS` it is preserved.
 
-### Hidden comments
+## Format limitations
 
-```
-run running //comment runs   ❌
-```
+The current dictionary format intentionally stays minimal:
 
-→ everything after `//` is ignored
+- no quoted tokens,
+- no escaping rules,
+- no multi-word entries,
+- no inline weighting syntax,
+- no explicit ambiguity syntax,
+- no sectioning or nested structure.
 
-## When to use this format
+Each dictionary item is simply one tab-separated word form after remark stripping and the configured case and diacritic normalization.
 
-This format is suitable for:
+## Authoring guidance
 
-- curated linguistic datasets
-- exported morphological dictionaries
-- domain-specific vocabularies
-- generated `(word, stem)` pairs from corpora
+For reliable results, keep dictionaries:
 
-## Next steps
+- consistent in normalization,
+- free of accidental duplicates unless repeated weighting is intentional,
+- focused on meaningful stem-to-variant groupings,
+- encoded in UTF-8,
+- easy to audit in plain text form.
+
+For most deployments, it is sensible to choose either preserved UTF-8 forms or a normalized ASCII/diacritic-stripped convention and keep that choice consistent across dictionary authoring, compilation, and runtime lookup.
+
+## Relationship to other documentation
+
+This page describes only the textual source format.
+
+To understand how those dictionary lines are transformed into compiled runtime artifacts, continue with:
 
 - [CLI compilation](cli-compilation.md)
 - [Programmatic usage](programmatic-usage.md)
-- [Quick start](quick-start.md)
-
-## Summary
-
-Radixor dictionaries are intentionally simple:
-
-- one line per stem
-- whitespace-separated tokens
-- optional remarks
-- no embedded rules
-
-This simplicity enables:
-
-- easy generation
-- fast parsing
-- deterministic behavior
-- efficient compilation into compact patch-command tries
+- [Architecture and reduction](architecture-and-reduction.md)

docs/fast-track.md

@@ -0,0 +1,130 @@
+# Fast Track
+
+This page is the shortest path from an empty Java project to a working Radixor stemmer.
+It deliberately uses a bundled dictionary and the preferred compiled-command runtime API, so the
+first result does not require writing a dictionary, running the CLI compiler, or understanding
+reduction internals.
+
+Use this page when the goal is:
+
+- add the dependency,
+- load a bundled language resource,
+- stem a token,
+- know where to go next.
+
+For deeper production guidance, see [Integration Deep Dive](integration-deep-dive.md).
+
+## 1. Add The Dependency
+
+Radixor is published as:
+
+```text
+groupId:    org.egothor
+artifactId: radixor
+```
+
+Use the current published version from Maven Central. The snippets below use `3.0.0`; replace it
+with the version you deploy if a newer release is available.
+
+For a Gradle project:
+
+```kotlin
+dependencies {
+    implementation("org.egothor:radixor:3.0.0")
+}
+```
+
+For a Maven project:
+
+```xml
+<dependency>
+    <groupId>org.egothor</groupId>
+    <artifactId>radixor</artifactId>
+    <version>3.0.0</version>
+</dependency>
+```
+
+Radixor targets modern Java and has a dependency-light runtime core. The project documentation and
+benchmarks assume a current JDK; Java 21 or newer is the practical baseline for current releases.
+
+## 2. Load A Bundled Dictionary
+
+The fastest path is to use a bundled dictionary through `StemmerPatchTrieLoader.Language`.
+This example uses the bundled English resource, `US_UK`.
+
+```java
+import java.io.IOException;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+public final class RadixorFirstStem {
+
+    private RadixorFirstStem() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<CompiledPatchCommand> stemmer = StemmerPatchTrieLoader.loadCompiled(
+                StemmerPatchTrieLoader.Language.US_UK,
+                true,
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+
+        final String token = "running";
+        final CompiledPatchCommand command = stemmer.get(token);
+        final String stem = command == null ? token : command.apply(token);
+
+        System.out.println(token + " -> " + stem);
+    }
+}
+```
+
+The loaded `FrequencyTrie<CompiledPatchCommand>` is immutable and can be shared across request
+threads. Load it once during application startup and reuse it for indexing and query processing.
+
+## 3. Choose A Language Resource
+
+Bundled dictionaries are exposed as enum constants. Common examples:
+
+| Language | Enum constant |
+| --- | --- |
+| English | `US_UK` |
+| German | `DE_DE` |
+| French | `FR_FR` |
+| Spanish | `ES_ES` |
+| Italian | `IT_IT` |
+| Polish | `PL_PL` |
+| Russian | `RU_RU` |
+| Czech | `CS_CZ` |
+
+The full list, writing-direction notes, and benchmark links are in
+[Built-in Languages](built-in-languages.md).
+
+## 4. Use The Same Stemmer On Both Sides
+
+For search, use the same Radixor configuration during indexing and query processing. A typical
+minimal integration flow is:
+
+1. tokenize text with your application or search platform,
+2. normalize tokens consistently,
+3. call `stemmer.get(token)`,
+4. apply the returned `CompiledPatchCommand`,
+5. index or query with the resulting stem.
+
+Do not load the trie per token. The compiled trie is the runtime artifact; per-token work should be
+limited to lookup and patch application.
+
+## 5. Next Step For Production
+
+The fast path compiles a bundled dictionary during startup. That is convenient for evaluation and
+small services. For larger deployments, compile once, persist a `.radixor.gz` artifact, and load
+that binary artifact at runtime.
+
+Continue with:
+
+- [Integration Deep Dive](integration-deep-dive.md) for production lifecycle guidance.
+- [Loading and Building Stemmers](programmatic-loading-and-building.md) for all loading APIs.
+- [Built-in Languages](built-in-languages.md) for bundled resources and dictionary locations.
+- [Benchmarking](benchmarking.md) for speed and quality interpretation.

docs/index.md

@@ -0,0 +1,39 @@
+<h1 class="visually-hidden">Home</h1>
+<p align="center">
+  <img src="assets/images/banner.jpg" alt="Radixor banner" style="width: 100%; max-width: 1100px;">
+</p>
+
+**Radixor** is a high-performance, multi-language stemmer for Java, built for production-grade search and text-processing systems.
+
+It modernizes the proven Egothor patch-command trie approach and extends it for deployment realities that classic stemming pipelines do not handle well.
+
+Traditional Egothor-style stemming workflows usually treat a compiled dictionary as a fixed artifact. Once built, its lexical knowledge is effectively closed unless the original source dictionary is recompiled. Radixor removes that constraint. An already compiled stemming structure can be extended with additional words and transformations, which makes it possible to evolve an existing dictionary for domain-specific, customer-specific, or deployment-specific vocabulary without rebuilding the entire lexical base from scratch.
+
+Radixor also improves how ambiguous reductions can be handled at runtime. Instead of always forcing a single result, it can return multiple plausible stems when the input token cannot be reduced unambiguously. This allows downstream systems to preserve linguistic ambiguity where that is operationally useful, whether for retrieval quality, ranking strategies, diagnostics, or domain-specific normalization policies.
+
+The project also has a clear research lineage. The historical idea behind this stemming family is described in Leo Galambos's paper *Lemmatizer for Document Information Retrieval Systems in JAVA* (SOFSEM 2001), which presents a semi-automatic stemming technique designed for Java-based information retrieval systems. In Radixor documentation, this reference serves as historical and algorithmic background rather than as technical documentation of the current implementation.
+
+> Unlike traditional Egothor-based deployments, Radixor can extend an already compiled stemmer dictionary and can return multiple stems when a word is not reducible to a single unambiguous form.
+
+Radixor delivers:
+
+- **Fast runtime stemming** with compact lookup structures
+- **Multi-language adaptability** through dictionary-driven compilation
+- **Extension of compiled stemmer structures** without full recompilation from source dictionaries
+- **Incremental vocabulary growth** for deployment-specific lexical refinement
+- **Support for multiple stemming results** when reduction is ambiguous
+- **Deterministic behavior** suitable for reproducible processing pipelines
+- **Flexible integration paths**, including CLI-based and programmatic workflows
+- **Operational transparency** through continuously published quality and benchmark reports
+
+Radixor is intended for teams that require consistent stemming quality at scale, while retaining the ability to evolve lexical resources after compilation and to handle ambiguous reductions with greater precision than traditional single-stem pipelines allow.
+
+## Start here
+
+- Read [Fast Track](fast-track.md) when you want the shortest path to a working bundled stemmer.
+- Use [Integration Deep Dive](integration-deep-dive.md) when you are wiring Radixor into a real application or search pipeline.
+- Read [Quick Start](quick-start.md) for the broader developer walkthrough after the first result works.
+- Use [Built-in Languages](built-in-languages.md) to find the bundled dictionaries exposed by Radixor.
+- Review [Benchmarking](benchmarking.md) and [Benchmark Results](benchmarks/index.md) for reproducible performance and quality methodology.
+- Open [CI Reports](reports.md) to inspect published build artifacts and quality metrics.
+- See the historical paper: [*Lemmatizer for Document Information Retrieval Systems in JAVA*](https://www.researchgate.net/publication/221512865_Lemmatizer_for_Document_Information_Retrieval_Systems_in_JAVA).

docs/integration-deep-dive.md

@@ -0,0 +1,235 @@
+# Integration Deep Dive
+
+This page explains how to integrate Radixor into a real Java application after the first
+fast-track experiment works. It covers dependencies, bundled dictionaries, runtime lifecycle,
+deployment artifacts, and the decisions that matter in search or text-processing systems.
+
+## Integration Model
+
+Radixor has two separate phases:
+
+| Phase | Work | Typical location |
+| --- | --- | --- |
+| Preparation | Parse dictionaries, derive patch commands, reduce and contract the trie, optionally persist a binary artifact. | Build pipeline, packaging job, admin tool, or startup for small services. |
+| Runtime | Load an immutable compiled trie, look up patch commands, apply them to tokens. | Search indexing, query processing, text normalization, enrichment pipelines. |
+
+The practical rule is simple: compile rarely, stem often.
+
+For production systems, prefer a startup-owned or dependency-injected singleton
+`FrequencyTrie<CompiledPatchCommand>` per language/configuration. The trie is immutable after
+construction and is suitable for concurrent reads.
+
+## Dependency Coordinates
+
+The Maven coordinates are:
+
+```text
+org.egothor:radixor
+```
+
+Gradle:
+
+```kotlin
+dependencies {
+    implementation("org.egothor:radixor:3.0.0")
+}
+```
+
+Maven:
+
+```xml
+<dependency>
+    <groupId>org.egothor</groupId>
+    <artifactId>radixor</artifactId>
+    <version>3.0.0</version>
+</dependency>
+```
+
+Replace `3.0.0` with the current release selected for your deployment.
+
+The core Java module is:
+
+```java
+module org.egothor.radixor;
+```
+
+A named consuming module declares:
+
+```java
+module example.search {
+    requires org.egothor.radixor;
+}
+```
+
+## Bundled Dictionaries
+
+Radixor ships bundled dictionaries inside the library artifact. The public API exposes them through:
+
+```java
+StemmerPatchTrieLoader.Language
+```
+
+The physical resources are packaged as compressed UTF-8 dictionaries under resource directories
+such as:
+
+```text
+us_uk/stemmer.gz
+de_de/stemmer.gz
+fr_fr/stemmer.gz
+pl_pl/stemmer.gz
+```
+
+Treat those resource paths as implementation details. Application code should load bundled
+dictionaries through `StemmerPatchTrieLoader.Language`, because the enum also carries the language
+metadata needed for correct traversal.
+
+See [Built-in Languages](built-in-languages.md) for the complete language list, writing-direction
+notes, and links to per-language benchmark pages.
+
+## Minimal Service Wrapper
+
+A small service wrapper keeps loading, null handling, and fallback behavior in one place.
+
+```java
+import java.io.IOException;
+import java.util.Objects;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+public final class RadixorStemmerService {
+
+    private final FrequencyTrie<CompiledPatchCommand> trie;
+
+    public RadixorStemmerService(final StemmerPatchTrieLoader.Language language) throws IOException {
+        this.trie = StemmerPatchTrieLoader.loadCompiled(
+                Objects.requireNonNull(language, "language"),
+                true,
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+    }
+
+    public String stem(final String token) {
+        final String checkedToken = Objects.requireNonNull(token, "token");
+        final CompiledPatchCommand command = trie.get(checkedToken);
+        return command == null ? checkedToken : command.apply(checkedToken);
+    }
+}
+```
+
+The fallback behavior preserves the original token when the trie has no patch command for it. That
+is usually the right default for search normalization, because unknown tokens should remain
+searchable.
+
+## Production Artifact Workflow
+
+For a controlled deployment, compile once and deploy the binary artifact:
+
+1. choose a bundled or custom dictionary,
+2. optionally extend it with domain vocabulary,
+3. compile a contracted trie,
+4. persist it as `.radixor.gz`,
+5. deploy that artifact with the application,
+6. load it with `StemmerPatchTrieLoader.loadBinaryCompiled(...)`.
+
+Runtime loading then avoids dictionary parsing and preparation-time memory pressure.
+
+```java
+import java.io.IOException;
+import java.nio.file.Path;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+public final class BinaryStemmerLoader {
+
+    private BinaryStemmerLoader() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static FrequencyTrie<CompiledPatchCommand> loadEnglish() throws IOException {
+        return StemmerPatchTrieLoader.loadBinaryCompiled(Path.of("stemmers", "english.radixor.gz"));
+    }
+}
+```
+
+Use [CLI Compilation](cli-compilation.md) for command-line artifact creation, or
+[Extending and Persisting Compiled Tries](programmatic-extending-and-persistence.md) for
+programmatic artifact generation.
+
+## Search Pipeline Guidance
+
+Use Radixor consistently across indexing and querying:
+
+- choose one language dictionary per field or per analysis chain,
+- apply the same token normalization before stemming on both sides,
+- keep the compiled trie in memory and reuse it,
+- use `get(...)` for a single preferred stem,
+- use `getAll(...)` when a retrieval model benefits from preserving alternatives,
+- version custom `.radixor.gz` artifacts with the application or index schema.
+
+For multilingual content, do not run every token through every language. Route text by field,
+document metadata, or language detection before stemming.
+
+## Choosing Bundled Versus Custom Dictionaries
+
+Start with bundled dictionaries when:
+
+- the language is supported,
+- the application needs a strong baseline quickly,
+- the vocabulary is general-purpose,
+- the team is evaluating Radixor or building an initial integration.
+
+Use custom or extended dictionaries when:
+
+- domain vocabulary changes search quality,
+- product names, technical terms, legal terms, or biomedical terms must be preserved consistently,
+- stemming behavior must be curated and reviewed,
+- a release process needs a versioned lexical artifact.
+
+The dictionary format is intentionally simple and documented in
+[Dictionary Format](dictionary-format.md). Contribution standards are described in
+[Contributing Dictionaries](contributing-dictionaries.md).
+
+## Performance Practices
+
+The hot path should be only:
+
+```text
+token -> trie lookup -> compiled command application -> stem
+```
+
+Avoid these patterns in production request paths:
+
+- loading or compiling dictionaries per request,
+- applying serialized patch strings repeatedly instead of `CompiledPatchCommand`,
+- rebuilding tries for short-lived batches,
+- mixing different stemmer configurations between indexing and querying,
+- interpreting speed without checking exact-root quality.
+
+The current benchmark documentation separates methodology, corpora, environment, and language
+results so performance claims remain auditable. Start with [Benchmarking](benchmarking.md), then
+use [Benchmark Results](benchmarks/index.md) for the detailed reference tree.
+
+## Operational Checklist
+
+Before production rollout:
+
+- dependency version is pinned,
+- language resource and reduction mode are documented,
+- indexing and query pipelines use the same stemming configuration,
+- custom artifacts are versioned and reproducible,
+- fallback behavior for unknown tokens is explicit,
+- benchmark expectations are read together with quality metrics,
+- CI includes at least a smoke test that stems representative project vocabulary.
+
+## Related Pages
+
+- [Fast Track](fast-track.md)
+- [Quick Start](quick-start.md)
+- [Built-in Languages](built-in-languages.md)
+- [Programmatic Usage](programmatic-usage.md)
+- [CLI Compilation](cli-compilation.md)
+- [Benchmarking](benchmarking.md)

docs/lookup-edge-optimization.md

@@ -0,0 +1,198 @@
+# Lookup Edge Optimization
+
+Compiled trie nodes (`CompiledNode`) use three lookup strategies when resolving child edges:
+
+1. dense array direct lookup,
+2. linear scan for very small child counts,
+3. binary search over sorted edge labels.
+
+This page explains the dense path, what `maxExpandedIndex` controls, and how to tune it. These
+edge lookup strategies operate after trie reduction and uniform-subtree contraction. If lookup
+reaches an accepting contracted leaf, no child edge search is needed for the remaining input
+characters.
+
+## Runtime model of one node
+
+For a node with sorted edge labels `char[] edges`, the implementation can materialize an
+index-aligned dense table when labels occupy a small compact code-point interval:
+
+```text
+span = maxEdge - minEdge
+use dense table iff (span <= maxExpandedIndex) and (maxExpandedIndex > 0)
+```
+
+When dense lookup is used, lookup is constant-time indexing:
+
+```text
+denseIndex = requestedEdge - minEdge
+return denseChildren[denseIndex]  // or null if outside interval
+```
+
+When dense lookup is not active (interval is too wide or the configured
+`maxExpandedIndex` is `0`), `CompiledNode` still chooses between two fallback
+strategies:
+
+- **linear scan** for very small child counts (`4` or fewer children),
+- **binary search** for larger child counts.
+
+This means the fallback method is selected by child count, not by “distance” alone.
+`linear scan` is therefore used when there are only a few edges even if those edges are
+spread across very distant code points.
+
+### Example: few edges, wide Unicode span
+
+```text
+edges = ['a', '中', '你']
+edge count = 3
+minEdge = 'a' (U+0061)
+maxEdge = '你' (U+4F60)
+span = 20319
+```
+
+- If `maxExpandedIndex = 512`, dense indexing is not used because `span > maxExpandedIndex`.
+- Because `edge count = 3` (<= 4), lookup falls back to a tiny linear scan of the
+  three labels.
+- This is exactly the case where you get benefit from the threshold even though the interval is wide.
+
+This is useful for non-Latin scripts as well: what matters is interval width in Unicode
+code points, not script name. A compact Arabic-range block can still benefit from dense
+lookups when keys stay in a tight code-point interval.
+
+## Why this is configurable
+
+`maxExpandedIndex` is only a performance/paging choice:
+
+- higher value:
+  - more compact intervals qualify for dense tables,
+  - more constant-time child lookup,
+  - more memory for dense tables in qualifying nodes.
+- lower value (or `0`):
+  - less dense-table allocation,
+  - fewer branches into constant-time path,
+  - lower materialization memory.
+
+The value never changes lookup semantics. It only changes the in-memory structure shape.
+
+## Persistence and loading model
+
+This threshold is **not** stored in `TrieMetadata`.
+
+- The binary format stores only trie payload and semantic metadata (`reduction`, `traversal`,
+  case/diacritic settings, contraction settings, and stream version).
+- `maxExpandedIndex` is chosen when materializing nodes in memory.
+- You can therefore keep one persisted artifact and load it with different in-memory
+  trade-offs depending on deployment constraints.
+
+## Default
+
+- `FrequencyTrie.DEFAULT_MAX_EXPANDED_INDEX == 512`
+- `CompiledNode.DEFAULT_MAX_EXPANDED_INDEX == 512`
+
+These are practical defaults for mixed-language text and Latin-like scripts where edge labels
+often cluster.
+
+## Tune during build (writable phase)
+
+Use the full `FrequencyTrie.Builder` constructor when you are compiling from source data.
+The builder threshold is applied while freezing reduced nodes into the immutable form.
+
+```java
+import org.egothor.stemmer.CaseProcessingMode;
+import org.egothor.stemmer.DiacriticProcessingMode;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.ReductionSettings;
+import org.egothor.stemmer.WordTraversalDirection;
+
+final ReductionSettings settings = ReductionSettings.withDefaults(
+        ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+
+final FrequencyTrie.Builder<String> fastBuilder =
+        new FrequencyTrie.Builder<>(String[]::new,
+                settings,
+                WordTraversalDirection.BACKWARD,
+                CaseProcessingMode.LOWERCASE_WITH_LOCALE_ROOT,
+                DiacriticProcessingMode.AS_IS,
+                1024); // prefer lookup speed
+
+// ... put(...) ...
+final FrequencyTrie<String> trie = fastBuilder.build();
+```
+
+Use `0` or `256` for lower memory while still building larger tries.
+
+```java
+final FrequencyTrie.Builder<String> compactBuilder =
+        new FrequencyTrie.Builder<>(String[]::new,
+                settings,
+                WordTraversalDirection.BACKWARD,
+                CaseProcessingMode.LOWERCASE_WITH_LOCALE_ROOT,
+                DiacriticProcessingMode.AS_IS,
+                256); // lower memory profile
+```
+
+## Tune when loading a binary artifact (runtime phase)
+
+At artifact load time, you can tune the same trade-off independently of persisted metadata.
+
+```java
+import java.nio.file.Path;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+final FrequencyTrie<CompiledPatchCommand> defaultLookup = StemmerPatchTrieLoader.loadBinaryCompiled(
+        Path.of("stemmers", "english.radixor.gz"));
+
+final FrequencyTrie<CompiledPatchCommand> fastLookup = StemmerPatchTrieLoader.loadBinaryCompiled(
+        Path.of("stemmers", "english.radixor.gz"), 1024);
+
+final FrequencyTrie<CompiledPatchCommand> compactLookup = StemmerPatchTrieLoader.loadBinaryCompiled(
+        Path.of("stemmers", "english.radixor.gz"), 0);
+```
+
+You can also set the threshold directly with `FrequencyTrie.readFrom(...)` when reading streams:
+
+```java
+import java.io.DataInputStream;
+import java.io.IOException;
+import java.io.InputStream;
+import java.nio.file.Files;
+import java.nio.file.Path;
+import java.util.zip.GZIPInputStream;
+
+import org.egothor.stemmer.FrequencyTrie;
+
+public final class StreamLoadExample {
+
+    private StreamLoadExample() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        try (InputStream fileInput = Files.newInputStream(Path.of("stemmers", "english.radixor.gz"));
+                GZIPInputStream gzip = new GZIPInputStream(fileInput);
+                DataInputStream dataInput = new DataInputStream(gzip)) {
+            final FrequencyTrie<String> compactOnLoad = FrequencyTrie.readFrom(
+                    dataInput,
+                    String[]::new,
+                    input -> input.readUTF(),
+                    256);
+        }
+    }
+}
+```
+
+Note: the string codec is intentionally inline in this snippet to keep it self-contained.
+
+## Practical guidance
+
+- Start with default (`512`) in production and profile before changing it.
+- Use `0` when memory is the priority and query throughput is not the bottleneck.
+- Use values around `1024` for workloads dominated by compact alphabets and very hot lookups.
+
+Trade-off expectation:
+
+- increasing `maxExpandedIndex` improves lookup speed when edges tend to occupy short spans,
+- decreasing it reduces per-node auxiliary memory in dense-span nodes.

docs/migration-and-backward-compatibility.md

@@ -0,0 +1,155 @@
+# Migration and Backward Compatibility
+
+This page describes the migration from repeated serialized patch-command application to compiled patch commands.
+
+## Summary
+
+Radixor patch commands are still encoded as compact strings when dictionaries are built and persisted. That serialized form remains the interchange format used by textual dictionaries, binary artifacts, and compilation tooling.
+
+Runtime stemming should no longer repeatedly apply those serialized strings directly. Since 2.3.0, the String-based patch application API is deprecated. Code that stems live input should load or create `CompiledPatchCommand` values and reuse them. The deprecated API remains available for compatibility during the transition, but applications should migrate before 3.0.0.
+
+The reason is performance. The old API parses the serialized P-command every time it is applied. `CompiledPatchCommand` parses it once and stores a concrete immutable command object, so repeated stemming avoids the same analysis work.
+
+## Deprecated Runtime APIs
+
+The following API family is kept for source compatibility but is no longer the preferred runtime path:
+
+- `PatchCommandEncoder.apply(String, String)`
+- `PatchCommandEncoder.apply(String, String, WordTraversalDirection)`
+- `PatchCommandEncoder.applyTo(..., String, WordTraversalDirection, ...)`
+- `PatchCommandEncoder.applyWithConfiguredDirection(String, String)`
+- `StemmerPatchTrieLoader.load(...)` overloads returning `FrequencyTrie<String>`
+- `StemmerPatchTrieLoader.loadBinary(...)` overloads returning `FrequencyTrie<String>`
+
+Use the compiled equivalents for runtime stemming:
+
+- `CompiledPatchCommand.compile(String, WordTraversalDirection)`
+- `PatchCommandEncoder.compile(String)`
+- `PatchCommandEncoder.compile(String, WordTraversalDirection)`
+- `StemmerPatchTrieLoader.loadCompiled(...)`
+- `StemmerPatchTrieLoader.loadBinaryCompiled(...)`
+
+## Loading A Text Dictionary
+
+Old runtime code:
+
+```java
+Path dictionary = Path.of("dictionary.txt");
+ReductionSettings settings = ReductionSettings.withDefaults(
+        ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+FrequencyTrie<String> trie = StemmerPatchTrieLoader.load(dictionary, true, settings);
+
+String word = "running";
+String patch = trie.get(word);
+String stem = patch == null
+        ? word
+        : PatchCommandEncoder.apply(word, patch, trie.traversalDirection());
+```
+
+New runtime code:
+
+```java
+Path dictionary = Path.of("dictionary.txt");
+ReductionSettings settings = ReductionSettings.withDefaults(
+        ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadCompiled(dictionary, true, settings);
+
+String word = "running";
+CompiledPatchCommand patch = trie.get(word);
+String stem = patch == null ? word : patch.apply(word);
+```
+
+## Loading A Binary Artifact
+
+Old runtime code:
+
+```java
+FrequencyTrie<String> trie = StemmerPatchTrieLoader.loadBinary(Path.of("us-uk.radixor.gz"));
+
+String word = "studies";
+String patch = trie.get(word);
+String stem = patch == null
+        ? word
+        : PatchCommandEncoder.apply(word, patch, trie.traversalDirection());
+```
+
+New runtime code:
+
+```java
+FrequencyTrie<CompiledPatchCommand> trie =
+        StemmerPatchTrieLoader.loadBinaryCompiled(Path.of("us-uk.radixor.gz"));
+
+String word = "studies";
+CompiledPatchCommand patch = trie.get(word);
+String stem = patch == null ? word : patch.apply(word);
+```
+
+Existing binary artifacts remain readable. `loadBinaryCompiled(...)` reads the stored serialized patch strings and compiles them during load setup, before live stemming begins.
+
+## Manual Patch Encoding
+
+Encoding still produces a serialized patch command because that is the compact stored representation:
+
+```java
+PatchCommandEncoder encoder = PatchCommandEncoder.builder().build();
+String patch = encoder.encode("running", "run");
+```
+
+Old repeated application:
+
+```java
+String stem = PatchCommandEncoder.apply("running", patch);
+```
+
+New repeated application:
+
+```java
+CompiledPatchCommand compiled = encoder.compile(patch);
+String stem = compiled.apply("running");
+```
+
+## Caller-Owned Output Buffers
+
+Old buffer-oriented code:
+
+```java
+char[] output = new char[32];
+int length = PatchCommandEncoder.applyTo(
+        "running",
+        patch,
+        WordTraversalDirection.BACKWARD,
+        output,
+        0,
+        output.length);
+```
+
+New buffer-oriented code:
+
+```java
+CompiledPatchCommand compiled = CompiledPatchCommand.compile(patch, WordTraversalDirection.BACKWARD);
+char[] output = new char[32];
+int length = compiled.applyTo("running", output, 0, output.length);
+```
+
+Both APIs return `CompiledPatchCommand.APPLY_INSUFFICIENT_CAPACITY` when the caller-owned output range is too small.
+
+## Compatibility Rules
+
+Serialized patch strings remain part of the dictionary and artifact format. The deprecation is about repeated runtime application of serialized strings, not about the stored representation itself.
+
+Compatibility tests may continue to exercise the deprecated API to prove that old artifacts and source code still work during the transition. New production code, examples, and benchmark runtime paths should use `CompiledPatchCommand`.
+
+The command-line compiler still writes artifacts containing serialized patch commands. Runtime loaders can expose those commands as compiled immutable objects through `loadCompiled(...)` and `loadBinaryCompiled(...)`.
+
+## Contracted Trie Artifacts
+
+Current compiled loaders and freshly written binary artifacts can use contracted compiled tries.
+Contraction replaces a subtree with an accepting leaf when every reachable entry below that subtree
+selects the same preferred patch command. This changes the physical trie shape and the binary
+stream version, but it does not change the serialized patch-command language.
+
+Existing binary artifacts remain readable through the compatibility reader. To obtain the
+contracted runtime representation, rebuild the artifact with the current compiler or load the
+source dictionary through the current `loadCompiled(...)` APIs. Applications that only consume
+`CompiledPatchCommand` values through `get()` and `apply(...)` do not need code changes for this
+optimization.

docs/programmatic-extending-and-persistence.md

@@ -0,0 +1,107 @@
+# Extending and Persisting Compiled Tries
+
+This document explains how compiled Radixor tries can be reopened, extended, rebuilt, and stored for deployment.
+
+## Reopen and extend a compiled trie
+
+`FrequencyTrieBuilders.copyOf(...)` reconstructs a mutable builder from a compiled trie. The reconstructed builder preserves the key-local value counts of the compiled trie as currently stored, making it suitable for subsequent modification and recompilation. Reconstruction is performed from the compiled state, not from the original unreduced insertion history.
+
+```java
+import java.io.IOException;
+import java.nio.file.Path;
+
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.FrequencyTrieBuilders;
+import org.egothor.stemmer.PatchCommandEncoder;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.ReductionSettings;
+import org.egothor.stemmer.StemmerPatchTrieBinaryIO;
+
+public final class ExtendCompiledStemmerExample {
+
+    private ExtendCompiledStemmerExample() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<String> compiledTrie = StemmerPatchTrieBinaryIO.read(
+                Path.of("stemmers", "english.radixor.gz"));
+
+        final ReductionSettings settings = ReductionSettings.withDefaults(
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+
+        final FrequencyTrie.Builder<String> builder = FrequencyTrieBuilders.copyOf(
+                compiledTrie,
+                String[]::new,
+                settings);
+
+        builder.put("microservices", "Na");
+
+        final FrequencyTrie<String> updatedTrie = builder.build();
+
+        StemmerPatchTrieBinaryIO.write(
+                updatedTrie,
+                Path.of("stemmers", "english-custom.radixor.gz"));
+    }
+}
+```
+
+This enables a layered workflow:
+
+1. start from a bundled or already compiled stemmer,
+2. reconstruct a builder,
+3. add custom lexical data,
+4. compile and persist a new binary artifact.
+
+## Persist and deploy compiled tries
+
+`StemmerPatchTrieBinaryIO` reads and writes patch-command tries as GZip-compressed binary files. `StemmerPatchTrieLoader` exposes convenience methods around the same persistence functionality.
+
+```java
+import java.io.IOException;
+import java.nio.file.Path;
+
+import org.egothor.stemmer.StemmerPatchTrieBinaryIO;
+
+StemmerPatchTrieBinaryIO.write(trie, Path.of("stemmers", "english.radixor.gz"));
+```
+
+In deployment terms, the cleanest model is usually:
+
+- compile once,
+- persist the binary artifact,
+- load the artifact directly in runtime services.
+
+## Binary-first operational model
+
+For larger dictionaries or controlled deployment environments, a binary-first workflow is usually the most robust choice:
+
+- prepare the compiled trie offline,
+- keep the preparation step outside the runtime startup path,
+- version and distribute the binary artifact,
+- load the finished trie directly in production.
+
+This model works especially well when domain-specific extensions are added in layers and then recompiled into a new read-only artifact.
+
+## Continue with
+
+- [Loading and Building Stemmers](programmatic-loading-and-building.md)
+- [Querying and Ambiguity Handling](programmatic-querying-and-ambiguity.md)
+
+
+## Inspecting persisted metadata
+
+After loading a compiled artifact, applications can inspect the persisted build descriptor directly:
+
+```java
+final FrequencyTrie<CompiledPatchCommand> trie =
+        StemmerPatchTrieLoader.loadBinaryCompiled("build/stemmers/cs_cz.dat.gz");
+final TrieMetadata metadata = trie.metadata();
+
+System.out.println(metadata.formatVersion());
+System.out.println(metadata.traversalDirection());
+System.out.println(metadata.reductionSettings().reductionMode());
+System.out.println(metadata.diacriticProcessingMode());
+```
+
+This is especially useful when a deployment manages multiple artifacts compiled under different traversal or reduction regimes.

docs/programmatic-loading-and-building.md

@@ -0,0 +1,186 @@
+# Loading and Building Stemmers
+
+This document explains how to acquire a compiled Radixor stemmer in Java.
+
+## Load a bundled language dictionary
+
+Bundled language resources are simple to use and compile directly into a `FrequencyTrie<CompiledPatchCommand>` during loading.
+
+```java
+import java.io.IOException;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+public final class BundledLanguageExample {
+
+    private BundledLanguageExample() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadCompiled(
+                StemmerPatchTrieLoader.Language.US_UK,
+                true,
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+    }
+}
+```
+
+The `storeOriginal` flag controls whether the canonical stem is inserted as a no-op patch entry for the stem itself.
+
+Bundled `loadCompiled(...)` entry points build the runtime trie with the same contracted
+representation used by the published benchmarks. During compilation, uniform preferred-command
+subtrees are collapsed into accepting leaves, so lookup can stop before consuming the entire input
+when the remaining characters cannot change the selected patch command.
+
+## Load a textual dictionary
+
+Loading from a dictionary file follows the same preparation model as bundled resources, but the source comes from your own file or path. The input may be plain UTF-8 text or GZip-compressed UTF-8 text; the loader detects GZip data from the stream header. The textual format is tab-separated values, meaning that columns are separated by the tab character. Each non-empty logical line starts with the stem column and may contain zero or more variant columns. Input case normalization is controlled by `CaseProcessingMode` (default: `LOWERCASE_WITH_LOCALE_ROOT`), trailing remarks introduced by `#` or `//` are ignored, and dictionary items containing embedded whitespace are currently ignored with warning-level diagnostics.
+
+```java
+import java.io.IOException;
+import java.nio.file.Path;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.ReductionSettings;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+public final class LoadTextDictionaryExample {
+
+    private LoadTextDictionaryExample() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadCompiled(
+                Path.of("data", "stemmer.tsv"),
+                true,
+                ReductionSettings.withDefaults(
+                        ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS));
+    }
+}
+```
+
+Additional `StemmerPatchTrieLoader.loadCompiled(...)` overloads let callers provide explicit `WordTraversalDirection`, `CaseProcessingMode`, `DiacriticProcessingMode`, or a complete `TrieMetadata` instance. Use those overloads when a custom dictionary must be compiled with forward traversal for right-to-left languages, case-sensitive keys, or diacritic stripping.
+
+When `ReductionSettings` are supplied through these compiled loader APIs, uniform-subtree
+contraction is still enabled as an internal pre-reduction step. The public `ReductionMode` remains
+the semantic policy for subtree equivalence after that contraction has removed regions whose
+preferred command is already uniform.
+
+## Load a compiled binary artifact
+
+Binary loading is typically the preferred runtime path because it avoids reparsing the textual source and skips the preparation step entirely.
+
+```java
+import java.io.IOException;
+import java.nio.file.Path;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+public final class LoadBinaryExample {
+
+    private LoadBinaryExample() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadBinaryCompiled(
+                Path.of("stemmers", "english.radixor.gz"));
+    }
+}
+```
+
+The binary format is the native `FrequencyTrie` serialization wrapped in GZip compression. It includes persisted `TrieMetadata`, so lookup after loading uses the traversal, case-processing, diacritic-processing, and reduction settings captured when the trie was compiled.
+
+## Tune child lookup density when loading binaries
+
+To optimize hot-path latency, you can tune direct child indexing by passing `maxExpandedIndex`
+at load time. This does not change persisted metadata, only the materialized in-memory form.
+
+```java
+import java.io.IOException;
+import java.nio.file.Path;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+public final class LoadBinaryWithDenseLookupExample {
+
+    private LoadBinaryWithDenseLookupExample() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<CompiledPatchCommand> balanced = StemmerPatchTrieLoader.loadBinaryCompiled(
+                Path.of("stemmers", "english.radixor.gz"));
+
+        final FrequencyTrie<CompiledPatchCommand> fast = StemmerPatchTrieLoader.loadBinaryCompiled(
+                Path.of("stemmers", "english.radixor.gz"),
+                1024);
+
+        final FrequencyTrie<CompiledPatchCommand> compact = StemmerPatchTrieLoader.loadBinaryCompiled(
+                Path.of("stemmers", "english.radixor.gz"),
+                0);
+    }
+}
+```
+
+Negative values still use `FrequencyTrie.DEFAULT_MAX_EXPANDED_INDEX`.
+
+[Lookup Edge Optimization](lookup-edge-optimization.md) describes the trade-off in detail and examples for build-time tuning as well.
+
+## Build directly with a mutable builder
+
+A `FrequencyTrie.Builder<V>` accepts repeated `put(key, value)` calls and compiles the final read-only trie through `build()`. Compilation performs bottom-up reduction and produces the compact immutable runtime representation.
+
+```java
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.PatchCommandEncoder;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.ReductionSettings;
+
+public final class BuilderExample {
+
+    private BuilderExample() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) {
+        final ReductionSettings settings = ReductionSettings.withDefaults(
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+
+        final FrequencyTrie.Builder<String> builder =
+                new FrequencyTrie.Builder<>(String[]::new, settings);
+
+        final PatchCommandEncoder encoder = PatchCommandEncoder.builder().build();
+
+        builder.put("running", encoder.encode("running", "run"));
+        builder.put("runs", encoder.encode("runs", "run"));
+        builder.put("ran", encoder.encode("ran", "run"));
+        builder.put("runner", encoder.encode("runner", "run"));
+
+        final FrequencyTrie<String> trie = builder.build();
+        System.out.println("Canonical node count: " + trie.size());
+    }
+}
+```
+
+## Preparation-time memory characteristics
+
+Compilation is commonly a one-time preparation activity and is generally fast enough not to be the main operational concern. The more important constraint is memory usage while building from textual dictionary data. Before reduction produces the compact immutable structure, the mutable build-time representation keeps the inserted data in memory. This is precisely why very large source dictionaries may require noticeably more memory during preparation than after compilation. The resulting compiled trie, by contrast, is designed as the compact runtime form.
+
+This makes offline preparation especially attractive for large dictionaries.
+
+## Continue with
+
+- [Querying and Ambiguity Handling](programmatic-querying-and-ambiguity.md)
+- [Extending and Persisting Compiled Tries](programmatic-extending-and-persistence.md)

docs/programmatic-querying-and-ambiguity.md

@@ -0,0 +1,125 @@
+# Querying and Ambiguity Handling
+
+This document explains how a compiled Radixor trie is queried and how ambiguity is represented.
+
+## Query a compiled trie
+
+### `get(...)`: preferred local value
+
+`FrequencyTrie.get(String)` returns the most frequent value stored at the node addressed by the supplied key. If several values have the same local frequency, the winner is chosen deterministically by shorter `toString()` value first, then by lexicographically lower `toString()`, and finally by stable first-seen order. If the key does not exist or no value is stored at the addressed node, `null` is returned.
+
+```java
+final String word = "running";
+final CompiledPatchCommand patch = trie.get(word);
+```
+
+### `getAll(...)`: ordered local values
+
+`FrequencyTrie.getAll(String)` returns all values stored at the addressed node, ordered by descending frequency using the same deterministic tie-breaking rules. The returned array is a defensive copy. If the key is missing or has no local values, an empty array is returned.
+
+```java
+final CompiledPatchCommand[] patches = trie.getAll("axes");
+```
+
+### `getEntries(...)`: values with counts
+
+`FrequencyTrie.getEntries(String)` returns immutable `ValueCount<V>` objects aligned with the same ordering used by `getAll(...)`.
+
+```java
+import java.util.List;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.ValueCount;
+
+final List<ValueCount<CompiledPatchCommand>> entries = trie.getEntries("axes");
+```
+
+### Visitor lookup for hot paths
+
+For allocation-sensitive token loops, use the visitor-style lookup methods. They visit the same ordered local values and counts without allocating a result array, list, or `ValueCount` objects.
+
+```java
+trie.getAll("axes", (patch, count, rank) -> {
+    // rank is zero-based and follows the same ordering as getAll(String).
+    return true; // return false to stop after this callback
+}, 8);
+```
+
+If the caller has already normalized the input exactly as required by `trie.metadata()`, the normalized methods avoid lookup normalization buffers too:
+
+```java
+final char[] token = "axes".toCharArray();
+trie.getAllNormalized(token, 0, token.length, (patch, count, rank) -> {
+    return true;
+}, 8);
+```
+
+`getAllNormalized(...)` bypasses `caseProcessingMode` and `diacriticProcessingMode`; callers are responsible for supplying canonical input. `maxResults == 0` visits nothing, negative values are rejected, and a sink returning `false` stops iteration after the current callback.
+
+## Apply compiled patch commands
+
+A patch command is not the final stem. It must be applied to the original input token. Runtime code should use `CompiledPatchCommand`, which parses the stored patch-command representation once during setup and then applies the concrete immutable command repeatedly.
+
+```java
+import org.egothor.stemmer.CompiledPatchCommand;
+
+final String word = "running";
+final CompiledPatchCommand patch = trie.get(word);
+final String stem = patch == null ? word : patch.apply(word);
+```
+
+Hot paths can apply a patch into caller-owned character storage:
+
+```java
+final char[] output = new char[32];
+final int produced = patch.applyTo(
+        word,
+        output,
+        0,
+        output.length);
+
+if (produced != CompiledPatchCommand.APPLY_INSUFFICIENT_CAPACITY) {
+    final String stem = new String(output, 0, produced);
+}
+```
+
+`applyTo(...)` returns the produced character count on success and `APPLY_INSUFFICIENT_CAPACITY` when the output range is too small. Capacity failure does not write partial output. The source and output ranges of the `char[]` overload must not overlap.
+
+For multiple candidates:
+
+```java
+final String word = "axes";
+for (final CompiledPatchCommand patch : trie.getAll(word)) {
+    final String stem = patch.apply(word);
+    System.out.println(word + " -> " + stem + " (" + patch + ")");
+}
+```
+
+The historical `PatchCommandEncoder.apply(...)` API still exists for compatibility with code that directly handles serialized patch-command strings, but it is deprecated because it reparses the command on every call. See [Migration and Backward Compatibility](migration-and-backward-compatibility.md) for the old and new forms side by side.
+
+## Understand reduction modes
+
+Reduction mode determines how mutable subtrees are merged during compilation. All modes operate on full subtree semantics rather than only on local node content.
+
+### `MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS`
+
+This mode merges subtrees whose `getAll()` results are equivalent for every reachable key suffix and whose local result ordering is the same. It ignores absolute frequencies when comparing subtree signatures, but it preserves ranked multi-result ordering semantics.
+
+### `MERGE_SUBTREES_WITH_EQUIVALENT_UNORDERED_GET_ALL_RESULTS`
+
+This mode also merges according to `getAll()` equivalence for every reachable key suffix, but it ignores local result ordering in addition to absolute frequencies. It is therefore more aggressive in what it considers equivalent.
+
+### `MERGE_SUBTREES_WITH_EQUIVALENT_DOMINANT_GET_RESULTS`
+
+This mode focuses on `get()` equivalence for every reachable key suffix, subject to dominance constraints. If a node does not satisfy the configured dominance thresholds, the implementation falls back to ranked `getAll()` semantics for that node to avoid unsafe over-reduction. The thresholds are configured through `ReductionSettings`. Defaults are 75 percent minimum winner share and a winner-over-second ratio of 3.
+
+## Practical guidance
+
+- choose a ranked `getAll()` mode when downstream ambiguity handling matters,
+- choose the dominant `get()` mode when the primary operational concern is the preferred result,
+- treat reduction mode as part of observable lookup semantics, not merely as an internal compression setting.
+
+## Continue with
+
+- [Extending and Persisting Compiled Tries](programmatic-extending-and-persistence.md)
+- [Loading and Building Stemmers](programmatic-loading-and-building.md)

docs/programmatic-usage.md

@@ -1,322 +1,80 @@
 # Programmatic Usage
 
-> ← Back to [README.md](../README.md)
+This document provides the programmatic entry point to **Radixor**.
 
-This document describes how to use **Radixor** programmatically from Java.
+Radixor follows a clear lifecycle:
 
-It covers:
+1. acquire a compiled stemmer,
+2. query it for patch commands,
+3. apply those commands to produce stems,
+4. reopen and extend the compiled structure when needed.
 
-- building a trie from dictionary data
-- compiling it into an immutable structure
-- loading compiled stemmers
-- querying for stems
-- working with multiple candidates
-- modifying existing compiled stemmers
+## Conceptual model
 
+Radixor is dictionary-driven, but runtime stemming does not operate by scanning raw dictionary files. A source dictionary is parsed as a sequence of canonical stems and their known variants. Each variant is converted into a compact patch command that transforms the variant into the stem, while the stem itself may optionally be stored as a canonical no-op patch. The mutable trie is then reduced into a compiled read-only structure that stores ordered values and their counts at addressed nodes.
 
+Two consequences matter for developers:
 
-## Overview
+- the quality and coverage of stemming behavior depend on dictionary richness,
+- runtime usage is based on compiled patch-command lookup rather than on direct dictionary traversal.
 
-Radixor separates the stemming lifecycle into three stages:
+This is why Radixor can generalize beyond explicitly listed forms and why compiled artifacts are well suited for deployment.
 
-1. **Build** – collect word–stem mappings in a mutable structure  
-2. **Compile** – reduce and convert to an immutable trie  
-3. **Query** – perform fast runtime lookups  
+## Documentation map
 
-These stages are represented by:
+The programmatic API is easier to understand when split by developer task:
 
-- `FrequencyTrie.Builder` (mutable)
-- `FrequencyTrie` (immutable, compiled)
-- `StemmerPatchTrieLoader` / `StemmerPatchTrieBinaryIO` (I/O)
+- [Fast Track](fast-track.md) gives the shortest dependency-to-first-stem path for a new Java project.
+- [Integration Deep Dive](integration-deep-dive.md) explains production integration, deployment artifacts, search-pipeline usage, and operational decisions.
+- [Loading and Building Stemmers](programmatic-loading-and-building.md) explains how to acquire a compiled stemmer from bundled resources, textual dictionaries, binary artifacts, or direct builder usage.
+- [Lookup Edge Optimization](lookup-edge-optimization.md) explains dense child lookup tuning and the speed/memory trade-off when materializing compiled tries.
+- [Querying and Ambiguity Handling](programmatic-querying-and-ambiguity.md) explains `get(...)`, `getAll(...)`, `getEntries(...)`, patch application, and the practical meaning of reduction modes.
+- [Extending and Persisting Compiled Tries](programmatic-extending-and-persistence.md) explains how to reopen compiled tries, add new lexical data, rebuild them, and store them as binary artifacts.
 
+## Core types
 
+The main types involved in programmatic usage are:
 
-## Building a trie programmatically
+- `FrequencyTrie.Builder<V>` for mutable construction and extension,
+- `FrequencyTrie<V>` for the compiled read-only trie,
+- `PatchCommandEncoder` for creating serialized patch commands,
+- `CompiledPatchCommand` for repeated runtime patch application,
+- `StemmerPatchTrieLoader` for loading bundled or textual dictionaries,
+- `StemmerPatchTrieBinaryIO` for reading and writing compressed binary artifacts,
+- `FrequencyTrieBuilders` for reconstructing a mutable builder from a compiled trie,
+- `ReductionMode` and `ReductionSettings` for controlling compilation semantics.
 
-You can construct a trie directly without using the CLI.
+## Java module system (JPMS)
+
+The core artifact is published as an explicit JPMS module:
 
 ```java
-import org.egothor.stemmer.*;
+module org.egothor.radixor;
+```
 
-public final class BuildExample {
+A named consuming module uses:
 
-    public static void main(String[] args) {
-        ReductionSettings settings = ReductionSettings.withDefaults(
-                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
-        );
-
-        FrequencyTrie.Builder<String> builder =
-                new FrequencyTrie.Builder<>(String[]::new, settings);
-
-        PatchCommandEncoder encoder = new PatchCommandEncoder();
-
-        builder.put("running", encoder.encode("running", "run"));
-        builder.put("runs", encoder.encode("runs", "run"));
-        builder.put("ran", encoder.encode("ran", "run"));
-
-        FrequencyTrie<String> trie = builder.build();
-    }
+```java
+module example.consumer {
+    requires org.egothor.radixor;
 }
 ```
 
+The core module is standalone and can be consumed directly as a normal Java module.
 
+## Recommended reading order
 
-## Loading from dictionary files
-
-To parse dictionary files directly:
-
-```java
-import java.io.IOException;
-import java.nio.file.Path;
-
-import org.egothor.stemmer.*;
-
-public final class LoadFromDictionaryExample {
-
-    public static void main(String[] args) throws IOException {
-        FrequencyTrie<String> trie = StemmerPatchTrieLoader.load(
-                Path.of("data/stemmer.txt"),
-                true,
-                ReductionSettings.withDefaults(
-                        ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
-                )
-        );
-    }
-}
-```
-
-
-
-## Loading a compiled binary trie
-
-```java
-import java.io.IOException;
-import java.nio.file.Path;
-
-import org.egothor.stemmer.*;
-
-public final class LoadBinaryExample {
-
-    public static void main(String[] args) throws IOException {
-        FrequencyTrie<String> trie =
-                StemmerPatchTrieLoader.loadBinary(Path.of("english.radixor.gz"));
-    }
-}
-```
-
-This is the **preferred production approach**.
-
-
-
-## Querying for stems
-
-### Preferred result
-
-```java
-String word = "running";
-String patch = trie.get(word);
-String stem = PatchCommandEncoder.apply(word, patch);
-```
-
-### All candidates
-
-```java
-String[] patches = trie.getAll(word);
-
-for (String patch : patches) {
-    String stem = PatchCommandEncoder.apply(word, patch);
-}
-```
-
-
-
-## Accessing value frequencies
-
-For diagnostic or advanced use cases:
-
-```java
-import org.egothor.stemmer.ValueCount;
-
-java.util.List<ValueCount<String>> entries = trie.getEntries("axes");
-
-for (ValueCount<String> entry : entries) {
-    String patch = entry.value();
-    int count = entry.count();
-}
-```
-
-This allows:
-
-* inspecting ambiguity
-* understanding ranking decisions
-* debugging dictionary quality
-
-
-
-## Using bundled language resources
-
-```java
-FrequencyTrie<String> trie = StemmerPatchTrieLoader.load(
-        StemmerPatchTrieLoader.Language.US_UK_PROFI,
-        true,
-        ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
-);
-```
-
-Bundled dictionaries are useful for:
-
-* quick integration
-* testing
-* reference behavior
-
-
-
-## Persisting a compiled trie
-
-```java
-import java.io.IOException;
-import java.nio.file.Path;
-
-import org.egothor.stemmer.*;
-
-public final class SaveExample {
-
-    public static void main(String[] args) throws IOException {
-        StemmerPatchTrieBinaryIO.write(trie, Path.of("english.radixor.gz"));
-    }
-}
-```
-
-
-
-## Modifying an existing trie
-
-A compiled trie can be reopened into a builder, extended, and rebuilt.
-
-```java
-import java.io.IOException;
-import java.nio.file.Path;
-
-import org.egothor.stemmer.*;
-
-public final class ModifyExample {
-
-    public static void main(String[] args) throws IOException {
-        FrequencyTrie<String> compiled =
-                StemmerPatchTrieBinaryIO.read(Path.of("english.radixor.gz"));
-
-        ReductionSettings settings = ReductionSettings.withDefaults(
-                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS
-        );
-
-        FrequencyTrie.Builder<String> builder =
-                FrequencyTrieBuilders.copyOf(compiled, String[]::new, settings);
-
-        builder.put("microservices", PatchCommandEncoder.NOOP_PATCH);
-
-        FrequencyTrie<String> updated = builder.build();
-
-        StemmerPatchTrieBinaryIO.write(updated,
-                Path.of("english-custom.radixor.gz"));
-    }
-}
-```
-
-
-
-## Thread safety
-
-* `FrequencyTrie` (compiled):
-
-  * **thread-safe**
-  * safe for concurrent reads
-
-* `FrequencyTrie.Builder`:
-
-  * **not thread-safe**
-  * intended for single-threaded construction
-
-
-
-## Performance characteristics
-
-### Querying
-
-* O(length of word)
-* minimal allocations
-* suitable for high-throughput pipelines
-
-### Loading
-
-* binary loading is fast
-* no preprocessing required
-
-### Building
-
-* depends on dictionary size
-* reduction phase may be CPU-intensive
-
-
-
-## Best practices
-
-### Reuse compiled trie instances
-
-* load once
-* share across threads
-
-### Prefer binary loading in production
-
-* avoid rebuilding at runtime
-* treat compiled files as deployable artifacts
-
-### Use `getAll()` only when needed
-
-* `get()` is faster and sufficient for most use cases
-
-### Keep builders short-lived
-
-* build → compile → discard
-
-
-
-## Integration patterns
-
-### Search systems
-
-* apply stemming during indexing and querying
-* ensure consistent dictionary usage
-
-### Text normalization pipelines
-
-* integrate as a transformation step
-* combine with tokenization and filtering
-
-### Domain adaptation
-
-* extend dictionaries with domain-specific vocabulary
-* rebuild compiled artifacts
-
+For most developers, the best order is:
 
+1. [Fast Track](fast-track.md)
+2. [Integration Deep Dive](integration-deep-dive.md)
+3. [Loading and Building Stemmers](programmatic-loading-and-building.md)
+4. [Querying and Ambiguity Handling](programmatic-querying-and-ambiguity.md)
+5. [Extending and Persisting Compiled Tries](programmatic-extending-and-persistence.md)
 
 ## Next steps
 
-* [Dictionary format](dictionary-format.md)
-* [CLI compilation](cli-compilation.md)
-* [Architecture and reduction](architecture-and-reduction.md)
-
-
-
-## Summary
-
-Programmatic usage of Radixor follows a clear pattern:
-
-* build or load a trie
-* query using patch commands
-* apply transformations
-
-The API is intentionally simple at the surface, while providing deeper control when needed for:
-
-* ambiguity handling
-* diagnostics
-* dictionary evolution
+- [Quick Start](quick-start.md)
+- [CLI compilation](cli-compilation.md)
+- [Dictionary format](dictionary-format.md)
+- [Architecture and reduction](architecture-and-reduction.md)

docs/quality-and-operations.md

@@ -1,317 +1,260 @@
 # Quality and Operations
 
-> ← Back to [README.md](../README.md)
+This document describes the engineering standards, quality posture, and operational model of **Radixor**.
 
-This document describes quality, testing, and operational practices for **Radixor**.
+It is intentionally broader than a test checklist. The purpose of the project is not only to provide a fast stemmer, but to provide one whose behavior is explainable, measurable, reproducible, and straightforward to audit. That objective influences both the implementation style and the surrounding operational practices.
 
-It focuses on:
+## Engineering position
 
-- reliability and determinism
-- testing strategies
-- deployment patterns
-- performance considerations
-- lifecycle management of stemmer data
+Radixor is developed with a strong preference for objective quality signals over informal confidence.
 
+In practical terms, that means the project emphasizes:
 
+- deterministic behavior,
+- reproducible compiled artifacts,
+- very high structural test coverage,
+- very high mutation resistance,
+- explicit benchmark methodology,
+- minimal operational ambiguity in deployment.
 
-## Overview
+This is not treated as a cosmetic quality layer added after the implementation. It is part of the design goal of the project itself.
 
-Radixor is designed to separate:
+## Why quality discipline matters here
 
-- **data preparation** (dictionary construction and compilation)
-- **runtime execution** (lookup and patch application)
+A stemmer can appear deceptively simple from the outside. In practice, however, correctness depends on several interacting layers:
 
-This separation enables:
-
-- predictable runtime behavior
-- reproducible builds
-- controlled evolution of stemming data
+- dictionary parsing,
+- patch-command generation,
+- trie construction,
+- reduction semantics,
+- binary persistence,
+- runtime lookup behavior.
 
+A defect in any one of these layers can produce subtle and difficult-to-detect errors, including silent ranking drift, loss of ambiguity information, reconstruction inconsistencies, or incorrect stemming outcomes under only a narrow subset of inputs.
 
+For that reason, Radixor aims to be validated not only by example-based tests, but by a broader quality model that combines functional testing, mutation testing, coverage analysis, benchmark visibility, and artifact publication.
 
 ## Determinism and reproducibility
 
-Radixor emphasizes deterministic behavior.
+Determinism is a foundational property of the project.
 
-### Deterministic outputs
+Given the same dictionary input and the same reduction settings, the project aims to produce:
 
-Given:
+- the same compiled trie semantics,
+- the same local value ordering,
+- the same observable `get()` and `getAll()` behavior,
+- the same persisted binary output structure in semantic terms.
 
-- the same dictionary input
-- the same reduction settings
+This matters for more than technical elegance. It enables:
 
-Radixor guarantees:
+- stable search behavior across deployments,
+- reproducible build outputs,
+- reliable regression analysis,
+- explainable differences when a dictionary or reduction setting changes.
 
-- identical compiled trie structure
-- identical value ordering
-- identical lookup results
+A deterministic system is easier to test, easier to reason about, and safer to integrate into production pipelines.
 
-### Why this matters
+## Test strategy
 
-- stable search behavior across deployments
-- reproducible builds
-- easier debugging and regression analysis
+The project is intended to maintain very high confidence in both core correctness and behavioral stability.
 
+The recommended execution strategy is defined by the tagged test profiles in [Test taxonomy and execution filtering](test-taxonomy-and-filtering.md). In practice, teams can execute profile tasks directly:
 
+- `./gradlew ciSmoke`: fast local/PR safety checks (`unit`, excluding `slow`; additionally excludes
+  `CompileIntegrationTest` as a defensive safeguard).
+- `./gradlew ciSlow`: enterprise heavy gate for all tests marked with `slow` (typically
+  production dictionary and large corpus verification). This should be used for scheduled/manual
+  hardening gates and not in standard release build.
+- `./gradlew ciCore`: behavioral coverage of trie and frequency-trie paths (`unit` + `property` where applicable)
+- `./gradlew ciIntegration`: pipeline and CLI integration path checks
+- `./gradlew ciCompat`: compatibility and regression verification for persisted artifacts
+- `./gradlew ciRelease`: full non-slow suite for release-confidence runs (all test tags except `slow`,
+  plus explicit name-based exclusion of `CompileIntegrationTest*` and
+  `StemmerPatchTrieLoaderTest$BundledDictionaryTests*` as additional guardrails)
+- `./gradlew ciNightly`: extended fuzz profile for robustness hardening
+- `./gradlew ci`: umbrella profile depending on smoke/core/integration/compat
 
-## Testing strategy
+## Test taxonomy and execution filtering
 
-### Unit testing
+The full tag taxonomy and executable filter examples are documented in
+[Test taxonomy and execution filtering](test-taxonomy-and-filtering.md).
 
-Core components should be tested independently:
+### Structural coverage
 
-- patch encoding and decoding
-- trie construction
-- reduction behavior
-- binary serialization and deserialization
+High code coverage is treated as a useful signal, but not as a sufficient goal on its own. Coverage is valuable only when the covered scenarios actually pressure the implementation in meaningful ways.
 
-### Dictionary validation tests
+In Radixor, strong coverage is expected across areas such as:
 
-A recommended pattern:
+- patch encoding and application,
+- mutable trie construction,
+- subtree reduction,
+- compiled trie lookup,
+- binary serialization and deserialization,
+- reconstruction from compiled state,
+- dictionary parsing and CLI behavior.
 
-1. load dictionary input
-2. compile trie
-3. re-apply all word → stem mappings
-4. verify that:
+### Mutation resistance
 
-- expected stem is present in `getAll()`
-- preferred result (`get()`) is correct when deterministic
+Mutation testing is especially important for this project because it helps distinguish superficial test execution from genuinely discriminating tests.
 
-This ensures:
+A project can report high line or branch coverage while still failing to detect semantically dangerous implementation drift. Mutation testing provides a stronger objective signal: whether the test suite actually notices meaningful behavioral changes.
 
-- no data loss during reduction
-- correctness of patch encoding
+For Radixor, very high mutation scores are therefore part of the intended engineering standard, not an optional vanity metric.
 
+### Boundary and negative-path validation
 
+The project also benefits from extensive negative and edge-case testing, for example around:
 
-## Regression testing
+- malformed patch commands,
+- missing or corrupt binary data,
+- invalid CLI arguments,
+- ambiguous mappings,
+- dominance-threshold edge conditions,
+- reconstruction of reduced compiled tries,
+- empty inputs and short words.
 
-Maintain a stable test dataset:
+These cases are important because many real integration failures occur at the boundary conditions, not in the central happy path.
 
-- representative vocabulary
-- edge cases (short words, long words, ambiguous forms)
+## Quality signals and published evidence
 
-Use it to:
+The project publishes durable quality artifacts through GitHub Pages so that important signals remain externally inspectable rather than existing only as transient CI output.
 
-- detect unintended changes
-- verify behavior after refactoring
-- validate reduction mode changes
+Those published surfaces include:
 
+- unit test results,
+- coverage reports,
+- mutation testing reports,
+- static analysis reports,
+- benchmark outputs,
+- software composition artifacts.
 
+This publication model improves transparency and makes it easier to inspect the project’s quality posture without having to reconstruct the CI environment locally.
 
-## Performance testing
+## Operational model
 
-Performance should be evaluated in terms of:
+Radixor is designed around a clean separation between preparation-time work and runtime execution.
 
-### Throughput
+### Preparation phase
 
-- words processed per second
+Preparation includes:
 
-### Latency
+- creating or refining dictionary data,
+- compiling the dictionary into a reduced read-only trie,
+- validating the resulting artifact,
+- persisting it as a deployable binary stemmer.
 
-- time per lookup
+### Runtime phase
 
-### Memory footprint
+Runtime usage is intentionally simpler:
 
-- size of compiled trie
-- runtime memory usage
+- load the compiled artifact,
+- reuse the resulting trie,
+- perform fast lookups and patch application,
+- avoid rebuilding or reparsing during live request handling.
 
-Benchmark with:
+This separation reduces startup unpredictability, keeps runtime behavior stable, and makes deployment artifacts explicit.
 
-- realistic token streams
-- production-like dictionaries
+## Production posture
 
+For production use, the preferred model is straightforward:
 
+1. prepare or refine the lexical resource,
+2. compile it offline,
+3. validate the resulting artifact,
+4. deploy the compiled binary,
+5. load it once and reuse it.
 
-## Deployment model
+This model has several advantages:
 
-### Recommended workflow
+- no runtime compilation cost,
+- no repeated parsing overhead,
+- clear versioning of stemming behavior,
+- better reproducibility across environments,
+- simpler operational diagnosis when results change.
 
-1. prepare dictionary data
-2. compile using CLI
-3. store `.radixor.gz` artifact
-4. deploy artifact with application
-5. load using `loadBinary(...)`
+## Auditability and dependency posture
 
-### Why this model
+Radixor deliberately avoids external runtime dependencies.
 
-- avoids runtime compilation overhead
-- reduces startup latency
-- ensures consistent behavior across environments
+That choice serves a practical engineering goal: the project should be easy to audit from both a correctness and a security perspective, without forcing downstream users to reason through a large dependency graph or a complex software supply chain for core functionality.
 
+A dependency-free core does not make a project automatically secure, but it does simplify several important activities:
 
+- source review,
+- behavioral auditing,
+- release inspection,
+- software composition analysis,
+- long-term maintenance.
 
-## Artifact management
+In operational terms, this means there is less hidden behavior outside the project’s own codebase and less need to evaluate third-party runtime libraries for the core implementation path.
 
-Compiled stemmers should be treated as versioned assets.
+## Security-minded operational guidance
 
-### Versioning
+The project’s operational simplicity should be preserved in deployment practice.
 
-- include version in filename or metadata
-- track dictionary source and reduction settings
+Recommended principles include:
 
-Example:
+- treat source dictionaries as controlled inputs,
+- generate compiled artifacts in known build environments,
+- version compiled artifacts explicitly,
+- avoid loading untrusted binary stemmer files,
+- keep benchmark, test, and quality outputs attached to the same revision that produced the artifact.
 
-```
-english-v1.2-ranked.radixor.gz
-```
+These practices support traceability and reduce ambiguity about what exactly is running in production.
 
-### Storage
+## Performance as a quality concern
 
-- store in repository or artifact storage
-- ensure consistent distribution across environments
+Performance is not isolated from quality; for Radixor, it is part of the project’s engineering contract.
 
+The benchmark suite exists to make throughput behavior measurable and historically visible. At the same time, benchmark interpretation must remain disciplined. Absolute numbers can vary by environment, especially when published through shared CI infrastructure. Sustained relative behavior and reproducible local benchmark methodology are more meaningful than one-off raw figures.
 
+This is why benchmarking belongs alongside testing and reporting rather than outside the quality discussion altogether.
 
-## Runtime usage
+## Operational observability
 
-### Loading
+Radixor itself is intentionally small and does not attempt to become an observability framework. Instead, integrations should provide the surrounding operational visibility that production systems require.
 
-- load once during application startup
-- reuse `FrequencyTrie` instance
+Typical integration-level observability includes:
 
-### Thread safety
+- reporting load failures,
+- monitoring startup artifact loading,
+- measuring lookup throughput in the host application,
+- tracking memory usage of loaded compiled tries,
+- optionally sampling ambiguity-heavy cases when `getAll()` is part of the application logic.
 
-- compiled trie is safe for concurrent access
-- no synchronization required for reads
+The project’s role is to remain deterministic and inspectable enough that such operational signals are meaningful.
 
-### Avoid repeated loading
+## What feedback is most valuable
 
-Do not:
+Feedback is especially valuable when it improves the objectivity or professional rigor of the project.
 
-- load trie per request
-- rebuild trie at runtime
+That includes, for example:
 
+- defects in behavioral correctness,
+- weaknesses in reduction semantics or edge-case handling,
+- benchmark methodology issues,
+- gaps in tests or mutation resistance,
+- ambiguities in published reports,
+- opportunities to improve auditability, reproducibility, or operational clarity.
 
+Project feedback is most useful when it helps strengthen the project as an implementation that can be trusted, reviewed, and maintained at a professional standard.
 
-## Memory considerations
+## Practical summary
 
-- compiled tries are compact but not negligible
-- size depends on:
-  - dictionary size
-  - reduction mode
+Radixor aims to combine:
 
-Recommendations:
+- strong algorithmic performance,
+- deterministic behavior,
+- very high validation standards,
+- transparent published quality evidence,
+- low operational ambiguity,
+- easy auditability of the core implementation.
 
-- monitor memory usage in production
-- choose reduction mode appropriately
+That combination is central to the identity of the project. The goal is not merely to be fast, but to be fast in a way that remains explainable, testable, reproducible, and professionally defensible.
 
+## Related documentation
 
-
-## Reduction mode in production
-
-Default recommendation:
-
-- use **ranked mode**
-
-Switch to other modes only when:
-
-- memory constraints are strict
-- multiple candidate results are not required
-
-Always validate behavior after changing reduction mode.
-
-
-
-## Dictionary lifecycle
-
-### Updating dictionaries
-
-When dictionary data changes:
-
-1. update source file
-2. recompile
-3. run validation tests
-4. deploy new artifact
-
-### Backward compatibility
-
-- changes in dictionary may affect stemming results
-- evaluate impact on search relevance
-
-
-
-## Observability
-
-Radixor itself does not provide observability features; integration should provide:
-
-- logging for loading failures
-- metrics for lookup throughput
-- monitoring of memory usage
-
-Optional:
-
-- sampling of ambiguous results (`getAll()`)
-
-
-
-## Error handling
-
-### During compilation
-
-Handle:
-
-- invalid dictionary format
-- I/O failures
-- invalid arguments
-
-### During runtime
-
-Handle:
-
-- missing dictionary files
-- corrupted binary artifacts
-
-Fail fast on initialization errors.
-
-
-
-## Operational best practices
-
-- compile dictionaries offline
-- version compiled artifacts
-- test before deployment
-- load once and reuse
-- monitor performance and memory
-- document reduction settings used
-
-
-
-## Security considerations
-
-- treat dictionary input as trusted data
-- validate external sources before compilation
-- avoid loading unverified binary artifacts
-
-
-
-## Integration checklist
-
-Before production deployment:
-
-- dictionary validated
-- compiled artifact generated
-- reduction mode documented
-- performance tested
-- memory usage verified
-- regression tests passing
-
-
-
-## Next steps
-
-- [Quick start](quick-start.md)
+- [Benchmarking](benchmarking.md)
+- [Reports](reports.md)
 - [CLI compilation](cli-compilation.md)
 - [Programmatic usage](programmatic-usage.md)
-
-
-
-## Summary
-
-Radixor is designed for:
-
-- deterministic behavior
-- efficient runtime execution
-- controlled data-driven evolution
-
-By separating compilation from runtime and following proper operational practices, it can be reliably integrated into production-grade systems.

docs/quick-start.md

@@ -1,104 +1,193 @@
 # Quick Start
 
-> ← Back to [README.md](../README.md)
+This guide introduces the fastest practical path to using **Radixor**.
 
-This guide shows the fastest way to start using **Radixor** and the most common next steps.
+If you are new to Radixor and want the shortest possible path to a first working stem, start with
+[Fast Track](fast-track.md). This Quick Start is a broader developer walkthrough: it introduces the
+main loading options, query methods, artifact workflow, and metadata model.
 
-## Hello world
+Radixor separates preparation from runtime usage. Source dictionaries are used to derive patch commands and reduce them into a compact read-only trie. Runtime stemming then operates on that compiled structure rather than on the original dictionary text. A richer dictionary usually improves the quality and coverage of inferred transformations, including transformations that are applicable to words not explicitly present in the source material. The reduction step also removes a large amount of redundant lexical information, which is why very large dictionaries can still produce compact runtime artifacts. These artifacts can be persisted and loaded directly when needed.
+
+A practical workflow usually consists of two independent phases:
+
+1. obtain a compiled stemmer,
+2. use the compiled stemmer.
+
+## 1. Obtain a compiled stemmer
+
+A compiled stemmer can be obtained in three common ways.
+
+### Use a bundled language dictionary
+
+Radixor ships with bundled dictionaries for a set of supported languages. These resources are line-oriented dictionaries stored with the library and compiled into a `FrequencyTrie<CompiledPatchCommand>` when loaded through the runtime API. The loader can also store the canonical stem itself as a no-op patch command. Compiled trie artifacts now persist self-describing metadata, including the traversal direction and compilation reduction settings used to build the artifact.
 
 ```java
 import java.io.IOException;
 
+import org.egothor.stemmer.CompiledPatchCommand;
 import org.egothor.stemmer.FrequencyTrie;
-import org.egothor.stemmer.PatchCommandEncoder;
 import org.egothor.stemmer.ReductionMode;
 import org.egothor.stemmer.StemmerPatchTrieLoader;
 
-public final class HelloRadixor {
+public final class BundledStemmerExample {
 
-    private HelloRadixor() {
+    private BundledStemmerExample() {
         throw new AssertionError("No instances.");
     }
 
     public static void main(final String[] arguments) throws IOException {
-        final FrequencyTrie<String> trie = StemmerPatchTrieLoader.load(
-                StemmerPatchTrieLoader.Language.US_UK_PROFI,
+        final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadCompiled(
+                StemmerPatchTrieLoader.Language.US_UK,
                 true,
                 ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
 
-        final String word = "running";
-        final String patch = trie.get(word);
-        final String stem = PatchCommandEncoder.apply(word, patch);
-
-        System.out.println(word + " -> " + stem);
+        System.out.println("Canonical node count: " + trie.size());
     }
 }
 ```
 
-This example shows the core workflow:
+### Load a previously compiled binary stemmer
 
-1. load a trie
-2. get a patch command for a word
-3. apply the patch
-4. obtain the stem
-
-## Retrieve multiple candidate stems
-
-If you need more than one candidate result, use `getAll(...)` instead of `get(...)`.
-
-```java
-final String word = "axes";
-final String[] patches = trie.getAll(word);
-
-for (String patch : patches) {
-    final String stem = PatchCommandEncoder.apply(word, patch);
-    System.out.println(word + " -> " + stem + " (" + patch + ")");
-}
-```
-
-## Load a compiled binary stemmer
-
-For production systems, the preferred approach is usually to precompile the dictionary and load the compressed binary artifact at runtime.
+Compiled stemmers can be stored as GZip-compressed binary artifacts and loaded directly. This is usually the most convenient production path because no dictionary parsing or recompilation is needed during application startup.
 
 ```java
 import java.io.IOException;
 import java.nio.file.Path;
 
+import org.egothor.stemmer.CompiledPatchCommand;
 import org.egothor.stemmer.FrequencyTrie;
-import org.egothor.stemmer.PatchCommandEncoder;
 import org.egothor.stemmer.StemmerPatchTrieLoader;
 
-public final class BinaryStemmerExample {
+public final class LoadBinaryStemmerExample {
 
-    private BinaryStemmerExample() {
+    private LoadBinaryStemmerExample() {
         throw new AssertionError("No instances.");
     }
 
     public static void main(final String[] arguments) throws IOException {
-        final Path path = Path.of("stemmers", "english.radixor.gz");
-        final FrequencyTrie<String> trie = StemmerPatchTrieLoader.loadBinary(path);
+        final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadBinaryCompiled(
+                Path.of("stemmers", "english.radixor.gz"));
 
-        final String word = "connected";
-        final String patch = trie.get(word);
-        final String stem = PatchCommandEncoder.apply(word, patch);
-
-        System.out.println(word + " -> " + stem);
+        System.out.println("Canonical node count: " + trie.size());
     }
 }
 ```
 
-## Compile a dictionary from the command line
+You can tune in-memory child lookup density at load time without changing the artifact:
 
-```bash
-java org.egothor.stemmer.Compile \
-    --input ./data/stemmer.txt \
-    --output ./build/english.radixor.gz \
-    --reduction-mode MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS \
-    --store-original \
-    --overwrite
+```java
+import java.io.IOException;
+import java.nio.file.Path;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+public final class LoadBinaryStemmerExampleTuned {
+
+    private LoadBinaryStemmerExampleTuned() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<CompiledPatchCommand> fast = StemmerPatchTrieLoader.loadBinaryCompiled(
+                Path.of("stemmers", "english.radixor.gz"),
+                1024);
+        final FrequencyTrie<CompiledPatchCommand> compact = StemmerPatchTrieLoader.loadBinaryCompiled(
+                Path.of("stemmers", "english.radixor.gz"),
+                128);
+
+        System.out.println("fast=" + fast.size() + ", compact=" + compact.size());
+    }
+}
 ```
 
-## Modify an existing compiled stemmer
+For the trade-off details, see [Lookup Edge Optimization](lookup-edge-optimization.md).
+
+### Build or extend a stemmer from dictionary data
+
+Radixor can also build a compiled trie from a custom dictionary. Dictionary lines consist of a canonical stem followed by zero or more variants. The input may be plain UTF-8 text or GZip-compressed UTF-8 text when loaded from a filesystem path. The parser applies `CaseProcessingMode` (default: `LOWERCASE_WITH_LOCALE_ROOT`), ignores leading and trailing whitespace around columns, supports line remarks introduced by `#` or `//`, and skips dictionary items that contain embedded whitespace.
+
+This path is also relevant when you extend an existing compiled stemmer with additional domain-specific entries and rebuild a new compact artifact.
+
+A dedicated CLI compilation workflow deserves its own focused page and should remain separate from Quick Start, but conceptually it is simply another way to prepare the compiled artifact before runtime use.
+
+## 2. Use the compiled stemmer
+
+A compiled `FrequencyTrie<CompiledPatchCommand>` stores patch commands, not final stems. Querying therefore has two steps:
+
+1. retrieve one or more patch commands from the trie,
+2. apply each patch command to the original input word.
+
+The trie returns values associated with the exact addressed node. `get(...)` returns the locally preferred value, while `getAll(...)` returns all locally stored values ordered by descending frequency with deterministic tie-breaking.
+
+### Get the preferred result
+
+Use `get(...)` when the application needs a single preferred transformation.
+
+```java
+import java.io.IOException;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+public final class SingleStemExample {
+
+    private SingleStemExample() {
+        throw new AssertionError("No instances.");
+    }
+
+    public static void main(final String[] arguments) throws IOException {
+        final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadCompiled(
+                StemmerPatchTrieLoader.Language.US_UK,
+                true,
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+
+        final String word = "running";
+        final CompiledPatchCommand patch = trie.get(word);
+        final String stem = patch == null ? word : patch.apply(word);
+
+        System.out.println(word + " -> " + stem + " (" + patch + ")");
+    }
+}
+```
+
+### Get all candidate results
+
+Use `getAll(...)` when the application should preserve ambiguity instead of collapsing everything into one result. The method is available on every compiled trie. What changes across reduction modes is the semantic strength with which multi-result behavior is preserved during reduction, not whether the method exists.
+
+```java
+final String word = "axes";
+final CompiledPatchCommand[] patches = trie.getAll(word);
+
+for (final CompiledPatchCommand patch : patches) {
+    final String stem = patch.apply(word);
+    System.out.println(word + " -> " + stem + " (" + patch + ")");
+}
+```
+
+### Inspect ranked values and counts
+
+For diagnostics or advanced ranking logic, use `getEntries(...)` to obtain value-count pairs in the same deterministic order as `getAll(...)`.
+
+```java
+import java.util.List;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.ValueCount;
+
+final List<ValueCount<CompiledPatchCommand>> entries = trie.getEntries("axes");
+
+for (final ValueCount<CompiledPatchCommand> entry : entries) {
+    System.out.println(entry.value() + " -> " + entry.count());
+}
+```
+
+## Extend an existing compiled stemmer
+
+A compiled trie is read-only, but it is not permanently closed. Radixor can reconstruct a mutable builder from a compiled trie, preserve the currently stored local counts, accept additional insertions, and then compile a new read-only trie. Reconstruction operates on the compiled form, so if the source trie was already reduced by subtree merging, the reopened builder reflects that compiled state rather than the original unreduced insertion history.
 
 ```java
 import java.io.IOException;
@@ -111,17 +200,15 @@ import org.egothor.stemmer.ReductionMode;
 import org.egothor.stemmer.ReductionSettings;
 import org.egothor.stemmer.StemmerPatchTrieBinaryIO;
 
-public final class ModifyCompiledExample {
+public final class ExtendCompiledStemmerExample {
 
-    private ModifyCompiledExample() {
+    private ExtendCompiledStemmerExample() {
         throw new AssertionError("No instances.");
     }
 
     public static void main(final String[] arguments) throws IOException {
-        final Path input = Path.of("stemmers", "english.radixor.gz");
-        final Path output = Path.of("stemmers", "english-custom.radixor.gz");
-
-        final FrequencyTrie<String> compiledTrie = StemmerPatchTrieBinaryIO.read(input);
+        final FrequencyTrie<String> compiledTrie = StemmerPatchTrieBinaryIO.read(
+                Path.of("stemmers", "english.radixor.gz"));
 
         final ReductionSettings settings = ReductionSettings.withDefaults(
                 ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
@@ -131,18 +218,36 @@ public final class ModifyCompiledExample {
                 String[]::new,
                 settings);
 
-        builder.put("microservices", PatchCommandEncoder.NOOP_PATCH);
+        final PatchCommandEncoder encoder = PatchCommandEncoder.builder()
+                .traversalDirection(compiledTrie.traversalDirection())
+                .build();
+
+        builder.put("microservices", encoder.encode("microservices", "microservice"));
 
         final FrequencyTrie<String> updatedTrie = builder.build();
-        StemmerPatchTrieBinaryIO.write(updatedTrie, output);
+
+        StemmerPatchTrieBinaryIO.write(
+                updatedTrie,
+                Path.of("stemmers", "english-custom.radixor.gz"));
     }
 }
 ```
 
+## Operational note on memory and preparation
+
+Dictionary compilation is usually a one-time preparation step and is generally fast. The more relevant operational constraint is memory consumption during preparation: before reduction, the mutable build-time structure keeps the full dictionary-derived content in RAM. Reduction then compacts it substantially, but very large source dictionaries can still require significant memory during the initial build phase. The best operational model is therefore to compile once, persist the resulting binary artifact, and load that artifact directly in runtime environments.
+
 ## Where to continue
 
-* [Dictionary format](dictionary-format.md)
-* [CLI compilation](cli-compilation.md)
-* [Programmatic usage](programmatic-usage.md)
-* [Built-in languages](built-in-languages.md)
-* [Architecture and reduction](architecture-and-reduction.md)
+- [Programmatic Usage](programmatic-usage.md)
+- [Dictionary format](dictionary-format.md)
+- [CLI compilation](cli-compilation.md)
+- [Built-in languages](built-in-languages.md)
+- [Architecture and reduction](architecture-and-reduction.md)
+
+
+## Persisted trie metadata
+
+Every compiled trie artifact stores a `TrieMetadata` descriptor together with the immutable trie payload. That metadata currently records the binary format version, the `WordTraversalDirection`, the `ReductionSettings` used during compilation, the declared `DiacriticProcessingMode`, and the selected `CaseProcessingMode`. Traversal, case processing, and diacritic processing are applied during runtime lookup (`get`, `getAll`), and case/diacritic processing are also applied during dictionary insertion when a trie is built.
+
+`DiacriticProcessingMode.AS_IS` keeps dictionary keys and lookup keys unchanged. `DiacriticProcessingMode.REMOVE` strips diacritics from dictionary keys and lookup keys (for Czech diacritics and broad European Latin-script variants). `DiacriticProcessingMode.AS_IS_AND_STRIPPED_FALLBACK` is currently not supported and raises an `UnsupportedOperationException`.

docs/reduction-semantics.md

@@ -0,0 +1,226 @@
+# Reduction Semantics
+
+This document explains how **Radixor** decides that two subtrees are equivalent, how the different reduction modes work, and how those choices affect observable runtime behavior.
+
+## Why reduction exists
+
+Without reduction, the trie would still work, but many subtrees that mean the same thing would remain duplicated. The result would be a much larger runtime artifact than necessary.
+
+Reduction solves that by merging semantically equivalent subtrees into one canonical representative.
+
+The key idea is simple:
+
+> if two subtrees behave the same way under the semantic contract chosen for compilation, only one physical copy is needed.
+
+## Reduction is semantic, not merely structural
+
+Radixor does not reduce nodes merely because they look similar locally. It reduces subtrees only when their **meaning** matches according to the selected mode.
+
+That is why reduction is based on a **signature** that captures both:
+
+1. the local semantics of the current node,
+2. the structure and semantics of all descendant edges.
+
+Conceptually:
+
+```text
+Signature = (LocalDescriptor, SortedChildDescriptors)
+```
+
+Two subtrees are merged only if their signatures are equal.
+
+## Local descriptors
+
+The local descriptor defines what “equivalent” means for the values stored at one node.
+
+Radixor supports three semantic views.
+
+### Ranked descriptor
+
+The ranked descriptor preserves the full ordered result semantics of `getAll()`.
+
+That means:
+
+- candidate membership is preserved,
+- local ordering is preserved,
+- observable ranked multi-result behavior remains stable.
+
+This is the most semantically faithful mode when ambiguity handling matters.
+
+### Unordered descriptor
+
+The unordered descriptor preserves the set of reachable results, but not their local ordering.
+
+That means:
+
+- candidate membership is preserved,
+- ordering differences may be ignored,
+- more subtrees can be merged than in ranked mode.
+
+This mode is useful when alternative candidates matter but exact ranking does not.
+
+### Dominant descriptor
+
+The dominant descriptor focuses on the preferred result returned by `get()`.
+
+This mode is used only when the dominant local candidate is strong enough according to configured thresholds:
+
+- minimum winner percentage,
+- winner-over-second ratio.
+
+If that local dominance is not strong enough, Radixor does not force dominant semantics anyway. It falls back to ranked semantics for that node to avoid unsafe over-reduction.
+
+That fallback is one of the most important safeguards in the design.
+
+## Child descriptors
+
+A subtree is not defined only by the values stored at the current node. It is also defined by what behavior is reachable through its children.
+
+Each child contributes:
+
+```text
+(edge character, child signature)
+```
+
+Children are sorted by edge character so that signatures remain deterministic and stable.
+
+This matters because reduction must not depend on incidental map iteration order or other non-semantic implementation details.
+
+## Canonicalization
+
+Once a subtree signature is computed, the reduction process checks whether an equivalent canonical subtree already exists.
+
+If yes, the existing reduced node is reused.
+
+If no, a new canonical reduced node is created and registered.
+
+This turns reduction into a canonicalization process:
+
+- compute semantic identity,
+- find canonical representative,
+- reuse or create,
+- continue bottom-up.
+
+That is how Radixor eliminates duplicated equivalent subtrees.
+
+## Uniform-subtree contraction
+
+Radixor performs one additional internal reduction before each public reduction mode is applied.
+When all reachable entries below a subtree have the same preferred patch command, the subtree is
+contracted into an accepting leaf for that command.
+
+This optimization is deliberately narrower than the public reduction modes:
+
+- it is based on preferred `get()` behavior,
+- it does not depend on child edge shape once the preferred command is uniform,
+- it removes lookup depth that cannot affect the selected command,
+- it preserves the standard single-result stemming path used by `StemmerPatchTrieLoader.loadCompiled(...)`.
+
+The effect is especially visible in large dictionary tries with many inflected forms that map to
+the same command class, such as no-op preservation or common suffix deletion. Runtime lookup can
+return the accepting leaf as soon as it is reached instead of traversing the remaining characters
+only to discover the same command deeper in the trie.
+
+## Count aggregation and compiled state
+
+When multiple original build-time subtrees collapse into one canonical reduced node, local counts may be aggregated.
+
+This is an important point for understanding compiled artifacts.
+
+A compiled trie is not always a verbatim replay of original insertion history. It is a canonical runtime structure that preserves the semantics guaranteed by the chosen reduction mode.
+
+This explains two things:
+
+- why compiled artifacts can become dramatically smaller,
+- why reconstructing a builder from a compiled trie reflects the compiled state rather than the full original unreduced history.
+
+## Reduction modes
+
+### `MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS`
+
+This mode merges subtrees only when their `getAll()` results are equivalent for every reachable key suffix and when local ordering is preserved.
+
+Use this mode when:
+
+- ambiguity handling matters,
+- `getAll()` ordering should remain meaningful,
+- behavioral fidelity is more important than maximum compression.
+
+This is the safest and most generally recommended mode.
+
+### `MERGE_SUBTREES_WITH_EQUIVALENT_UNORDERED_GET_ALL_RESULTS`
+
+This mode also preserves `getAll()`-level membership equivalence for every reachable key suffix, but it ignores local ordering differences.
+
+Use this mode when:
+
+- alternative candidates still matter,
+- exact ordering is less important,
+- stronger reduction is acceptable.
+
+This mode is more aggressive than ranked mode, but less semantically rich.
+
+### `MERGE_SUBTREES_WITH_EQUIVALENT_DOMINANT_GET_RESULTS`
+
+This mode focuses on preserving dominant `get()` semantics for every reachable key suffix, subject to dominance thresholds.
+
+Use this mode when:
+
+- the main operational concern is the preferred result,
+- richer alternative-result behavior is less important,
+- stronger reduction is desirable.
+
+Because non-dominant nodes fall back to ranked semantics, this mode is not simply “discard everything except the winner”. It is a controlled reduction strategy with a built-in safety condition.
+
+## Practical effect on runtime behavior
+
+Reduction mode is not just a storage optimization setting. It affects what distinctions remain visible after compilation.
+
+### When ranked mode is used
+
+You can rely on full ranked `getAll()` semantics being preserved.
+
+### When unordered mode is used
+
+You can rely on candidate membership, but not necessarily on preserving the same local ranking distinctions.
+
+### When dominant mode is used
+
+You optimize primarily for preferred-result semantics. Alternative-result behavior may still exist, but it is no longer the primary semantic contract of the reduction.
+
+## Choosing a mode
+
+A practical rule of thumb is:
+
+- choose **ranked** if you are unsure,
+- choose **unordered** if alternative membership matters but ranking does not,
+- choose **dominant** only when your application is fundamentally driven by `get()` and you understand the trade-off.
+
+## Why this design works well
+
+The reduction model succeeds because it does not confuse “smaller” with “acceptable”.
+
+Instead, it makes the semantic contract explicit:
+
+- what exactly must be preserved,
+- what differences may be ignored,
+- when a more aggressive mode is safe,
+- when the system must fall back to a stricter interpretation.
+
+That explicitness is what makes the compression trustworthy.
+
+## Mental model to keep
+
+If you want one concise mental model for reduction, use this one:
+
+- build-time insertion collects examples,
+- reduction asks which subtrees mean the same thing,
+- the answer depends on the chosen semantic contract,
+- canonical representatives are shared,
+- the compiled trie preserves the behavior promised by that contract.
+
+## Continue with
+
+- [Architecture](architecture.md)
+- [Programmatic usage](programmatic-usage.md)
+- [CLI compilation](cli-compilation.md)

docs/reports.md

@@ -0,0 +1,60 @@
+# Reports and Published Build Artifacts
+
+Radixor publishes durable build outputs to GitHub Pages from qualifying runs of `.github/workflows/pages.yml`.
+
+This page is the central entry point for published project artifacts, including build summaries, API documentation, test and quality reports, benchmark outputs, and software composition materials. It is intended both for routine project inspection and for linking stable report surfaces from external references such as the README, release notes, or development workflows.
+
+## Stable entry points
+
+The following links are the primary stable locations for the most recent published build outputs:
+
+- [Latest build summary](https://leogalambos.github.io/Radixor/builds/latest/)
+- [Browse historical build reports](https://leogalambos.github.io/Radixor/builds/)
+
+Use `builds/latest/` when you want the current published report surface. Use `builds/` when you need to inspect or compare retained historical runs.
+
+## API and developer documentation
+
+These reports are primarily useful when reviewing the published API surface and generated developer-facing documentation:
+
+- [Javadoc](https://leogalambos.github.io/Radixor/builds/latest/javadoc/)
+
+## Verification and code quality reports
+
+These reports describe the outcome of core verification and static-analysis stages for the latest published build:
+
+- [Release verification test report (ciRelease)](https://leogalambos.github.io/Radixor/builds/latest/test/)
+- [PMD report](https://leogalambos.github.io/Radixor/builds/latest/pmd/main.html)
+- [JaCoCo coverage report](https://leogalambos.github.io/Radixor/builds/latest/coverage/)
+- [PIT mutation testing report](https://leogalambos.github.io/Radixor/builds/latest/pitest/)
+- [Dependency vulnerability report](https://leogalambos.github.io/Radixor/builds/latest/dependency-check/dependency-check-report.html)
+
+Together, these reports provide the most direct published view of functional correctness, static quality signals, coverage, mutation resistance, and dependency-level security review outputs.
+
+## Software composition artifacts
+
+These artifacts expose the published software bill of materials for the latest build:
+
+- [SBOM (JSON)](https://leogalambos.github.io/Radixor/builds/latest/sbom/radixor-sbom.json)
+- [SBOM (XML)](https://leogalambos.github.io/Radixor/builds/latest/sbom/radixor-sbom.xml)
+
+They are useful for dependency inspection, downstream integration, compliance-oriented workflows, and artifact traceability.
+
+## Benchmark outputs and badge metadata
+
+These resources expose benchmark results and generated badge metadata derived from the latest published build. JMH benchmark reports are published as TXT and CSV files; the historical Porter comparison badge is no longer generated.
+
+- [JMH benchmark results (TXT)](https://leogalambos.github.io/Radixor/builds/latest/jmh/jmh-results.txt)
+- [JMH benchmark results (CSV)](https://leogalambos.github.io/Radixor/builds/latest/jmh/jmh-results.csv)
+- [Coverage badge metadata](https://leogalambos.github.io/Radixor/builds/latest/metrics/coverage-badge.json)
+- [Mutation badge metadata](https://leogalambos.github.io/Radixor/builds/latest/metrics/pitest-badge.json)
+
+The benchmark outputs provide direct access to the published JMH result files. Coverage and mutation badge metadata endpoints are intended for status surfaces such as the project README or other generated dashboards.
+
+## Practical usage
+
+In most cases, the recommended entry path is:
+
+1. start with the [Latest build summary](https://leogalambos.github.io/Radixor/builds/latest/),
+2. open the specific report category relevant to your task,
+3. use [Browse historical build reports](https://leogalambos.github.io/Radixor/builds/) when historical inspection is needed.

docs/stemming-quality.md

@@ -0,0 +1,82 @@
+# Stemming quality evaluation
+
+The explicit `stemmingQuality` analysis measures agreement between stemmer outputs and gold-standard equivalence classes represented by bundled multilingual dictionary rows. Dictionary text remains unchanged; reports and diagnostics use English.
+
+JMH adapters, registries, third-party versions, language mappings, and preparation remain in `src/jmh`. The evaluator, reports, audits, and tests reside in the standard `src/test` source set. The former `src/stemmingQualityTest` source set was removed, and neither analytical nor JMH classes enter the production JAR.
+
+## Language and adapter coverage
+
+The authoritative Radixor universe is the validated one-to-one reconciliation of `src/main/resources/*/stemmer.gz` and every `StemmerPatchTrieLoader.Language` value. All 20 current values have exactly one resource; no sentinel or alias is excluded. Radixor is evaluated for all 20 languages, independently of third-party support. Third-party combinations come only from explicit JMH adapter metadata. Unsupported combinations are documented and never fabricated as zero-valued rows.
+
+The expected matrix is constructed before evaluation from stemmer, language, dictionary mode, and supported output policy. Generation fails on missing, duplicate, unexpected, or stale keys.
+
+## Dictionary groups and modes
+
+Every usable parsed row is one gold-standard group. Exact duplicate strings are removed only within that row; identical forms in different rows remain distinct. `ALL_WORDS` preserves every valid form. `LOWERCASE_GROUPS_ONLY` excludes a complete group containing an uppercase or titlecase Unicode code point. Retained words are not lowercased or normalized by the evaluator.
+
+## Output policies
+
+`PRIMARY_OUTPUT` uses the deterministic JMH output and defines a strict partition.
+
+For multi-output adapters, `C(w)` is the immutable, sorted, exactly deduplicated candidate set. It is non-null, non-empty, contains no null, and contains the primary output. Radixor obtains alternatives through `getAll`. The repository's Morphologik lookups can return distinct lemma strings and are multi-output. Configured Hunspell filters can emit several stems at one token position. Other adapters emit only primary rows.
+
+`ANY_CANDIDATE` is an optimistic oracle-assisted pairwise upper bound. A same-group pair succeeds when its sets intersect. A cross-group pair is an error only when both sets are the same singleton; otherwise unequal candidates can be selected for that pair. Choices may vary between pairs and need not form one realizable global assignment.
+
+`ALL_CANDIDATES` activates every candidate. Two forms are related when their sets intersect, for both same-group and cross-group pairs. This relation can overlap and need not be transitive. A pair sharing several candidates is counted once.
+
+The evaluator verifies:
+
+```text
+ANY under <= PRIMARY under
+ALL under <= PRIMARY under
+ANY under = ALL under
+ANY over <= PRIMARY over
+ALL over >= PRIMARY over
+```
+
+## Pair definitions and efficient counting
+
+For `C2(n) = n(n-1)/2`:
+
+```text
+underPossible = sum_g C2(n_g)
+overPossible = C2(N) - sum_g C2(n_g)
+```
+
+Under-stemming counts unrelated same-group pairs. Over-stemming counts related cross-group pairs. Primary output uses global and per-group stem frequencies. Candidate sets are canonical signatures counted globally and per group. An inverted candidate-to-signature index discovers intersections, and signature pairs shared through several candidates are deduplicated. `ANY_CANDIDATE` over-stemming uses only equal singleton signatures. All pair arithmetic uses checked `long` operations; complete production word pairs are never enumerated.
+
+## Confusion and aggregate metrics
+
+```text
+TP = underPossible - underError
+FN = underError
+FP = overError
+TN = overPossible - overError
+```
+
+Under-stemming is `FN/(TP+FN)` and over-stemming is `FP/(TN+FP)`; their denominators differ. The CSV also publishes precision, recall, specificity, accuracy, balanced accuracy, F0.5, F1, F2, Jaccard, Fowlkes-Mallows, Matthews correlation coefficient, and pairwise error rate. F0.5 emphasizes precision and over-stemming, F1 balances precision and recall, and F2 emphasizes recall and under-stemming. Accuracy and error rate can be dominated by the large cross-group true-negative population. Metrics use raw counts, not rounded rates. Zero denominators produce `n/a` in Markdown and empty CSV fields.
+
+Only `PRIMARY_OUTPUT` receives partition metrics: Adjusted Rand Index, homogeneity, completeness, V-measure, and normalized mutual information with arithmetic-mean entropy normalization. Candidate policies remain inapplicable rather than being forced into artificial partitions.
+
+Micro summaries sum confusion counts before calculation. Macro summaries average defined language values and retain coverage counts. Common-language comparisons use the exact language intersection and never score unsupported languages as zero. Rankings are separated by policy and metric; the default F0.5 choice is navigation, not a universal scientific preference.
+
+Pearson and average-tie-rank Spearman reports use unrounded values and separate dictionary-mode and output-policy cohorts. Fewer than three observations, undefined inputs, and zero variance produce documented missing values. The reports provide reproducible data and make no automatic scientific conclusion.
+
+## Exact accuracy and pairwise under-stemming
+
+Exact textual accuracy and pairwise grouping use different denominators. One erroneous form in a 12-form group creates 11 erroneous pairs: with 88 singleton groups, exact accuracy can be 99% while pairwise under-stemming is `11/C2(12) = 16.666667%`. Singleton groups affect word accuracy but add no within-group pairs.
+
+## Running the analysis
+
+```bash
+./gradlew stemmingQuality
+./gradlew stemmingQuality -PstemmingQualityStemmer=Radixor -PstemmingQualityLanguage=DE_DE -PstemmingQualityMode=ALL_WORDS -PstemmingQualityAudit=true
+```
+
+Optional properties are `stemmingQualityLanguage`, `stemmingQualityStemmer`, `stemmingQualityMode`, `stemmingQualityOutputPolicy`, `stemmingQualityRankMetric`, `stemmingQualityAudit`, and `stemmingQualityAuditLimit`. Policies are `PRIMARY_OUTPUT`, `ANY_CANDIDATE`, and `ALL_CANDIDATES`. Filtered reports carry `-filtered` and cannot overwrite complete output.
+
+Generated files under `build/reports/stemming-quality/` include `stemming-quality.md`, `stemming-quality.csv`, `metric-correlations-pearson.csv`, `metric-correlations-spearman.csv`, and optional audit Markdown.
+
+## Limitations
+
+These measurements evaluate agreement with the available dictionary grouping. They do not capture every semantic, morphological, downstream, or dataset-specific property. `ANY_CANDIDATE` is optimistic and may not be globally realizable. `ALL_CANDIDATES` measures an overlap graph rather than a partition. Language coverage must remain visible in cross-stemmer comparisons. No single published metric establishes universal superiority; multiple metrics and their correlations are provided for transparent scientific assessment.

docs/test-taxonomy-and-filtering.md

@@ -0,0 +1,224 @@
+# Test Tag Taxonomy and Execution Guide
+
+Radixor uses JUnit tags as an explicit execution policy for its test suite.
+
+The project uses three orthogonal axes:
+
+1. **Scope** (how the test is executed in the pipeline)
+2. **Domain** (where in the system it belongs)
+3. **Intent** (what behavior it verifies)
+
+## Canonical scope tags
+
+| Tag | Description | Typical usage |
+| --- | --- | --- |
+| `unit` | Fast, deterministic tests that exercise a specific class or behavior without external processes. | Default developer feedback; should stay near-zero flakiness and low run time. |
+| `integration` | Tests that span multiple components or end-to-end flows of the public pipeline. | Parser/loader/CLI/IO integration checks and multi-step compile-then-load validations. |
+| `property` | Property-based tests with generator-driven coverage for invariants. | Semantics-preserving laws and edge-case exploration beyond curated fixtures. |
+| `fuzz` | Randomized stress checks with bounded runtime. | Heavier probabilistic verification of robustness and reduction invariants. |
+| `compat` | Backward/forward compatibility and reproducibility checks for persisted artifacts. | Artifact fingerprints, deterministic rebuild, and regression fixtures. |
+| `slow` | Long-running or expensive tests that should not execute in every fast gate. | Heavy fuzz/property budgets or high-duration integration checks. |
+
+## Canonical domain tags
+
+| Tag | Description | Typical usage |
+| --- | --- | --- |
+| `core` | Core algorithm and foundational platform behavior. | Traversal direction, base data structures, low-level helpers. |
+| `trie` | All mutable/compiled trie behaviors and traversal internals. | Lookup path selection, node shape, child representation, subtree behavior. |
+| `frequency-trie` | Algorithms and corner cases specific to frequency-aware trie logic. | Ranking, weighted reductions, persistence of weighted nodes. |
+| `stemmer` | End-user stemming pipeline semantics. | Parse-encode-apply flows and output invariants. |
+| `patch` | Patch encoding, decoding, and application semantics. | `PatchCommandEncoder` behavior and related compatibility contracts. |
+| `io` | Input/output and resource loading boundaries. | Filesystem readers, streams, and stream lifecycle handling. |
+| `serialization` | Binary persistence contract of compiled artifacts. | Versioned format reads/writes and checksum/consistency checks. |
+| `parser` | Dictionary and metadata parsing concerns. | Dictionary input parsing and malformed-source rejection. |
+| `cli` | Command-line entrypoint and command orchestration behavior. | Compile CLI integration and CLI argument validation. |
+| `metadata` | Trie metadata semantics, compatibility fields, and schema expectations. | Version flags, structural properties, and metadata round-trips. |
+| `compile` | Compile-time pipeline and build-oriented behavior. | Building, reduction-mode behavior, and compiled artifact generation. |
+| `diacritic` | Unicode diacritic normalization and stripping behavior. | Accent-removal correctness and locale-safe normalization checks. |
+
+## Canonical intent tags
+
+| Tag | Description | Typical usage |
+| --- | --- | --- |
+| `construction` | Tests around construction and assembly of runtime structures. | Builders, loaders, and compile-time object construction contracts. |
+| `lookup` | Read behavior and retrieval semantics. | `get()`, `getAll()`, traversal and missing-key behavior. |
+| `persistence` | Storage lifecycle semantics. | Serialization/deserialization and round-trip correctness. |
+| `reduction` | Reduction algorithm correctness and corner cases. | Dominance threshold, subtree deduplication, rank-preservation invariants. |
+| `encoding` | Encoding transformation direction. | `PatchCommandEncoder.encode` and serialized command form generation. |
+| `decoding` | Decoding/interpretation of persisted or runtime commands. | Optional consumers that parse and apply encoded command payloads. |
+| `apply` | Patch application and transformation behavior. | Verifies that applied patches produce expected derived forms. |
+| `normalization` | Canonicalization and cleanup behavior. | String normalization around case/shape and mirrored input paths. |
+| `validation` | Input rejection and defensive checks. | Null/empty/invalid contracts and explicit failure conditions. |
+| `regression` | Guard tests for behavior changes over time. | Known historical bugs and behavioral drift prevention. |
+| `determinism` | Repeatable results under fixed input and settings. | Compile determinism, stable ordering, and artifact reproducibility. |
+| `error-handling` | Exception surface and robustness expectations. | Recovery/failure modes and diagnostics quality. |
+
+## Class-level rules
+
+1. Every test class has **exactly one** scope tag.
+2. Every test class has at least one domain tag.
+3. Additional tags describe intent and may be used on classes or nested tests.
+4. For each test class, intent tags should reflect the primary behavior under test, not historical naming conventions.
+
+## Governance and execution policy
+
+The following rules are used to keep the suite auditable and stable:
+
+| Rule | Required state | Why |
+| --- | --- | --- |
+| Scope discipline | Exactly one scope tag per class. | Prevents accidental promotion of integration-only behavior into fast unit runs. |
+| Coverage breadth | At least one domain tag per class. | Ensures tests can be grouped by subsystem for targeted review. |
+| Intent specificity | Use at least one intent tag when behavior is non-trivial. | Makes failure triage faster and profile composition explicit. |
+| Runtime policy | Never run `slow` tests in the default `unit` profile unless explicitly required. | Preserves turnaround for PR feedback while preserving deep checks. |
+| Change risk | Any persistence or compatibility-affecting change must include `compat` in validation. | Protects long-lived binary artifact contracts. |
+| Mutation resistance | `fuzz`/`property` sets should be gated to dedicated profiles. | Limits flakiness exposure and controls CI resource cost. |
+
+## Suggested CI profiles
+
+These are recommended launch profiles for local and CI usage and are also exposed as Gradle tasks:
+
+- **Profile: `ci-smoke` (fast feedback):**
+
+```
+./gradlew test -DincludeTags=unit -DexcludeTags=slow
+./gradlew ciSmoke
+```
+
+`ciSmoke` also excludes `org.egothor.stemmer.CompileIntegrationTest*` at test-name filter level as a
+defensive fallback in case of future tag drift.
+`ciRelease` also excludes
+`org.egothor.stemmer.StemmerPatchTrieLoaderTest$BundledDictionaryTests*` at filter level.
+
+- **Profile: `ci-core` (core behavioral coverage):**
+
+```
+./gradlew test -DincludeTags=unit,trie,frequency-trie,property
+./gradlew ciCore
+```
+
+- **Profile: `ci-integration` (pipeline correctness):**
+
+```
+./gradlew test -DincludeTags=integration
+./gradlew ciIntegration
+```
+
+- **Profile: `ci-slow` (explicit heavy validation):**
+
+```
+./gradlew ciSlow
+```
+
+- **Profile: `ci-compat` (artifact stability):**
+
+```
+./gradlew test -DincludeTags=compat,regression
+./gradlew ciCompat
+```
+
+- **Profile: `ci-release` (strong confidence before release):**
+
+```
+./gradlew test -DexcludeTags=slow
+./gradlew ciRelease
+```
+`ciRelease` is non-slow by policy and uses the same defensive name-based exclusion for
+`org.egothor.stemmer.CompileIntegrationTest*` and
+`org.egothor.stemmer.StemmerPatchTrieLoaderTest$BundledDictionaryTests*` in addition to tag filtering.
+
+- **Profile: `ci-nightly` (extended hardening):**
+
+```
+./gradlew test -DincludeTags=fuzz
+./gradlew ciNightly
+```
+
+- **Profile: `ci` (enterprise umbrella):**
+
+```
+./gradlew ci
+```
+
+`ci` and `ciRelease` intentionally do **not** include `slow` paths. Run `ciSlow` explicitly for production-dictionary stress and long-running corpus checks.
+
+## Practical examples
+
+All examples use Gradle with JUnit Platform integration:
+
+- Default fast test run:
+
+```
+./gradlew test
+```
+
+The default `test` task excludes `slow` tests. Supplying `-DincludeTags` or `-PincludeTags` still excludes `slow` unless the include expression contains `slow`; supplying an explicit exclude expression replaces the default. Long-running bundled-dictionary compilation and full-language loading checks therefore run only through an explicit tag expression such as `-DincludeTags=slow` or a dedicated profile such as `ciSlow`.
+
+- Only unit tests:
+
+```
+./gradlew test -DincludeTags=unit
+```
+
+- Integration tests only:
+
+```
+./gradlew test -DincludeTags=integration -DexcludeTags=slow
+```
+
+- Only trie subsystem tests:
+
+```
+./gradlew test -DincludeTags=trie
+```
+
+- Deterministic fuzz checks:
+
+```
+./gradlew test -DincludeTags=fuzz
+```
+
+- Property tests:
+
+```
+./gradlew test -DincludeTags=property
+```
+
+- Stemmer + patch command behavior:
+
+```
+./gradlew test -DincludeTags=stemmer,patch
+```
+
+- Compatibility artifacts and regression checks:
+
+```
+./gradlew test -DincludeTags=compat
+```
+
+- Keep regression suite and remove long-running cases:
+
+```
+./gradlew test -DincludeTags=regression -DexcludeTags=slow
+```
+
+- Core + patch behavior:
+
+```
+./gradlew test -DincludeTags=trie,patch
+```
+
+- Deterministic compatibility and persistence checks:
+
+```
+./gradlew test -DincludeTags=compat,determinism,serialization
+```
+
+## Notes
+
+- `-DincludeTags` and `-DexcludeTags` are interpreted by Gradle task filtering and forwarded into
+  JUnit tag filtering.
+- Class-name filtering is also available via Gradle test selectors where needed
+  (for example, `--tests *CompileTest`), but tag filtering remains the default
+  execution strategy.
+- `-DincludeTags` supports comma-separated literal tags. When you need a single exact tag with special
+  characters, quote the argument for the shell.

gradle.lockfile

@@ -7,8 +7,13 @@ com.google.code.gson:gson:2.13.2=pmd
 com.google.errorprone:error_prone_annotations:2.41.0=pmd
 net.bytebuddy:byte-buddy-agent:1.17.7=jmhRuntimeClasspath,testCompileClasspath,testRuntimeClasspath
 net.bytebuddy:byte-buddy:1.17.7=jmhRuntimeClasspath,testCompileClasspath,testRuntimeClasspath
+net.jqwik:jqwik-api:1.9.3=jmhRuntimeClasspath,testCompileClasspath,testRuntimeClasspath
+net.jqwik:jqwik-engine:1.9.3=jmhRuntimeClasspath,testRuntimeClasspath
+net.jqwik:jqwik-time:1.9.3=jmhRuntimeClasspath,testCompileClasspath,testRuntimeClasspath
+net.jqwik:jqwik-web:1.9.3=jmhRuntimeClasspath,testCompileClasspath,testRuntimeClasspath
+net.jqwik:jqwik:1.9.3=jmhRuntimeClasspath,testCompileClasspath,testRuntimeClasspath
 net.sf.jopt-simple:jopt-simple:4.9=pitest
-net.sf.jopt-simple:jopt-simple:5.0.4=jmh,jmhCompileClasspath,jmhRuntimeClasspath
+net.sf.jopt-simple:jopt-simple:5.0.4=jmh,jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
 net.sf.saxon:Saxon-HE:12.9=pmd
 net.sourceforge.pmd:pmd-ant:7.20.0=pmd
 net.sourceforge.pmd:pmd-core:7.20.0=pmd
@@ -17,9 +22,17 @@ org.antlr:antlr4-runtime:4.9.3=pmd
 org.antlr:stringtemplate:3.2.1=pitest
 org.apache.commons:commons-lang3:3.18.0=pitest
 org.apache.commons:commons-lang3:3.20.0=pmd
-org.apache.commons:commons-math3:3.6.1=jmh,jmhCompileClasspath,jmhRuntimeClasspath
+org.apache.commons:commons-math3:3.6.1=jmh,jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
 org.apache.commons:commons-text:1.14.0=pitest
-org.apiguardian:apiguardian-api:1.1.2=testCompileClasspath
+org.apache.lucene:lucene-analysis-common:10.5.0=jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
+org.apache.lucene:lucene-analysis-morfologik:10.5.0=jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
+org.apache.lucene:lucene-analysis-stempel:10.5.0=jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
+org.apache.lucene:lucene-core:10.5.0=jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
+org.apache.opennlp:opennlp-tools:2.5.4=jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
+org.apiguardian:apiguardian-api:1.1.2=jmhRuntimeClasspath,testCompileClasspath,testRuntimeClasspath
+org.carrot2:morfologik-fsa:2.1.9=jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
+org.carrot2:morfologik-polish:2.1.9=jmhRuntimeClasspath,stemmingQualityJmhRuntime
+org.carrot2:morfologik-stemming:2.1.9=jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
 org.checkerframework:checker-qual:3.52.1=pmd
 org.jacoco:org.jacoco.agent:0.8.14=jacocoAgent,jacocoAnt
 org.jacoco:org.jacoco.ant:0.8.14=jacocoAnt
@@ -36,10 +49,10 @@ org.junit:junit-bom:5.14.3=jmhRuntimeClasspath,testCompileClasspath,testRuntimeC
 org.mockito:mockito-core:5.23.0=jmhRuntimeClasspath,mockitoAgent,testCompileClasspath,testRuntimeClasspath
 org.mockito:mockito-junit-jupiter:5.23.0=jmhRuntimeClasspath,testCompileClasspath,testRuntimeClasspath
 org.objenesis:objenesis:3.3=jmhRuntimeClasspath,testRuntimeClasspath
-org.openjdk.jmh:jmh-core:1.37=jmh,jmhCompileClasspath,jmhRuntimeClasspath
-org.openjdk.jmh:jmh-generator-asm:1.37=jmh,jmhCompileClasspath,jmhRuntimeClasspath
-org.openjdk.jmh:jmh-generator-bytecode:1.37=jmh,jmhCompileClasspath,jmhRuntimeClasspath
-org.openjdk.jmh:jmh-generator-reflection:1.37=jmh,jmhCompileClasspath,jmhRuntimeClasspath
+org.openjdk.jmh:jmh-core:1.37=jmh,jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
+org.openjdk.jmh:jmh-generator-asm:1.37=jmh,jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
+org.openjdk.jmh:jmh-generator-bytecode:1.37=jmh,jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
+org.openjdk.jmh:jmh-generator-reflection:1.37=jmh,jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
 org.opentest4j:opentest4j:1.3.0=jmhRuntimeClasspath,testCompileClasspath,testRuntimeClasspath
 org.ow2.asm:asm-analysis:9.9.1=pitest
 org.ow2.asm:asm-commons:9.9=jacocoAnt
@@ -47,7 +60,7 @@ org.ow2.asm:asm-commons:9.9.1=pitest
 org.ow2.asm:asm-tree:9.9=jacocoAnt
 org.ow2.asm:asm-tree:9.9.1=pitest
 org.ow2.asm:asm-util:9.9.1=pitest
-org.ow2.asm:asm:9.0=jmh,jmhCompileClasspath,jmhRuntimeClasspath
+org.ow2.asm:asm:9.0=jmh,jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
 org.ow2.asm:asm:9.9=jacocoAnt
 org.ow2.asm:asm:9.9.1=pitest,pmd
 org.pcollections:pcollections:4.0.2=pmd
@@ -57,5 +70,7 @@ org.pitest:pitest-html-report:1.22.1=pitest
 org.pitest:pitest-junit5-plugin:1.2.3=pitest
 org.pitest:pitest:1.22.1=pitest
 org.slf4j:jul-to-slf4j:1.7.36=pmd
+org.slf4j:slf4j-api:2.0.17=jmhCompileClasspath,jmhRuntimeClasspath,stemmingQualityJmhRuntime
 org.xmlresolver:xmlresolver:5.3.3=pmd
+ua.net.nlp:morfologik-ukrainian-search:4.9.1=jmhRuntimeClasspath,stemmingQualityJmhRuntime
 empty=annotationProcessor,compileClasspath,cyclonedxBom,jmhAnnotationProcessor,mainPmdAuxClasspath,runtimeClasspath,testAnnotationProcessor

gradle.properties

@@ -17,3 +17,6 @@ pomScmDeveloperConnection=scm:git:ssh://git@github.com/leogalambos/Radixor.git
 
 pomLicenseName=BSD-3-Clause
 pomLicenseUrl=https://spdx.org/licenses/BSD-3-Clause.html
+
+pomStemmerDataLicenseName=Stemmer Data License Policy
+pomStemmerDataLicenseUrl=https://github.com/leogalambos/Radixor/blob/main/LICENSE-stemmer-data

gradle/cistem-benchmarks.gradle

@@ -0,0 +1,69 @@
+def cistemGoldStandardBaseUrl = 'https://raw.githubusercontent.com/LeonieWeissweiler/CISTEM/refs/heads/master/gold_standards'
+def cistemGoldStandardFiles = [
+        'goldstandard1.txt',
+        'goldstandard2.txt'
+]
+def cistemGoldStandardDownloadDirectory = layout.buildDirectory.dir('third-party/cistem-gold-standards')
+def cistemGoldStandardGeneratedResourcesDirectory = layout.buildDirectory.dir('generated/resources/cistem-gold-standards')
+
+def cistemGoldStandardDownloadedFiles = cistemGoldStandardFiles.collect { String fileName ->
+    cistemGoldStandardDownloadDirectory.map { it.file(fileName) }
+}
+
+tasks.register('downloadCistemGoldStandards') {
+    group = 'build setup'
+    description = 'Downloads benchmark-only CISTEM German gold standards.'
+
+    outputs.files(cistemGoldStandardDownloadedFiles)
+
+    doLast {
+        cistemGoldStandardFiles.each { String fileName ->
+            final File targetFile = cistemGoldStandardDownloadDirectory.get().file(fileName).asFile
+            targetFile.parentFile.mkdirs()
+
+            if (!targetFile.exists()) {
+                final URL sourceUrl = new URL("${cistemGoldStandardBaseUrl}/${fileName}")
+                try {
+                    sourceUrl.withInputStream { inputStream ->
+                        targetFile.withOutputStream { outputStream ->
+                            outputStream << inputStream
+                        }
+                    }
+                } catch (FileNotFoundException exception) {
+                    throw new GradleException(
+                            "Unable to download CISTEM gold standard ${fileName} from ${sourceUrl}.",
+                            exception)
+                }
+            }
+
+            if (targetFile.length() <= 0L) {
+                throw new GradleException("Downloaded CISTEM gold standard ${fileName} was empty.")
+            }
+        }
+    }
+}
+
+tasks.register('prepareCistemGoldStandardResources', Copy) {
+    group = 'build setup'
+    description = 'Copies benchmark-only CISTEM German gold standards into the JMH resource output.'
+
+    dependsOn(tasks.named('downloadCistemGoldStandards'))
+
+    from(cistemGoldStandardDownloadDirectory) {
+        include 'goldstandard1.txt'
+        include 'goldstandard2.txt'
+    }
+    into(cistemGoldStandardGeneratedResourcesDirectory)
+}
+
+sourceSets {
+    jmh {
+        resources {
+            srcDir(cistemGoldStandardGeneratedResourcesDirectory)
+        }
+    }
+}
+
+tasks.named('processJmhResources') {
+    dependsOn(tasks.named('prepareCistemGoldStandardResources'))
+}

gradle/hunspell-benchmarks.gradle

@@ -0,0 +1,121 @@
+import org.gradle.plugins.ide.eclipse.model.SourceFolder
+
+
+def hunspellDictionaryBaseUrl = 'https://raw.githubusercontent.com/wooorm/dictionaries/main/dictionaries'
+def hunspellDictionaryLanguages = [
+        en: 'English',
+        cs: 'Czech',
+        de: 'German',
+        es: 'Spanish',
+        fr: 'French',
+        nl: 'Dutch',
+        pl: 'Polish',
+        uk: 'Ukrainian'
+]
+def hunspellDownloadDirectory = layout.buildDirectory.dir('third-party/hunspell')
+def hunspellGeneratedResourcesDirectory = layout.buildDirectory.dir('generated/resources/hunspell')
+def hunspellGeneratedResourcesPath = provider {
+    project.relativePath(hunspellGeneratedResourcesDirectory.get().asFile)
+}
+def hunspellEclipseClasspathAttributes = [
+        gradle_scope        : 'jmh',
+        gradle_used_by_scope: 'jmh',
+        test                : 'true'
+]
+def hunspellIsAbsolutePath = { String path ->
+    path.startsWith('/') || path ==~ /^[A-Za-z]:[\\\/].*/
+}
+
+def hunspellDownloadedFiles = hunspellDictionaryLanguages.keySet().collectMany { String code ->
+    [
+            hunspellDownloadDirectory.map { it.file("${code}/index.aff") },
+            hunspellDownloadDirectory.map { it.file("${code}/index.dic") },
+            hunspellDownloadDirectory.map { it.file("${code}/license") }
+    ]
+}
+
+tasks.register('downloadHunspellBenchmarkDictionaries') {
+    group = 'build setup'
+    description = 'Downloads benchmark-only Hunspell dictionaries from wooorm/dictionaries.'
+
+    outputs.files(hunspellDownloadedFiles)
+
+    doLast {
+        hunspellDictionaryLanguages.each { String code, String displayName ->
+            ['index.aff', 'index.dic', 'license'].each { String fileName ->
+                final File targetFile = hunspellDownloadDirectory.get().file("${code}/${fileName}").asFile
+                targetFile.parentFile.mkdirs()
+
+                if (!targetFile.exists()) {
+                    final URL sourceUrl = new URL("${hunspellDictionaryBaseUrl}/${code}/${fileName}")
+                    try {
+                        sourceUrl.withInputStream { inputStream ->
+                            targetFile.withOutputStream { outputStream ->
+                                outputStream << inputStream
+                            }
+                        }
+                    } catch (FileNotFoundException exception) {
+                        throw new GradleException(
+                                "Unable to download Hunspell ${fileName} file for ${displayName} (${code}) from ${sourceUrl}.",
+                                exception)
+                    }
+                }
+
+                if (targetFile.length() <= 0L) {
+                    throw new GradleException("Downloaded Hunspell ${fileName} file for ${displayName} was empty.")
+                }
+            }
+        }
+    }
+}
+
+tasks.register('prepareHunspellBenchmarkResources', Copy) {
+    group = 'build setup'
+    description = 'Copies benchmark-only Hunspell dictionaries into the JMH resource output.'
+
+    dependsOn(tasks.named('downloadHunspellBenchmarkDictionaries'))
+
+    from(hunspellDownloadDirectory) {
+        include '**/index.aff'
+        include '**/index.dic'
+        include '**/license'
+        into 'hunspell'
+    }
+    into(hunspellGeneratedResourcesDirectory)
+}
+
+sourceSets {
+    jmh {
+        resources {
+            srcDir(hunspellGeneratedResourcesDirectory)
+        }
+    }
+}
+
+tasks.named('processJmhResources') {
+    dependsOn(tasks.named('prepareHunspellBenchmarkResources'))
+}
+
+eclipse {
+    classpath {
+        file {
+            whenMerged { classpath ->
+                String generatedPath = hunspellGeneratedResourcesPath.get()
+
+                classpath.entries.removeAll { entry ->
+                    entry.hasProperty('path') && (
+                            entry.path == generatedPath ||
+                            hunspellIsAbsolutePath(entry.path)
+                    )
+                }
+
+                SourceFolder hunspellEntry = new SourceFolder(generatedPath, null)
+                hunspellEntry.output = 'bin/jmh'
+                hunspellEclipseClasspathAttributes.each { String name, String value ->
+                    hunspellEntry.entryAttributes[name] = value
+                }
+                classpath.entries.add(hunspellEntry)
+            }
+        }
+    }
+}

gradle/libs.versions.toml

@@ -1,12 +1,14 @@
 #
 # After changing dependency versions:
 #
+# unlock temporarily: LockMode.STRICT -> LockMode.LENIENT
+#
+# refresh verification metadata:
+#     ./gradlew --write-verification-metadata sha256 test jmh distZip cyclonedxBom
+#
 # run:
 #     ./gradlew --write-locks classes testClasses jmh distZip cyclonedxBom
 #
-# if needed, refresh verification metadata:
-#     ./gradlew --write-verification-metadata sha256 test jmh distZip cyclonedxBom
-#
 # (optional - for Eclipse IDE)
 # insert trusted-artifacts into gradle/verification-metadata.xml/verification-metadata/configuration:
 #     <trusted-artifacts>
@@ -21,6 +23,7 @@
 [versions]
 junit = "5.14.3"
 mockito = "5.23.0"
+jqwik = "1.9.3"
 
 [libraries]
 junit-bom = { module = "org.junit:junit-bom", version.ref = "junit" }
@@ -29,3 +32,5 @@ junit-platform-launcher = { module = "org.junit.platform:junit-platform-launcher
 
 mockito-core = { module = "org.mockito:mockito-core", version.ref = "mockito" }
 mockito-junit-jupiter = { module = "org.mockito:mockito-junit-jupiter", version.ref = "mockito" }
+
+jqwik = { module = "net.jqwik:jqwik", version.ref = "jqwik" }

gradle/lucene-benchmarks.gradle

@@ -0,0 +1,223 @@
+import org.gradle.plugins.ide.eclipse.model.SourceFolder
+
+
+def luceneVersion = '10.5.0'
+def luceneRootRelativePath = 'third-party/lucene'
+def luceneSourceArtifacts = ['lucene-analysis-common', 'lucene-analyzers-common']
+def luceneSourceDirectory = layout.buildDirectory.dir("${luceneRootRelativePath}/source/analyzers-common")
+
+def luceneGeneratedSourceDirectory = layout.buildDirectory.dir('generated/sources/lucene')
+def luceneGeneratedPorterFile = luceneGeneratedSourceDirectory.map { it.file('org/egothor/stemmer/benchmark/LucenePorterStemmerCopied.java') }
+
+def luceneSourceDownloadFile = layout.buildDirectory.file("${luceneRootRelativePath}/lucene-${luceneVersion}-sources.jar")
+
+dependencies {
+    jmhImplementation "org.apache.lucene:lucene-analysis-common:${luceneVersion}"
+    jmhImplementation "org.apache.lucene:lucene-analysis-stempel:${luceneVersion}"
+    jmhImplementation "org.apache.lucene:lucene-analysis-morfologik:${luceneVersion}"
+}
+
+def buildLuceneSourcesName = { final String artifact ->
+    "${artifact}-${luceneVersion}-sources.jar"
+}
+def buildLuceneSourcesUrl = { final String artifact ->
+    "https://repo1.maven.org/maven2/org/apache/lucene/${artifact}/${luceneVersion}/${buildLuceneSourcesName(artifact)}"
+}
+
+def isLuceneSourcesDownloadable = { final String artifact ->
+    try {
+        final URL sourceUrl = new URL(buildLuceneSourcesUrl(artifact))
+        final java.net.HttpURLConnection connection = (java.net.HttpURLConnection) sourceUrl.openConnection()
+        connection.requestMethod = 'HEAD'
+        connection.instanceFollowRedirects = true
+        connection.connectTimeout = 10000
+        connection.readTimeout = 10000
+        final int responseCode = connection.responseCode
+        connection.disconnect()
+        return responseCode == 200
+    } catch (Exception ignored) {
+        return false
+    }
+}
+
+def downloadLuceneSourcesJar = { ->
+    final File targetFile = luceneSourceDownloadFile.get().asFile
+
+    for (String artifact : luceneSourceArtifacts) {
+        if (!isLuceneSourcesDownloadable(artifact)) {
+            continue
+        }
+
+        final String sourceUrl = buildLuceneSourcesUrl(artifact)
+        final File tempFile = new File(targetFile.parentFile, "${artifact}.${luceneVersion}.tmp")
+        try {
+            new URL(sourceUrl).withInputStream { inputStream ->
+                tempFile.parentFile.mkdirs()
+                tempFile.withOutputStream { outputStream ->
+                    outputStream << inputStream
+                }
+            }
+            if (!tempFile.exists() || tempFile.length() <= 0L) {
+                throw new GradleException("Downloaded Lucene source artifact for ${artifact} was empty.")
+            }
+            targetFile.delete()
+            if (!tempFile.renameTo(targetFile)) {
+                throw new GradleException("Failed to persist downloaded Lucene source artifact for ${artifact}.")
+            }
+            return
+        } catch (Exception ignored) {
+            tempFile.delete()
+        }
+    }
+
+    throw new GradleException(
+            "Failed to download Apache Lucene source artifacts ${luceneSourceArtifacts} for version ${luceneVersion}.")
+}
+
+def luceneSourceClasspathPath = provider {
+    project.relativePath(luceneGeneratedSourceDirectory.get().asFile)
+}
+def luceneEclipseClasspathAttributes = [
+        gradle_scope        : 'jmh',
+        gradle_used_by_scope: 'jmh',
+        test                : 'true'
+]
+
+def isAbsoluteClasspathPath = { String path ->
+    path.startsWith('/') || path ==~ /^[A-Za-z]:[\\\/].*/
+}
+
+def luceneGeneratedPorterNotice = '''
+/**
+ * Generated at benchmark execution time from Apache Lucene source.
+ *
+ * This source copy is compiled only for the JMH benchmark source set and is
+ * not committed as production code.
+ */
+'''
+
+def transformPorterStemmerSource = { final File sourceFile, final File targetFile ->
+    if (!sourceFile.exists()) {
+        throw new GradleException("Apache Lucene PorterStemmer source was not available at ${sourceFile}.")
+    }
+
+    final String sourceText = sourceFile.getText('UTF-8')
+    String transformedText = sourceText
+
+    if (transformedText.contains('package org.apache.lucene.analysis.en;')) {
+        transformedText = transformedText.replaceFirst(/(?m)^\s*package\s+org\.apache\.lucene\.analysis\.en\s*;/,
+                'package org.egothor.stemmer.benchmark;')
+    } else {
+        throw new GradleException(
+                'Expected Lucene package-private PorterStemmer in org.apache.lucene.analysis.en package was not found in downloaded source.')
+    }
+
+    transformedText = transformedText.replaceFirst(/(?m)^\s*class\s+PorterStemmer\s*\{/, 'public final class LucenePorterStemmerCopied {')
+    transformedText = transformedText.replaceFirst(/(?m)^\s*public\s+PorterStemmer\(\)/, 'public LucenePorterStemmerCopied()')
+
+    if (!transformedText.contains('class LucenePorterStemmerCopied')) {
+        throw new GradleException("Failed to rename PorterStemmer class when generating ${targetFile}.")
+    }
+
+    targetFile.parentFile.mkdirs()
+    targetFile.text = transformedText
+}
+
+def resolveLucenePorterStemmerSource = { ->
+    final File sourceRoot = luceneSourceDirectory.get().asFile
+    final List<String> candidates = [
+            'org/apache/lucene/analysis/en/org/apache/lucene/analysis/en/PorterStemmer.java',
+            'org/apache/lucene/analysis/en/PorterStemmer.java',
+            'org/apache/lucene/analysis/en/org/tartarus/snowball/ext/PorterStemmer.java'
+    ]
+
+    for (String candidate : candidates) {
+        final File file = new File(sourceRoot, candidate)
+        if (file.exists()) {
+            return file
+        }
+    }
+
+    final FileTree porterCandidates = fileTree(sourceRoot).matching { include '**/PorterStemmer.java' }
+    for (File file : porterCandidates.files) {
+        if (file.text.contains('class PorterStemmer') && file.text.contains('package org.apache.lucene.analysis.en;')) {
+            return file
+        }
+    }
+
+    throw new GradleException('Unable to resolve Lucene PorterStemmer source file from extracted artifact.')
+}
+
+tasks.register('downloadLuceneAnalyzersSources') {
+    group = 'build setup'
+    description = 'Downloads Apache Lucene analysis sources for benchmark-only code generation.'
+
+    outputs.file(luceneSourceDownloadFile)
+
+    doLast {
+        if (!luceneSourceDownloadFile.get().asFile.exists()) {
+            downloadLuceneSourcesJar()
+        }
+    }
+}
+
+tasks.register('extractLuceneAnalyzersSources', Copy) {
+    group = 'build setup'
+    description = 'Extracts Apache Lucene analysis source JAR for benchmark-only extraction.'
+
+    dependsOn(tasks.named('downloadLuceneAnalyzersSources'))
+
+    from(zipTree(luceneSourceDownloadFile))
+    into(luceneSourceDirectory)
+}
+
+tasks.register('generateLucenePorterStemmerCopied') {
+    group = 'build setup'
+    description = 'Generates LucenePorterStemmerCopied into the build-only benchmark source directory.'
+
+    dependsOn(tasks.named('extractLuceneAnalyzersSources'))
+
+    inputs.dir(luceneSourceDirectory)
+    outputs.file(luceneGeneratedPorterFile)
+
+    doLast {
+        final File sourceFile = resolveLucenePorterStemmerSource()
+        transformPorterStemmerSource(sourceFile, luceneGeneratedPorterFile.get().asFile)
+    }
+}
+
+sourceSets {
+    jmh {
+        java {
+            srcDir(luceneGeneratedSourceDirectory)
+        }
+    }
+}
+
+tasks.named('compileJmhJava') {
+    dependsOn(tasks.named('generateLucenePorterStemmerCopied'))
+}
+
+eclipse {
+    classpath {
+        file {
+            whenMerged { classpath ->
+                String generatedPath = luceneSourceClasspathPath.get()
+
+                classpath.entries.removeAll { entry ->
+                    entry.hasProperty('path') && (
+                            entry.path == generatedPath ||
+                            isAbsoluteClasspathPath(entry.path)
+                    )
+                }
+
+                SourceFolder luceneEntry = new SourceFolder(generatedPath, null)
+                luceneEntry.output = 'bin/jmh'
+                luceneEclipseClasspathAttributes.each { String name, String value ->
+                    luceneEntry.entryAttributes[name] = value
+                }
+                classpath.entries.add(luceneEntry)
+            }
+        }
+    }
+}

gradle/maven-pom.gradle

@@ -13,6 +13,12 @@ def pomScmDeveloperConnection = providers.gradleProperty('pomScmDeveloperConnect
 def pomLicenseName = providers.gradleProperty('pomLicenseName').orNull
 def pomLicenseUrl = providers.gradleProperty('pomLicenseUrl').orNull
 def pomLicenseDistribution = providers.gradleProperty('pomLicenseDistribution').orElse('repo').get()
+def pomStemmerDataLicenseName = providers.gradleProperty('pomStemmerDataLicenseName')
+        .orElse('Stemmer Data License Policy')
+        .get()
+def pomStemmerDataLicenseUrl = providers.gradleProperty('pomStemmerDataLicenseUrl')
+        .orElse('https://github.com/leogalambos/Radixor/blob/main/LICENSE-stemmer-data')
+        .get()
 def pomDeveloperId = providers.gradleProperty('pomDeveloperId').orElse('egothor').get()
 def pomDeveloperName = providers.gradleProperty('pomDeveloperName').orElse('Leo Galambos').get()
 def pomDeveloperEmail = providers.gradleProperty('pomDeveloperEmail').orElse('egothor@gmail.com').get()
@@ -45,6 +51,11 @@ publishing {
                         url = pomLicenseUrl
                         distribution = pomLicenseDistribution
                     }
+                    license {
+                        name = pomStemmerDataLicenseName
+                        url = pomStemmerDataLicenseUrl
+                        distribution = pomLicenseDistribution
+                    }
                 }
 
                 developers {
@@ -73,7 +84,7 @@ publishing {
 }
 
 signing {
-    required { !version.toString().endsWith('-SNAPSHOT') }
+    required = !version.toString().endsWith('-SNAPSHOT')
     if (signingKey != null && !signingKey.isBlank()) {
         useInMemoryPgpKeys(signingKey, signingPassword)
         sign publishing.publications.mavenJava
@@ -93,6 +104,8 @@ tasks.register('validateReleaseMetadata') {
         if (pomScmDeveloperConnection == null || pomScmDeveloperConnection.isBlank()) missing.add('pomScmDeveloperConnection')
         if (pomLicenseName == null || pomLicenseName.isBlank()) missing.add('pomLicenseName')
         if (pomLicenseUrl == null || pomLicenseUrl.isBlank()) missing.add('pomLicenseUrl')
+        if (pomStemmerDataLicenseName == null || pomStemmerDataLicenseName.isBlank()) missing.add('pomStemmerDataLicenseName')
+        if (pomStemmerDataLicenseUrl == null || pomStemmerDataLicenseUrl.isBlank()) missing.add('pomStemmerDataLicenseUrl')
         if (signingKey == null || signingKey.isBlank()) missing.add('pomSigningKey / SIGNING_KEY')
         if (signingPassword == null || signingPassword.isBlank()) missing.add('pomSigningPassword / SIGNING_PASSWORD')
 
@@ -131,7 +144,10 @@ tasks.register('centralBundle', Zip) {
 
     dependsOn(tasks.named('createCentralChecksums'))
 
-    from(centralStagingRepositoryDirectory)
+    from(centralStagingRepositoryDirectory) {
+        exclude '**/maven-metadata*.xml*'
+    }
+    
     destinationDirectory = centralBundleDirectory
     archiveFileName = "radixor-${project.version}-central-bundle.zip"
 }

gradle/opennlp-benchmarks.gradle

@@ -0,0 +1,5 @@
+def openNlpVersion = '2.5.4'
+
+dependencies {
+    jmhImplementation "org.apache.opennlp:opennlp-tools:${openNlpVersion}"
+}

gradle/paicehusk-benchmarks.gradle

@@ -0,0 +1,219 @@
+import org.gradle.plugins.ide.eclipse.model.SourceFolder
+
+
+def paicehuskVersion = 'master'
+def paicehuskArchiveName = "paice-husk-stemmer-${paicehuskVersion}.zip"
+def paicehuskDownloadUrl = "https://github.com/Hopper262/paice-husk-stemmer/archive/refs/heads/${paicehuskVersion}.zip"
+def paicehuskDownloadFile = layout.buildDirectory.file("third-party/paicehusk/${paicehuskArchiveName}")
+def paicehuskExtractDirectory = layout.buildDirectory.dir('third-party/paicehusk/source')
+def paicehuskArchiveDirectory = paicehuskExtractDirectory.map { it.dir('paice-husk-stemmer-master') }
+def paicehuskJavaFile = paicehuskArchiveDirectory.map { it.file('paicehusk_java.java') }
+def paicehuskRulesFile = paicehuskArchiveDirectory.map { it.file('paicehusk_rules.txt') }
+
+def paicehuskGeneratedSourceDirectory = layout.buildDirectory.dir('generated/sources/paicehusk')
+def paicehuskGeneratedStemmerFile = paicehuskGeneratedSourceDirectory.map { it.file('org/egothor/stemmer/benchmark/PaiceHuskLancasterStemmer.java') }
+
+def paicehuskGeneratedSourcePath = provider {
+    project.relativePath(paicehuskGeneratedSourceDirectory.get().asFile)
+}
+def paicehuskSourceEclipseClasspathAttributes = [
+        gradle_scope        : 'jmh',
+        gradle_used_by_scope: 'jmh',
+        test                : 'true'
+]
+
+def paicehuskIsAbsolutePath = { String path ->
+    path.startsWith('/') || path ==~ /^[A-Za-z]:[\\\/].*/
+}
+
+def paicehuskGeneratedNotice = '''
+/**
+ * Generated at benchmark execution time from upstream
+ * https://github.com/Hopper262/paice-husk-stemmer .
+ *
+ * This source copy is compiled only for the JMH benchmark source set and is
+ * not committed as production code.
+ */
+'''
+
+def escapeForJava = { final String text ->
+    return text.replace('\\\\', '\\\\\\\\')
+            .replace('\"', '\\\"')
+}
+
+def toRuleLines = { final File rulesFile ->
+    final List<String> lines = rulesFile.readLines('UTF-8')
+    final StringBuilder ruleLines = new StringBuilder()
+    for (int index = 0; index < lines.size(); index++) {
+        final String line = lines.get(index)
+        ruleLines.append('            "')
+        ruleLines.append(escapeForJava(line))
+        ruleLines.append('"')
+        if (index < lines.size() - 1) {
+            ruleLines.append(',')
+        }
+        ruleLines.append('\n')
+    }
+    return ruleLines.toString()
+}
+
+def paicehuskEngineInsertion = { final String ruleLines ->
+    return """
+
+    public static final String[] RULE_LINES = {
+${ruleLines}
+    };
+
+    private static final java.util.HashMap RULES = createRulesFromEmbeddedRules();
+
+    /**
+     * Creates benchmark stemmer instance.
+     */
+    public PaiceHuskLancasterStemmer() {
+    }
+
+    /**
+     * Applies Paice/Husk stemming to one token.
+     *
+     * @param token input token
+     * @return stemmed token
+     */
+    public String stem(final String token) {
+        if (token == null) {
+            return null;
+        }
+        return stemWord(token, RULES, null);
+    }
+
+    /**
+     * Loads bundled rule lines directly from the generated benchmark source.
+     *
+     * @return initialized rule map
+     */
+    private static java.util.HashMap createRulesFromEmbeddedRules() {
+        try {
+            final java.io.File ruleFile = java.io.File.createTempFile("paicehusk-rules", ".txt");
+            ruleFile.deleteOnExit();
+            try (java.io.PrintWriter writer = new java.io.PrintWriter(new java.io.FileWriter(ruleFile))) {
+                for (String line : RULE_LINES) {
+                    writer.println(line);
+                }
+            }
+            return loadRules(ruleFile.getAbsolutePath());
+        } catch (Exception exception) {
+            throw new IllegalStateException("Unable to initialize benchmark Paice/Husk rules.", exception);
+        }
+    }
+"""
+}
+
+def transformPaiceHuskSource = { final File sourceFile, final File rulesFile, final File targetFile ->
+    if (!sourceFile.exists()) {
+        throw new GradleException("Paice/Husk Java source was not available at ${sourceFile}.")
+    }
+    if (!rulesFile.exists()) {
+        throw new GradleException("Paice/Husk rule file was not available at ${rulesFile}.")
+    }
+
+    final String sourceText = sourceFile.getText('UTF-8')
+    String transformedText = sourceText
+
+    transformedText = 'package org.egothor.stemmer.benchmark;' + '\n\n' + transformedText
+    transformedText = transformedText.replaceFirst(/(?m)^\s*class\s+PaiceHusk\s*\{/, 'public final class PaiceHuskLancasterStemmer {')
+
+    final int packageEnd = transformedText.indexOf('\n', transformedText.indexOf('package org.egothor.stemmer.benchmark;'))
+    if (packageEnd >= 0) {
+        transformedText = transformedText.substring(0, packageEnd + 1) + '\n' + paicehuskGeneratedNotice + transformedText.substring(packageEnd + 1)
+    }
+
+    final String marker = '\n} // end class PaiceHusk'
+    final int markerIndex = transformedText.lastIndexOf(marker)
+    if (markerIndex < 0) {
+        throw new GradleException("Unexpected Paice/Husk source structure at ${sourceFile}.")
+    }
+    final String replacement = paicehuskEngineInsertion(toRuleLines(rulesFile))
+    transformedText = transformedText.substring(0, markerIndex) + '\n' + replacement + '\n}' 
+
+    targetFile.parentFile.mkdirs()
+    targetFile.text = transformedText
+}
+
+tasks.register('downloadPaiceHuskStemmer') {
+    group = 'build setup'
+    description = 'Downloads the upstream Paice/Husk benchmark source for dynamic extraction.'
+
+    outputs.file(paicehuskDownloadFile)
+
+    doLast {
+        final File targetFile = paicehuskDownloadFile.get().asFile
+        targetFile.parentFile.mkdirs()
+
+        if (!targetFile.exists()) {
+            new URL(paicehuskDownloadUrl).withInputStream { inputStream ->
+                targetFile.withOutputStream { outputStream ->
+                    outputStream << inputStream
+                }
+            }
+        }
+    }
+}
+
+tasks.register('extractPaiceHuskStemmer', Copy) {
+    group = 'build setup'
+    description = 'Extracts the upstream Paice/Husk benchmark archive.'
+
+    dependsOn(tasks.named('downloadPaiceHuskStemmer'))
+
+    from(zipTree(paicehuskDownloadFile))
+    into(paicehuskExtractDirectory)
+}
+
+tasks.register('generatePaiceHuskLancasterStemmer') {
+    group = 'build setup'
+    description = 'Generates PaiceHuskLancasterStemmer into a benchmark-only generated source directory.'
+
+    dependsOn(tasks.named('extractPaiceHuskStemmer'))
+
+    inputs.files(paicehuskJavaFile, paicehuskRulesFile)
+    outputs.file(paicehuskGeneratedStemmerFile)
+
+    doLast {
+        transformPaiceHuskSource(paicehuskJavaFile.get().asFile, paicehuskRulesFile.get().asFile, paicehuskGeneratedStemmerFile.get().asFile)
+    }
+}
+
+sourceSets {
+    jmh {
+        java {
+            srcDir(paicehuskGeneratedSourceDirectory)
+        }
+    }
+}
+
+tasks.named('compileJmhJava') {
+    dependsOn(tasks.named('generatePaiceHuskLancasterStemmer'))
+}
+
+eclipse {
+    classpath {
+        file {
+            whenMerged { classpath ->
+                String generatedPath = paicehuskGeneratedSourcePath.get()
+
+                classpath.entries.removeAll { entry ->
+                    entry.hasProperty('path') && (
+                            entry.path == generatedPath ||
+                            paicehuskIsAbsolutePath(entry.path)
+                    )
+                }
+
+                SourceFolder paicehuskEntry = new SourceFolder(generatedPath, null)
+                paicehuskEntry.output = 'bin/jmh'
+                paicehuskSourceEclipseClasspathAttributes.each { String name, String value ->
+                    paicehuskEntry.entryAttributes[name] = value
+                }
+                classpath.entries.add(paicehuskEntry)
+            }
+        }
+    }
+}

gradle/snowball-benchmarks.gradle

@@ -1,10 +1,57 @@
+import org.gradle.plugins.ide.eclipse.model.SourceFolder
+
+
 def snowballVersion = '3.0.1'
 def snowballArchiveName = "libstemmer_java-${snowballVersion}.tar.gz"
+def snowballDistributionDirectoryName = "libstemmer_java-${snowballVersion}"
+def snowballRootRelativePath = 'third-party/snowball'
+def snowballSourceRelativePath = "${snowballRootRelativePath}/source"
+def snowballJavaSourceRelativePath = "${snowballSourceRelativePath}/${snowballDistributionDirectoryName}/java"
+def snowballGeneratedSourceRelativePath = 'generated/sources/snowball'
 def snowballDownloadUrl = "https://snowballstem.org/dist/${snowballArchiveName}"
-def snowballDownloadFile = layout.buildDirectory.file("third-party/snowball/${snowballArchiveName}")
-def snowballExtractDirectory = layout.buildDirectory.dir('third-party/snowball/source')
-def snowballJavaSourceDirectory = layout.buildDirectory.dir(
-        "third-party/snowball/source/libstemmer_java-${snowballVersion}/java")
+def snowballDownloadFile = layout.buildDirectory.file("${snowballRootRelativePath}/${snowballArchiveName}")
+def snowballExtractDirectory = layout.buildDirectory.dir(snowballSourceRelativePath)
+def snowballJavaSourceDirectory = layout.buildDirectory.dir(snowballJavaSourceRelativePath)
+def snowballGeneratedSourceDirectory = layout.buildDirectory.dir(snowballGeneratedSourceRelativePath)
+def snowballJavaSourceClasspathPath = provider {
+    project.relativePath(snowballJavaSourceDirectory.get().asFile)
+}
+def snowballGeneratedSourceClasspathPath = provider {
+    project.relativePath(snowballGeneratedSourceDirectory.get().asFile)
+}
+def transformSnowballSourceText = { final String sourceText ->
+    String transformedText = sourceText
+
+    transformedText = transformedText.replaceAll(/(?m)^\s*package\s+org\.tartarus\.snowball\.ext\s*;/,
+            'package org.egothor.stemmer.benchmark.snowball.ext;')
+    transformedText = transformedText.replaceAll(/(?m)^\s*package\s+org\.tartarus\.snowball\s*;/,
+            'package org.egothor.stemmer.benchmark.snowball;')
+    transformedText = transformedText.replace('org.tartarus.snowball.', 'org.egothor.stemmer.benchmark.snowball.')
+
+    return transformedText
+}
+def copySnowballSourcesWithPackageIsolation = { final File sourceDirectory, final File targetDirectory ->
+    final FileTree sourceFiles = fileTree(sourceDirectory).matching { include '**/*.java' }
+    if (targetDirectory.exists()) {
+        targetDirectory.deleteDir()
+    }
+
+    for (File sourceFile : sourceFiles.files) {
+        final String relativePath = sourceDirectory.toPath().relativize(sourceFile.toPath()).toString()
+        final File outputFile = new File(targetDirectory, relativePath)
+        outputFile.parentFile.mkdirs()
+        outputFile.text = transformSnowballSourceText(sourceFile.getText('UTF-8'))
+    }
+}
+
+def snowballEclipseClasspathAttributes = [
+        gradle_scope        : 'jmh',
+        gradle_used_by_scope: 'jmh',
+        test                : 'true'
+]
+def isAbsoluteClasspathPath = { String path ->
+    path.startsWith('/') || path ==~ /^[A-Za-z]:[\\\/].*/
+}
 
 tasks.register('downloadSnowballJava') {
     group = 'build setup'
@@ -36,14 +83,60 @@ tasks.register('extractSnowballJava', Copy) {
     into(snowballExtractDirectory)
 }
 
+tasks.register('generateIsolatedSnowballSources') {
+    group = 'build setup'
+    description = 'Copies Snowball source to benchmark-only package-isolated package paths.'
+
+    dependsOn(tasks.named('extractSnowballJava'))
+
+    inputs.dir(snowballJavaSourceDirectory)
+    outputs.dir(snowballGeneratedSourceDirectory)
+
+    doLast {
+        copySnowballSourcesWithPackageIsolation(
+                snowballJavaSourceDirectory.get().asFile,
+                snowballGeneratedSourceDirectory.get().asFile
+        )
+    }
+}
+
 sourceSets {
     jmh {
         java {
-            srcDir(snowballJavaSourceDirectory)
+            srcDir(snowballGeneratedSourceDirectory)
         }
     }
 }
 
 tasks.named('compileJmhJava') {
-    dependsOn(tasks.named('extractSnowballJava'))
+    dependsOn(tasks.named('generateIsolatedSnowballSources'))
+}
+
+eclipse {
+    classpath {
+        file {
+            whenMerged { classpath ->
+                String generatedSnowballPath = snowballJavaSourceClasspathPath.get()
+                String generatedIsolatedSnowballPath = snowballGeneratedSourceClasspathPath.get()
+                String modelSnowballPath = snowballJavaSourceRelativePath
+
+                classpath.entries.removeAll { entry ->
+                    entry.hasProperty('path') && (
+                            entry.path == generatedSnowballPath ||
+                            entry.path == generatedIsolatedSnowballPath ||
+                            entry.path == modelSnowballPath ||
+                            isAbsoluteClasspathPath(entry.path)
+                    )
+                }
+
+                SourceFolder snowballEntry = new SourceFolder(generatedIsolatedSnowballPath, null)
+                snowballEntry.output = 'bin/jmh'
+                snowballEclipseClasspathAttributes.each { String name, String value ->
+                    snowballEntry.entryAttributes[name] = value
+                }
+
+                classpath.entries.add(snowballEntry)
+            }
+        }
+    }
 }

gradle/verification-metadata.xml

@@ -568,6 +568,46 @@
             <sha256 value="1af699f8d9ddab67f9a0d202fbd7915eb0362a5a6dfd5ffc54cafa3465c9cb0a" origin="Generated by Gradle"/>
          </artifact>
       </component>
+      <component group="net.jqwik" name="jqwik" version="1.9.3">
+         <artifact name="jqwik-1.9.3.jar">
+            <sha256 value="562931e1667308180056a8ce85791f71ab8c37ca8efc2006a163ba5d650e5f73" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="jqwik-1.9.3.module">
+            <sha256 value="681316f856db4ea3cac8fcced811127fc1d7016875e5b50aa4a55024513a93d7" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
+      <component group="net.jqwik" name="jqwik-api" version="1.9.3">
+         <artifact name="jqwik-api-1.9.3.jar">
+            <sha256 value="4bce7e80beb6d9d7092a799fa8a509d76cc31dbb20c938a9952965c15d1dd9b2" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="jqwik-api-1.9.3.module">
+            <sha256 value="69984416ea2e9f7fde40cfac983d2f540d3a37e9766fd3b0a06fada8f9b4cff2" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
+      <component group="net.jqwik" name="jqwik-engine" version="1.9.3">
+         <artifact name="jqwik-engine-1.9.3.jar">
+            <sha256 value="b85592ee78e30239ccfdca7a134f918ee94ebec51ad29a313fc9a676d97b3ede" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="jqwik-engine-1.9.3.module">
+            <sha256 value="2c68479ebda9e334bc9033abd2ef227353808f20114f197947b5c7b9646ab8e5" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
+      <component group="net.jqwik" name="jqwik-time" version="1.9.3">
+         <artifact name="jqwik-time-1.9.3.jar">
+            <sha256 value="9fd09021d8f03d44990457bf3095cf0aaf34d2785d1108ff22590286c233b3e5" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="jqwik-time-1.9.3.module">
+            <sha256 value="c2b056576c8767bfcd7efdd982890fbc71e608fb5c9c80fc145cfee6adeeaa24" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
+      <component group="net.jqwik" name="jqwik-web" version="1.9.3">
+         <artifact name="jqwik-web-1.9.3.jar">
+            <sha256 value="6aee9d583c1ff9efe319b2fa0bc9d75fc616de6d1f240ddbd2af9eabda483dbe" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="jqwik-web-1.9.3.module">
+            <sha256 value="38c86130c8b86c1657b4f8256e065ee08551f7c5ce728d1a5be8f63133b14554" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
       <component group="net.sf.jopt-simple" name="jopt-simple" version="4.9">
          <artifact name="jopt-simple-4.9.jar">
             <sha256 value="26c5856e954b5f864db76f13b86919b59c6eecf9fd930b96baa8884626baf2f5" origin="Generated by Gradle"/>
@@ -670,6 +710,11 @@
             <sha256 value="d78bd8524c5f8380a190a6525686629a95dfe512df21111383a6d8c0923a4415" origin="Generated by Gradle"/>
          </artifact>
       </component>
+      <component group="org.apache" name="apache" version="34">
+         <artifact name="apache-34.pom">
+            <sha256 value="3671ae9d4d062ae3bb985731c76088bb2f6f7d7254e2d304ee9f690b97651328" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
       <component group="org.apache" name="apache" version="35">
          <artifact name="apache-35.pom">
             <sha256 value="ea297dcd114136e8b8e8b630230d52a76c2fc69f6c5db25d672b1857000728b8" origin="Generated by Gradle"/>
@@ -865,6 +910,14 @@
             <sha256 value="d8ef04000565affac019b7a55de5bb7cc82ab0403295285ef49f6c8c2745afeb" origin="Generated by Gradle"/>
          </artifact>
       </component>
+      <component group="org.apache.lucene" name="lucene-analysis-common" version="10.5.0">
+         <artifact name="lucene-analysis-common-10.5.0.jar">
+            <sha256 value="922e217fe5cc88305b5a8e057cd80a30b7de96be2ff20dc4327fb524d26e6a25" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="lucene-analysis-common-10.5.0.pom">
+            <sha256 value="8ad3288be355a6dca42678ae6abaa4726fe4a320b1657e7a7a455ae684edad27" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
       <component group="org.apache.lucene" name="lucene-analysis-common" version="9.12.3">
          <artifact name="lucene-analysis-common-9.12.3.jar">
             <sha256 value="fa571bd7caf0f0b4faf46a72ca004a7836f348c31d92bd522dddcc3d128d287e" origin="Generated by Gradle"/>
@@ -873,6 +926,30 @@
             <sha256 value="3242d6696252c6ce33744087bbb821c384373c7819c3a054539b59876e3df39a" origin="Generated by Gradle"/>
          </artifact>
       </component>
+      <component group="org.apache.lucene" name="lucene-analysis-morfologik" version="10.5.0">
+         <artifact name="lucene-analysis-morfologik-10.5.0.jar">
+            <sha256 value="7438fa8afd11dc9b606e911ef5a922cf642a3d45e32c5bbd456e54d731810dbe" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="lucene-analysis-morfologik-10.5.0.pom">
+            <sha256 value="df9863e0db7416ba1d0d1fa657b2728e31a2ac8150bdd87fa272b100aa0838fb" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
+      <component group="org.apache.lucene" name="lucene-analysis-stempel" version="10.5.0">
+         <artifact name="lucene-analysis-stempel-10.5.0.jar">
+            <sha256 value="f46699a4457e1cec1035737c37be1f061eaf32d96b6679968bbda19d1b12ed2b" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="lucene-analysis-stempel-10.5.0.pom">
+            <sha256 value="da627cd84d6e29eeae63fd1881420b1b9f1b3619c106016a5ba0338351a292ac" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
+      <component group="org.apache.lucene" name="lucene-core" version="10.5.0">
+         <artifact name="lucene-core-10.5.0.jar">
+            <sha256 value="ec05ee432860dc6116765fc6bd9ffccf65311cb0430a42a3a90f09fffcd310b1" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="lucene-core-10.5.0.pom">
+            <sha256 value="356efef0e44ed7e1979af9d39c4403fea268edc5ede1c6fe4c006e58f218a0dc" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
       <component group="org.apache.lucene" name="lucene-core" version="9.12.3">
          <artifact name="lucene-core-9.12.3.jar">
             <sha256 value="b64a3f8098a7572034fb30085cdee01b34ec81fb0e5a31b471536af58dc6c01b" origin="Generated by Gradle"/>
@@ -1066,6 +1143,19 @@
             <sha256 value="a941745d7faeb8dc9a75edc2c330c994b7440b9a44d21142716b6053967a41c1" origin="Generated by Gradle"/>
          </artifact>
       </component>
+      <component group="org.apache.opennlp" name="opennlp" version="2.5.4">
+         <artifact name="opennlp-2.5.4.pom">
+            <sha256 value="433d0873ec27cfe1b1a20ea643773a2f29a40a1fdea1bb5d1828b2f35350c29b" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
+      <component group="org.apache.opennlp" name="opennlp-tools" version="2.5.4">
+         <artifact name="opennlp-tools-2.5.4.jar">
+            <sha256 value="5efedb26d0e97e53707a8d2f0e2e28f05fcf654fce2a33f8bfe11c2d4c2a4fe9" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="opennlp-tools-2.5.4.pom">
+            <sha256 value="fd4557916e65775956c37e8cdda7d9ad2bca778a29c98d6a5c6e31b61f83d560" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
       <component group="org.apache.velocity" name="velocity-engine-core" version="2.4.1">
          <artifact name="velocity-engine-core-2.4.1.jar">
             <sha256 value="1c19157d1171d560088e485be97c93a7a2f7e9f56e517f0a30273c5c39df6231" origin="Generated by Gradle"/>
@@ -1097,6 +1187,35 @@
             <sha256 value="22b87dda9aab83fa1d0f3ea409b524e7a44921cf8f5f87999cf59046f0fe0bc3" origin="Generated by Gradle"/>
          </artifact>
       </component>
+      <component group="org.carrot2" name="morfologik-fsa" version="2.1.9">
+         <artifact name="morfologik-fsa-2.1.9.jar">
+            <sha256 value="1bfefce937df14cc94d32a98ce59c33f4d5b6c0eddbb436b6bfe27ff2120a23d" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="morfologik-fsa-2.1.9.pom">
+            <sha256 value="1097b12e6ede04b5a4e09b77233ac0943d8a6020edced7fd65ec97f2b02e103c" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
+      <component group="org.carrot2" name="morfologik-parent" version="2.1.9">
+         <artifact name="morfologik-parent-2.1.9.pom">
+            <sha256 value="59c72168787ba151785125e34472e7841c5ff18bde176c2db7349c06917a0627" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
+      <component group="org.carrot2" name="morfologik-polish" version="2.1.9">
+         <artifact name="morfologik-polish-2.1.9.jar">
+            <sha256 value="e503682b3f4e8bb7a5d05820b0e2a4a19d4bad43dae20f64741786658a9cf478" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="morfologik-polish-2.1.9.pom">
+            <sha256 value="85595c01c592576b91f59600c040bf2ceabaf61afc25ffa8507e4da981a54fb5" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
+      <component group="org.carrot2" name="morfologik-stemming" version="2.1.9">
+         <artifact name="morfologik-stemming-2.1.9.jar">
+            <sha256 value="6170895b2315b697f4da5630caf57c6c441f1cb419d89d1cb5326b0673293e8a" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="morfologik-stemming-2.1.9.pom">
+            <sha256 value="0b1495ad4d54b8dd4d309e1b445625ff2c788cc68818b5ded5cfe4f0d2f891a2" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
       <component group="org.checkerframework" name="checker-qual" version="3.52.1">
          <artifact name="checker-qual-3.52.1.jar">
             <sha256 value="934641a18c8461bf66d7e939b2b054bf2a518ed4188fd7d6836a65b038f5364a" origin="Generated by Gradle"/>
@@ -1915,6 +2034,14 @@
             <sha256 value="ba01ae7a744cb52fe8ecf3b023cbc32e0ccc8c6beef9f26de77a47808239447d" origin="Generated by Gradle"/>
          </artifact>
       </component>
+      <component group="ua.net.nlp" name="morfologik-ukrainian-search" version="4.9.1">
+         <artifact name="morfologik-ukrainian-search-4.9.1.jar">
+            <sha256 value="463d9054b8d4cfacb9cd69566395826a6fe32fb6e7da91dbf79d37ddf7d56ba0" origin="Generated by Gradle"/>
+         </artifact>
+         <artifact name="morfologik-ukrainian-search-4.9.1.pom">
+            <sha256 value="6a66d8efe6a5c774932401b04411a8feca7d6628631e4ef80819cfafa133dc38" origin="Generated by Gradle"/>
+         </artifact>
+      </component>
       <component group="us.springett" name="cpe-parser" version="3.0.1">
          <artifact name="cpe-parser-3.0.1.jar">
             <sha256 value="f98a50dce0a381e08f0f0ac067801c7f4c51dc7f8f1fe7a3c9960c684a809705" origin="Generated by Gradle"/>

mkdocs.yml

@@ -0,0 +1,104 @@
+site_name: Radixor
+site_description: High-performance multi-language stemming toolkit for Java
+site_url: https://leogalambos.github.io/Radixor/
+repo_url: https://github.com/leogalambos/Radixor
+repo_name: leogalambos/Radixor
+copyright: "&copy; 2026 Egothor. Licensed under <a href='https://github.com/leogalambos/Radixor/blob/main/LICENSE'>BSD-3-Clause</a>."
+
+theme:
+  name: material
+  language: en
+  features:
+    - navigation.instant
+    - navigation.sections
+    - navigation.top
+    - search.suggest
+    - search.highlight
+    - content.code.copy
+  palette:
+    - scheme: default
+      primary: indigo
+      accent: indigo
+
+extra: 
+  generator: false
+
+extra_css:
+  - assets/stylesheets/extra.css
+
+markdown_extensions:
+  - admonition
+  - attr_list
+  - md_in_html
+  - pymdownx.details
+  - pymdownx.highlight
+  - pymdownx.superfences
+  - tables
+
+nav:
+  - Home: index.md
+
+  - Start:
+     - Fast Track: fast-track.md
+     - Quick Start: quick-start.md
+     - Integration Deep Dive: integration-deep-dive.md
+
+  - Integration:
+     - Overview: programmatic-usage.md
+     - Loading and Building Stemmers: programmatic-loading-and-building.md
+     - Querying and Ambiguity Handling: programmatic-querying-and-ambiguity.md
+     - Extending and Persisting Compiled Tries: programmatic-extending-and-persistence.md
+     - Migration and Backward Compatibility: migration-and-backward-compatibility.md
+     - CLI Compilation: cli-compilation.md
+
+  - Dictionaries and Languages:
+     - Built-in Languages: built-in-languages.md
+     - Dictionary Format: dictionary-format.md
+     - Contributing Dictionaries: contributing-dictionaries.md
+
+  - Architecture and Semantics:
+     - Overview: architecture-and-reduction.md
+     - Architecture: architecture.md
+     - Reduction Semantics: reduction-semantics.md
+     - Lookup Edge Optimization: lookup-edge-optimization.md
+     - Compatibility and Guarantees: compatibility-and-guarantees.md
+
+  - Benchmarks:
+     - How to Read Benchmarks: benchmarking.md
+     - Benchmark Results: benchmarks/index.md
+     - Reference:
+        - Methodology: benchmarks/reference/methodology.md
+        - Linguistic Quality Methodology: benchmarks/reference/linguistic-quality.md
+        - Tested Stemmers: benchmarks/reference/tested-stemmers.md
+        - Reproducibility and Raw Data: benchmarks/reference/reproducibility.md
+        - Corpora: benchmarks/reference/corpora.md
+        - Environment and Reports: benchmarks/reference/environment.md
+        - English Dictionary Coverage: benchmarks/reference/english-coverage.md
+        - Candidate Evaluation: benchmarks/reference/candidates.md
+     - Language Results:
+        - Overview: benchmarks/languages/index.md
+        - Czech: benchmarks/languages/czech.md
+        - Danish: benchmarks/languages/danish.md
+        - Dutch: benchmarks/languages/dutch.md
+        - English: benchmarks/languages/english.md
+        - Finnish: benchmarks/languages/finnish.md
+        - French: benchmarks/languages/french.md
+        - German: benchmarks/languages/german.md
+        - Hungarian: benchmarks/languages/hungarian.md
+        - Italian: benchmarks/languages/italian.md
+        - Norwegian Bokmal: benchmarks/languages/norwegian-bokmal.md
+        - Norwegian Nynorsk: benchmarks/languages/norwegian-nynorsk.md
+        - Persian: benchmarks/languages/persian.md
+        - Polish: benchmarks/languages/polish.md
+        - Portuguese: benchmarks/languages/portuguese.md
+        - Russian: benchmarks/languages/russian.md
+        - Spanish: benchmarks/languages/spanish.md
+        - Swedish: benchmarks/languages/swedish.md
+        - Ukrainian: benchmarks/languages/ukrainian.md
+        - Yiddish: benchmarks/languages/yiddish.md
+
+  - Quality and Operations:
+     - Quality and Operations: quality-and-operations.md
+     - Stemming Quality: stemming-quality.md
+     - Reports: reports.md
+     - Test taxonomy and execution filtering: test-taxonomy-and-filtering.md

src/jmh/java/org/egothor/stemmer/benchmark/BenchmarkCorpusSupport.java

@@ -113,12 +113,12 @@ final class BenchmarkCorpusSupport {
             dictionaryBuilder.append(stem);
             lookupKeys.add(stem);
             for (String variant : variants) {
-                dictionaryBuilder.append(' ').append(variant);
+                dictionaryBuilder.append('\t').append(variant);
                 lookupKeys.add(variant);
             }
 
             final String homograph = createHomograph(index);
-            dictionaryBuilder.append(' ').append(homograph);
+            dictionaryBuilder.append('\t').append(homograph);
             lookupKeys.add(homograph);
             ambiguousLookupKeys.add(homograph);
 
@@ -149,7 +149,7 @@ final class BenchmarkCorpusSupport {
         Objects.requireNonNull(reductionSettings, "reductionSettings");
 
         final FrequencyTrie.Builder<String> builder = new FrequencyTrie.Builder<>(String[]::new, reductionSettings);
-        final PatchCommandEncoder encoder = new PatchCommandEncoder();
+        final PatchCommandEncoder encoder = PatchCommandEncoder.builder().build();
 
         StemmerDictionaryParser.parse(
                 new StringReader(corpusText),

src/jmh/java/org/egothor/stemmer/benchmark/BenchmarkTokenSequence.java

@@ -0,0 +1,185 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ * 
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ * 
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ * 
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.util.Objects;
+
+/**
+ * Reusable deterministic token sequence for benchmark-only token streams.
+ *
+ * <p>
+ * The sequence keeps stable token ordering and offset progression while avoiding
+ * per-token object creation during iteration.
+ * </p>
+ */
+final class BenchmarkTokenSequence {
+
+    /**
+     * Shared backing corpus as character arrays.
+     */
+    private char[][] tokenCharacters;
+
+    /**
+     * Number of active tokens in the sequence.
+     */
+    private int tokenCount;
+
+    /**
+     * Cursor index for the currently emitted token.
+     */
+    private int cursor;
+
+    /**
+     * Current token character array.
+     */
+    private char[] currentToken;
+
+    /**
+     * Start offset of the current token.
+     */
+    private int currentStartOffset;
+
+    /**
+     * End offset of the current token.
+     */
+    private int currentEndOffset;
+
+    /**
+     * Offset of the next token start.
+     */
+    private int nextOffset;
+
+    /**
+     * Creates a reusable token sequence.
+ *
+     * @param tokens token corpus source
+     */
+    BenchmarkTokenSequence(final String[] tokens) {
+        setTokens(tokens);
+    }
+
+    /**
+     * Sets a new token corpus for this sequence.
+     *
+     * <p>
+     * The sequence stores copied character arrays so token reads can be reused
+     * without creating per-token objects during benchmark iteration.
+     * </p>
+     *
+     * @param tokens new token corpus
+     */
+    void setTokens(final String[] tokens) {
+        Objects.requireNonNull(tokens, "tokens");
+        this.tokenCharacters = new char[tokens.length][];
+        for (int index = 0; index < tokens.length; index++) {
+            final String token = Objects.requireNonNull(tokens[index], "tokens[" + index + "]");
+            this.tokenCharacters[index] = token.toCharArray();
+        }
+
+        this.tokenCount = this.tokenCharacters.length;
+        reset();
+    }
+
+    /**
+     * Resets stream position for reuse.
+     */
+    void reset() {
+        this.cursor = 0;
+        this.nextOffset = 0;
+        this.currentStartOffset = 0;
+        this.currentEndOffset = 0;
+        this.currentToken = null;
+    }
+
+    /**
+     * Returns whether at least one token remains in the sequence.
+     *
+     * @return true if a token can be emitted
+     */
+    boolean hasNext() {
+        return this.cursor < this.tokenCount;
+    }
+
+    /**
+     * Advances to the next token.
+     *
+     * @return true if a token was emitted
+     */
+    boolean advance() {
+        if (!hasNext()) {
+            return false;
+        }
+
+        final char[] token = this.tokenCharacters[this.cursor];
+        this.currentToken = token;
+        this.currentStartOffset = this.nextOffset;
+        this.currentEndOffset = this.currentStartOffset + token.length;
+        this.nextOffset = this.currentEndOffset + 1;
+        this.cursor++;
+        return true;
+    }
+
+    /**
+     * Returns the current token in the sequence.
+     *
+     * @return current token character array
+     */
+    char[] currentToken() {
+        return this.currentToken;
+    }
+
+    /**
+     * Returns current token start offset for token stream attributes.
+     *
+     * @return start offset
+     */
+    int currentStartOffset() {
+        return this.currentStartOffset;
+    }
+
+    /**
+     * Returns current token end offset for token stream attributes.
+ *
+     * @return end offset
+     */
+    int currentEndOffset() {
+        return this.currentEndOffset;
+    }
+
+    /**
+     * Returns final stream offset value used by {@code end()}.
+     *
+     * @return final offset
+     */
+    int endOffset() {
+        return this.nextOffset > 0 ? this.nextOffset - 1 : 0;
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/BenchmarkTokenStream.java

@@ -0,0 +1,146 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.IOException;
+
+import org.apache.lucene.analysis.TokenStream;
+import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
+import org.apache.lucene.analysis.tokenattributes.OffsetAttribute;
+import org.apache.lucene.analysis.tokenattributes.PositionIncrementAttribute;
+
+/**
+ * Reusable Lucene {@link TokenStream} backed by a deterministic token array.
+ *
+ * <p>
+ * Instances are mutable and intended for one JMH worker thread. The stream
+ * copies configured token text into reusable character storage so benchmark
+ * iteration can replay the same token sequence without mutating the shared
+ * source array.
+ * </p>
+ */
+final class BenchmarkTokenStream extends TokenStream {
+
+    /**
+     * Current token text attribute.
+     */
+    private final CharTermAttribute charTermAttribute;
+
+    /**
+     * Offset attribute used by Lucene filters that inspect offsets.
+     */
+    private final OffsetAttribute offsetAttribute;
+
+    /**
+     * Position increment attribute for one-token-at-a-time streams.
+     */
+    private final PositionIncrementAttribute positionIncrementAttribute;
+
+    /**
+     * Reusable token sequence.
+     */
+    private final BenchmarkTokenSequence tokenSequence;
+
+    /**
+     * Creates a stream over the supplied tokens.
+     *
+     * @param tokens initial token corpus
+     */
+    BenchmarkTokenStream(final String[] tokens) {
+        this.tokenSequence = new BenchmarkTokenSequence(tokens);
+        this.charTermAttribute = addAttribute(CharTermAttribute.class);
+        this.offsetAttribute = addAttribute(OffsetAttribute.class);
+        this.positionIncrementAttribute = addAttribute(PositionIncrementAttribute.class);
+    }
+
+    /**
+     * Replaces the configured token corpus.
+     *
+     * @param tokens new token corpus
+     */
+    void setTokens(final String[] tokens) {
+        this.tokenSequence.setTokens(tokens);
+    }
+
+    /**
+     * Returns whether all configured tokens have been emitted.
+     *
+     * @return {@code true} after the current pass is exhausted
+     */
+    boolean isDrained() {
+        return !this.tokenSequence.hasNext();
+    }
+
+    /**
+     * {@inheritDoc}
+     */
+    @Override
+    public boolean incrementToken() throws IOException {
+        if (!this.tokenSequence.advance()) {
+            return false;
+        }
+
+        clearAttributes();
+        final char[] token = this.tokenSequence.currentToken();
+        this.charTermAttribute.copyBuffer(token, 0, token.length);
+        this.positionIncrementAttribute.setPositionIncrement(1);
+        this.offsetAttribute.setOffset(this.tokenSequence.currentStartOffset(), this.tokenSequence.currentEndOffset());
+        return true;
+    }
+
+    /**
+     * {@inheritDoc}
+     */
+    @Override
+    public void reset() throws IOException {
+        super.reset();
+        this.tokenSequence.reset();
+    }
+
+    /**
+     * {@inheritDoc}
+     */
+    @Override
+    public void end() throws IOException {
+        super.end();
+        final int endOffset = this.tokenSequence.endOffset();
+        this.offsetAttribute.setOffset(endOffset, endOffset);
+    }
+
+    /**
+     * {@inheritDoc}
+     */
+    @Override
+    public void close() throws IOException {
+        super.close();
+        this.charTermAttribute.setEmpty();
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/Cistem.java

@@ -0,0 +1,178 @@
+/*******************************************************************************
+ * MIT License
+ *
+ * Copyright (c) 2017 Leonie Weißweiler
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * Source: CISTEM German stemmer
+ *   Authors: Leonie Weissweiler, Alexander Fraser
+ *   https://github.com/LeonieWeissweiler/CISTEM
+ *   https://www.cis.lmu.de/~weissweiler/cistem/
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.util.regex.Pattern;
+
+public final class Cistem {
+
+    private static final Pattern GE_PATTERN = Pattern.compile("^ge(.{4,})");
+    private static final Pattern DOLLAR1_PATTERN = Pattern.compile("(.)\\1");
+    private static final Pattern ND_PATTERN = Pattern.compile("nd$");
+    private static final Pattern EMR_PATTERN = Pattern.compile("e[mr]$");
+    private static final Pattern T_PATTERN = Pattern.compile("t$");
+    private static final Pattern ESN_PATTERN = Pattern.compile("[esn]$");
+    private static final Pattern STAR_PATTERN = Pattern.compile("(.)\\*");
+
+    private Cistem() {
+    }
+
+    public static String stem(final String word) {
+        return stem(word, false);
+    }
+
+    public static String stem(final String word, final boolean caseInsensitive) {
+        if (word.isEmpty()) {
+            return word;
+        }
+
+        String normalized = word;
+        normalized = normalized.replace("Ü", "U");
+        normalized = normalized.replace("Ö", "O");
+        normalized = normalized.replace("Ä", "A");
+        normalized = normalized.replace("ü", "u");
+        normalized = normalized.replace("ö", "o");
+        normalized = normalized.replace("ä", "a");
+
+        final boolean uppercase = Character.isUpperCase(normalized.charAt(0));
+
+        normalized = normalized.toLowerCase();
+        normalized = normalized.replace("ß", "ss");
+        normalized = GE_PATTERN.matcher(normalized).replaceAll("$1");
+        normalized = normalized.replace("sch", "$");
+        normalized = normalized.replace("ei", "%");
+        normalized = normalized.replace("ie", "&");
+        normalized = DOLLAR1_PATTERN.matcher(normalized).replaceAll("$1*");
+
+        while (normalized.length() > 3) {
+            if (normalized.length() > 5) {
+                String newWord = EMR_PATTERN.matcher(normalized).replaceAll("");
+                if (!normalized.equals(newWord)) {
+                    normalized = newWord;
+                    continue;
+                }
+
+                newWord = ND_PATTERN.matcher(normalized).replaceAll("");
+                if (!normalized.equals(newWord)) {
+                    normalized = newWord;
+                    continue;
+                }
+            }
+
+            if (!uppercase || caseInsensitive) {
+                final String newWord = T_PATTERN.matcher(normalized).replaceAll("");
+                if (!normalized.equals(newWord)) {
+                    normalized = newWord;
+                    continue;
+                }
+            }
+
+            final String newWord = ESN_PATTERN.matcher(normalized).replaceAll("");
+            if (!normalized.equals(newWord)) {
+                normalized = newWord;
+            } else {
+                break;
+            }
+        }
+
+        normalized = STAR_PATTERN.matcher(normalized).replaceAll("$1$1");
+        normalized = normalized.replace("&", "ie");
+        normalized = normalized.replace("%", "ei");
+        normalized = normalized.replace("$", "sch");
+
+        return normalized;
+    }
+
+    public static String[] segment(final String word) {
+        return segment(word, false);
+    }
+
+    public static String[] segment(final String word, final boolean caseInsensitive) {
+        if (word.isEmpty()) {
+            return new String[] {"", ""};
+        }
+
+        int restLength = 0;
+        final boolean uppercase = Character.isUpperCase(word.charAt(0));
+        String normalized = word.toLowerCase();
+        final String original = new String(normalized);
+
+        normalized = normalized.replace("sch", "$");
+        normalized = normalized.replace("ei", "%");
+        normalized = normalized.replace("ie", "&");
+        normalized = DOLLAR1_PATTERN.matcher(normalized).replaceAll("$1*");
+
+        while (normalized.length() > 3) {
+            if (normalized.length() > 5) {
+                String newWord = normalized.replaceAll("e[mr]$", "");
+                if (!normalized.equals(newWord)) {
+                    restLength += 2;
+                    normalized = newWord;
+                    continue;
+                }
+
+                newWord = normalized.replaceAll("nd$", "");
+                if (!normalized.equals(newWord)) {
+                    restLength += 2;
+                    normalized = newWord;
+                    continue;
+                }
+            }
+
+            if (!uppercase || caseInsensitive) {
+                final String newWord = normalized.replaceAll("t$", "");
+                if (!normalized.equals(newWord)) {
+                    restLength += 1;
+                    normalized = newWord;
+                    continue;
+                }
+            }
+
+            final String newWord = normalized.replaceAll("[esn]$", "");
+            if (!normalized.equals(newWord)) {
+                restLength += 1;
+                normalized = newWord;
+            } else {
+                break;
+            }
+        }
+
+        normalized = normalized.replaceAll("(.)\\*", "$1$1");
+        normalized = normalized.replace("&", "ie");
+        normalized = normalized.replace("%", "ei");
+        normalized = normalized.replace("$", "sch");
+
+        String rest = "";
+        if (restLength != 0) {
+            rest = original.substring(original.length() - restLength);
+        }
+
+        return new String[] {normalized, rest};
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/EnglishComparisonCorpus.java

@@ -30,43 +30,24 @@
  ******************************************************************************/
 package org.egothor.stemmer.benchmark;
 
-import java.util.ArrayList;
-import java.util.List;
-import java.util.Locale;
+import java.io.IOException;
+
+import org.egothor.stemmer.StemmerPatchTrieLoader;
 
 /**
  * Builds a deterministic English token corpus for side-by-side stemming
- * benchmarks.
+ * benchmarks from the bundled Radixor English dictionary resource.
  *
  * <p>
- * The generated corpus mixes:
- * </p>
- * <ul>
- * <li>simple inflections</li>
- * <li>common derivational forms</li>
- * <li>US/UK spelling families</li>
- * <li>forms that are suitable for comparison against the bundled
- * {@code US_UK_PROFI} Radixor dictionary</li>
- * </ul>
- *
- * <p>
- * The goal is not to simulate natural language frequency distribution exactly,
- * but to provide a stable and reproducible comparison workload for benchmark
- * runs and regression tracking.
+ * The dictionary resource stores the expected stem as the first tab-separated
+ * field on each line and its surface variants on the same line. This helper
+ * uses only token/root pairs where the token differs from the expected root for
+ * timing. Resources smaller than the shared timing minimum are repeated
+ * deterministically by {@link LanguageBenchmarkCorpus}.
  * </p>
  */
 final class EnglishComparisonCorpus {
 
-    /**
-     * Canonical lexical bases used to generate the token workload.
-     */
-    private static final String[] BASES = { "analyze", "analyse", "color", "colour", "center", "centre", "organize",
-            "organise", "optimize", "optimise", "characterize", "characterise", "connect", "construct", "compute",
-            "design", "develop", "engineer", "govern", "improve", "index", "inform", "manage", "model", "observe",
-            "operate", "perform", "predict", "prepare", "process", "project", "protect", "publish", "query", "reduce",
-            "refresh", "render", "resolve", "return", "search", "select", "signal", "store", "structure", "support",
-            "transform", "update", "validate", "value" };
-
     /**
      * Utility class.
      */
@@ -77,64 +58,21 @@ final class EnglishComparisonCorpus {
     /**
      * Creates a deterministic token corpus for English stemming comparison.
      *
-     * @param familyCount number of generated lexical families
      * @return token array in stable order
+     * @throws IOException if the bundled English resource cannot be read
      */
-    static String[] createTokens(final int familyCount) {
-        if (familyCount < 1) {
-            throw new IllegalArgumentException("familyCount must be at least 1.");
-        }
-
-        final List<String> tokens = new ArrayList<>(familyCount * 14);
-
-        for (int index = 0; index < familyCount; index++) {
-            final String base = createBase(index);
-
-            tokens.add(base);
-            tokens.add(base + "s");
-            tokens.add(base + "ed");
-            tokens.add(base + "ing");
-            tokens.add(base + "er");
-            tokens.add(base + "ers");
-            tokens.add(base + "ly");
-            tokens.add(base + "ness");
-            tokens.add(base + "ment");
-            tokens.add(base + "ments");
-            tokens.add(base + "able");
-            tokens.add(base + "ability");
-
-            if (base.endsWith("ize")) {
-                tokens.add(base.substring(0, base.length() - 3) + "isation");
-                tokens.add(base.substring(0, base.length() - 3) + "ised");
-            }
-
-            if (base.endsWith("ise")) {
-                tokens.add(base.substring(0, base.length() - 3) + "ization");
-                tokens.add(base.substring(0, base.length() - 3) + "ized");
-            }
-        }
-
-        return tokens.toArray(String[]::new);
+    static String[] createTokens() throws IOException {
+        return createCorpus().tokens();
     }
 
     /**
-     * Creates one deterministic base token.
+     * Creates a deterministic changed-token corpus and expected root array for
+     * English stemming comparison.
      *
-     * @param index base ordinal
-     * @return generated lexical base
+     * @return token corpus with expected roots
+     * @throws IOException if the bundled English resource cannot be read
      */
-    private static String createBase(final int index) {
-        return (BASES[index % BASES.length] + suffix(index)).toLowerCase(Locale.ROOT);
-    }
-
-    /**
-     * Creates a compact discriminator suffix so that large corpora remain unique
-     * while retaining stable lexical families.
-     *
-     * @param value ordinal value
-     * @return compact discriminator
-     */
-    private static String suffix(final int value) {
-        return Integer.toString(value, Character.MAX_RADIX);
+    static LanguageBenchmarkCorpus.Corpus createCorpus() throws IOException {
+        return LanguageBenchmarkCorpus.createChangedCorpus(StemmerPatchTrieLoader.Language.US_UK);
     }
 }

src/jmh/java/org/egothor/stemmer/benchmark/EnglishHunspellStemmerComparisonBenchmarkQuality.java

@@ -0,0 +1,280 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.IOException;
+import java.io.InputStream;
+import java.text.ParseException;
+import java.util.List;
+import java.util.Objects;
+import java.util.concurrent.TimeUnit;
+
+import org.apache.lucene.analysis.LowerCaseFilter;
+import org.apache.lucene.analysis.TokenStream;
+import org.apache.lucene.analysis.hunspell.Dictionary;
+import org.apache.lucene.analysis.hunspell.HunspellStemFilter;
+import org.apache.lucene.analysis.hunspell.SortingStrategy;
+import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
+import org.apache.lucene.analysis.tokenattributes.PositionIncrementAttribute;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+import org.openjdk.jmh.annotations.AuxCounters;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Fork;
+import org.openjdk.jmh.annotations.Level;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.Setup;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.Warmup;
+import org.openjdk.jmh.infra.Blackhole;
+
+/**
+ * Emits exact-root agreement metrics for the benchmark-only English Hunspell
+ * comparison.
+ */
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.NANOSECONDS)
+@Warmup(iterations = 0)
+@Measurement(iterations = 1, time = 1, timeUnit = TimeUnit.MILLISECONDS)
+@Fork(0)
+public class EnglishHunspellStemmerComparisonBenchmarkQuality {
+
+    /**
+     * Shared English quality corpus and Hunspell dictionary.
+     */
+    @State(Scope.Benchmark)
+    public static class SharedState {
+
+        /**
+         * Complete English resource-derived corpus.
+         */
+        private LanguageBenchmarkCorpus.Corpus corpus;
+
+        /**
+         * Parsed benchmark-only Hunspell dictionary.
+         */
+        private Dictionary dictionary;
+
+        /**
+         * Initializes quality resources.
+         *
+         * @throws IOException    if corpus or dictionary loading fails
+         * @throws ParseException if the Hunspell dictionary cannot be parsed
+         */
+        @Setup(Level.Trial)
+        public void setUp() throws IOException, ParseException {
+            this.corpus = LanguageBenchmarkCorpus.createFullCorpus(StemmerPatchTrieLoader.Language.US_UK);
+            this.dictionary = loadEnglishDictionary();
+        }
+    }
+
+    /**
+     * JMH auxiliary counters for exact-root agreement.
+     */
+    @State(Scope.Thread)
+    @AuxCounters(AuxCounters.Type.EVENTS)
+    public static class AccuracyCounters {
+
+        /**
+         * Number of exact-root matches.
+         */
+        public long correctMatches;
+
+        /**
+         * Number of evaluated tokens.
+         */
+        public long evaluatedTokens;
+
+        /**
+         * Number of exact-root matches where the input token differs from the
+         * expected root.
+         */
+        public long changedCorrectMatches;
+
+        /**
+         * Number of evaluated tokens where the input token differs from the expected
+         * root.
+         */
+        public long changedEvaluatedTokens;
+
+        /**
+         * Number of exact-root matches where the input token is already the expected
+         * root.
+         */
+        public long rootPreservedMatches;
+
+        /**
+         * Number of evaluated tokens where the input token is already the expected
+         * root.
+         */
+        public long rootEvaluatedTokens;
+
+        /**
+         * Resets counters before the measured iteration.
+         */
+        @Setup(Level.Iteration)
+        public void reset() {
+            this.correctMatches = 0L;
+            this.evaluatedTokens = 0L;
+            this.changedCorrectMatches = 0L;
+            this.changedEvaluatedTokens = 0L;
+            this.rootPreservedMatches = 0L;
+            this.rootEvaluatedTokens = 0L;
+        }
+    }
+
+    /**
+     * Evaluates exact-root agreement for English Hunspell.
+     *
+     * @param sharedState shared English quality state
+     * @param counters    JMH auxiliary counters
+     * @param blackhole   result sink
+     * @return exact-root match count
+     * @throws IOException if Lucene token streaming fails
+     */
+    @Benchmark
+    public int luceneHunspellStemFilterAccuracy(final SharedState sharedState, final AccuracyCounters counters,
+            final Blackhole blackhole) throws IOException {
+        final String[] actualStems = firstHunspellOutputs(sharedState.corpus.tokens(), sharedState.dictionary,
+                blackhole);
+        final String[] tokens = sharedState.corpus.tokens();
+        final String[] expectedRoots = sharedState.corpus.expectedRoots();
+
+        int correct = 0;
+        int changedCorrect = 0;
+        int changedEvaluated = 0;
+        int rootPreserved = 0;
+        int rootEvaluated = 0;
+        for (int index = 0; index < actualStems.length; index++) {
+            final String token = tokens[index];
+            final String expectedRoot = expectedRoots[index];
+            final boolean exact = Objects.equals(expectedRoot, actualStems[index]);
+            if (exact) {
+                correct++;
+            }
+            if (Objects.equals(token, expectedRoot)) {
+                rootEvaluated++;
+                if (exact) {
+                    rootPreserved++;
+                }
+            } else {
+                changedEvaluated++;
+                if (exact) {
+                    changedCorrect++;
+                }
+            }
+        }
+
+        counters.correctMatches += correct;
+        counters.evaluatedTokens += actualStems.length;
+        counters.changedCorrectMatches += changedCorrect;
+        counters.changedEvaluatedTokens += changedEvaluated;
+        counters.rootPreservedMatches += rootPreserved;
+        counters.rootEvaluatedTokens += rootEvaluated;
+        return correct;
+    }
+
+    /**
+     * Extracts the first emitted Hunspell stem for each input token.
+     *
+     * @param tokens     token corpus
+     * @param dictionary Hunspell dictionary
+     * @param blackhole  result sink
+     * @return first emitted term per input token
+     * @throws IOException if Lucene streaming fails
+     */
+    private static String[] firstHunspellOutputs(final String[] tokens, final Dictionary dictionary,
+            final Blackhole blackhole) throws IOException {
+        final String[] outputs = new String[tokens.length];
+        final BenchmarkTokenStream input = new BenchmarkTokenStream(tokens);
+        final TokenStream output = new HunspellStemFilter(new LowerCaseFilter(input), dictionary, true);
+        final CharTermAttribute termAttribute = output.addAttribute(CharTermAttribute.class);
+        final PositionIncrementAttribute positionAttribute = output.addAttribute(PositionIncrementAttribute.class);
+        int inputIndex = -1;
+        boolean recordedForPosition = false;
+
+        output.reset();
+        while (output.incrementToken()) {
+            final int positionIncrement = positionAttribute.getPositionIncrement();
+            if (positionIncrement > 0) {
+                inputIndex += positionIncrement;
+                recordedForPosition = false;
+            }
+            if (inputIndex >= 0 && inputIndex < outputs.length && !recordedForPosition) {
+                outputs[inputIndex] = termAttribute.toString();
+                recordedForPosition = true;
+            }
+            blackhole.consume(termAttribute);
+        }
+        output.end();
+        output.close();
+
+        for (int index = 0; index < outputs.length; index++) {
+            if (outputs[index] == null) {
+                outputs[index] = tokens[index];
+            }
+        }
+        return outputs;
+    }
+
+    /**
+     * Loads the benchmark-only English Hunspell dictionary.
+     *
+     * @return parsed Hunspell dictionary
+     * @throws IOException    if dictionary resources cannot be read
+     * @throws ParseException if the Hunspell dictionary cannot be parsed
+     */
+    private static Dictionary loadEnglishDictionary() throws IOException, ParseException {
+        final ClassLoader classLoader = EnglishHunspellStemmerComparisonBenchmarkQuality.class.getClassLoader();
+        try (InputStream affixStream = openRequiredResource(classLoader, "hunspell/en/index.aff");
+                InputStream dictionaryStream = openRequiredResource(classLoader, "hunspell/en/index.dic")) {
+            return new Dictionary(affixStream, List.of(dictionaryStream), true, SortingStrategy.inMemory());
+        }
+    }
+
+    /**
+     * Opens a required classpath resource.
+     *
+     * @param classLoader class loader
+     * @param path        resource path
+     * @return resource stream
+     */
+    private static InputStream openRequiredResource(final ClassLoader classLoader, final String path) {
+        final InputStream stream = classLoader.getResourceAsStream(path);
+        if (stream == null) {
+            throw new IllegalStateException("Missing benchmark-only Hunspell resource: " + path);
+        }
+        return stream;
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/EnglishRadixorDictionaryCoverageBenchmark.java

@@ -0,0 +1,431 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.BufferedReader;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.InputStreamReader;
+import java.nio.charset.StandardCharsets;
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.HashMap;
+import java.util.HashSet;
+import java.util.List;
+import java.util.Map;
+import java.util.Objects;
+import java.util.Set;
+import java.util.concurrent.TimeUnit;
+import java.util.zip.GZIPInputStream;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.FrequencyTrieBuilders;
+import org.egothor.stemmer.PatchCommandEncoder;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.ReductionSettings;
+import org.egothor.stemmer.StemmerDictionaryParser;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+import org.egothor.stemmer.WordTraversalDirection;
+import org.openjdk.jmh.annotations.AuxCounters;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Level;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+import org.openjdk.jmh.annotations.Param;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.Setup;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.Warmup;
+import org.openjdk.jmh.infra.Blackhole;
+
+/**
+ * Measures Radixor English stemming quality and changed-token speed when the
+ * runtime trie is trained from a deterministic percentage of dictionary rows.
+ *
+ * <p>
+ * The measured stemmer always uses {@link CompiledPatchCommand} values. Quality
+ * is evaluated against the complete English dictionary corpus, while speed is
+ * measured over the complete changed-token English corpus used by the comparison
+ * benchmarks.
+ * </p>
+ */
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.NANOSECONDS)
+@Warmup(iterations = 3, time = 1, timeUnit = TimeUnit.SECONDS)
+@Measurement(iterations = 5, time = 1, timeUnit = TimeUnit.SECONDS)
+public class EnglishRadixorDictionaryCoverageBenchmark {
+
+    /**
+     * Shared benchmark state for one dictionary-row coverage percentage.
+     */
+    @State(Scope.Benchmark)
+    public static class CoverageState {
+
+        /**
+         * Percentage of parsed English dictionary rows used to build the Radixor trie.
+         */
+        @Param({ "100", "90", "80", "70", "60", "50", "40", "30", "20", "10" })
+        public int coveragePercent;
+
+        /**
+         * Full English corpus used for exact-root accounting.
+         */
+        private LanguageBenchmarkCorpus.Corpus fullCorpus;
+
+        /**
+         * Complete changed-token English corpus used for speed measurement.
+         */
+        private LanguageBenchmarkCorpus.Corpus changedCorpus;
+
+        /**
+         * Radixor stemmer backed by a trie built from selected dictionary rows.
+         */
+        private RadixorBenchmarkStemmer stemmer;
+
+        /**
+         * Parsed dictionary row count before deterministic coverage selection.
+         */
+        private int totalRowCount;
+
+        /**
+         * Selected dictionary row count for the configured coverage percentage.
+         */
+        private int selectedRowCount;
+
+        /**
+         * Builds the reduced dictionary trie and shared corpora before measurement.
+         *
+         * @throws IOException if the English dictionary resource cannot be read
+         */
+        @Setup(Level.Trial)
+        public void setUp() throws IOException {
+            final List<DictionaryRow> rows = readEnglishRows();
+            this.totalRowCount = rows.size();
+            final List<DictionaryRow> selectedRows = selectRows(rows, this.coveragePercent);
+            this.selectedRowCount = selectedRows.size();
+            this.fullCorpus = LanguageBenchmarkCorpus.createFullCorpus(StemmerPatchTrieLoader.Language.US_UK);
+            this.changedCorpus = LanguageBenchmarkCorpus.createChangedCorpus(StemmerPatchTrieLoader.Language.US_UK);
+            this.stemmer = new RadixorBenchmarkStemmer(buildCompiledTrie(selectedRows));
+        }
+    }
+
+    /**
+     * JMH auxiliary counters for dictionary-row coverage and exact-root agreement.
+     */
+    @State(Scope.Thread)
+    @AuxCounters(AuxCounters.Type.EVENTS)
+    public static class CoverageCounters {
+
+        /**
+         * Number of exact output/root matches over the full dictionary corpus.
+         */
+        public long correctMatches;
+
+        /**
+         * Number of evaluated tokens over the full dictionary corpus.
+         */
+        public long evaluatedTokens;
+
+        /**
+         * Number of exact output/root matches where token and root differ.
+         */
+        public long changedCorrectMatches;
+
+        /**
+         * Number of evaluated tokens where token and root differ.
+         */
+        public long changedEvaluatedTokens;
+
+        /**
+         * Number of exact output/root matches where token already equals root.
+         */
+        public long rootPreservedMatches;
+
+        /**
+         * Number of evaluated tokens where token already equals root.
+         */
+        public long rootEvaluatedTokens;
+
+        /**
+         * Number of parsed dictionary rows used for trie construction.
+         */
+        public long selectedRows;
+
+        /**
+         * Total number of parsed dictionary rows available.
+         */
+        public long totalRows;
+
+        /**
+         * Resets counters before each measured iteration.
+         */
+        @Setup(Level.Iteration)
+        public void reset() {
+            this.correctMatches = 0L;
+            this.evaluatedTokens = 0L;
+            this.changedCorrectMatches = 0L;
+            this.changedEvaluatedTokens = 0L;
+            this.rootPreservedMatches = 0L;
+            this.rootEvaluatedTokens = 0L;
+            this.selectedRows = 0L;
+            this.totalRows = 0L;
+        }
+    }
+
+    /**
+     * Measures direct Radixor stemming over the complete English changed-token
+     * corpus.
+     *
+     * @param state     shared coverage state
+     * @param blackhole result sink
+     */
+    @Benchmark
+    public void changedTokenStemmingSpeed(final CoverageState state, final Blackhole blackhole) {
+        final String[] tokens = state.changedCorpus.tokens();
+        final RadixorBenchmarkStemmer stemmer = state.stemmer;
+        for (String token : tokens) {
+            blackhole.consume(stemmer.stem(token));
+        }
+    }
+
+    /**
+     * Measures exact-root agreement over the complete English dictionary corpus.
+     *
+     * @param state     shared coverage state
+     * @param counters  auxiliary exact-root counters
+     * @param blackhole result sink
+     * @return exact-root match count for one benchmark operation
+     */
+    @Benchmark
+    public int exactRootAgreement(final CoverageState state, final CoverageCounters counters,
+            final Blackhole blackhole) {
+        final QualityCounts counts = evaluate(state.fullCorpus, state.stemmer, blackhole);
+        counters.correctMatches += counts.correctMatches();
+        counters.evaluatedTokens += counts.evaluatedTokens();
+        counters.changedCorrectMatches += counts.changedCorrectMatches();
+        counters.changedEvaluatedTokens += counts.changedEvaluatedTokens();
+        counters.rootPreservedMatches += counts.rootPreservedMatches();
+        counters.rootEvaluatedTokens += counts.rootEvaluatedTokens();
+        counters.selectedRows += state.selectedRowCount;
+        counters.totalRows += state.totalRowCount;
+        return counts.correctMatches();
+    }
+
+    private static QualityCounts evaluate(final LanguageBenchmarkCorpus.Corpus corpus,
+            final RadixorBenchmarkStemmer stemmer, final Blackhole blackhole) {
+        final String[] tokens = corpus.tokens();
+        final String[] roots = corpus.expectedRoots();
+        int correct = 0;
+        int changedCorrect = 0;
+        int changedEvaluated = 0;
+        int rootPreserved = 0;
+        int rootEvaluated = 0;
+        for (int index = 0; index < tokens.length; index++) {
+            final String token = tokens[index];
+            final String root = roots[index];
+            final String actual = stemmer.stem(token);
+            blackhole.consume(actual);
+            final boolean exact = Objects.equals(root, actual);
+            if (exact) {
+                correct++;
+            }
+            if (Objects.equals(token, root)) {
+                rootEvaluated++;
+                if (exact) {
+                    rootPreserved++;
+                }
+            } else {
+                changedEvaluated++;
+                if (exact) {
+                    changedCorrect++;
+                }
+            }
+        }
+        return new QualityCounts(correct, tokens.length, changedCorrect, changedEvaluated, rootPreserved,
+                rootEvaluated);
+    }
+
+    private static FrequencyTrie<CompiledPatchCommand> buildCompiledTrie(final List<DictionaryRow> rows) {
+        final ReductionSettings settings = new ReductionSettings(
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS,
+                ReductionSettings.DEFAULT_DOMINANT_WINNER_MIN_PERCENT,
+                ReductionSettings.DEFAULT_DOMINANT_WINNER_OVER_SECOND_RATIO,
+                true);
+        final FrequencyTrie.Builder<String> builder = new FrequencyTrie.Builder<>(String[]::new, settings,
+                WordTraversalDirection.BACKWARD);
+        final PatchCommandEncoder encoder = PatchCommandEncoder.builder()
+                .traversalDirection(WordTraversalDirection.BACKWARD)
+                .build();
+
+        for (DictionaryRow row : rows) {
+            builder.put(row.stem(), encoder.encode(row.stem(), row.stem()));
+            for (String variant : row.variants()) {
+                if (!variant.equals(row.stem())) {
+                    builder.put(variant, encoder.encode(variant, row.stem()));
+                }
+            }
+        }
+
+        final FrequencyTrie<String> trie = builder.build();
+        final Map<String, CompiledPatchCommand> compiledCommands = new HashMap<String, CompiledPatchCommand>(4096);
+        return FrequencyTrieBuilders.mapValues(trie, CompiledPatchCommand[]::new, trie.metadata().reductionSettings(),
+                patch -> compiledCommands.computeIfAbsent(patch,
+                        value -> CompiledPatchCommand.compile(value, trie.traversalDirection())));
+    }
+
+    private static List<DictionaryRow> selectRows(final List<DictionaryRow> rows, final int coveragePercent) {
+        if (coveragePercent < 1 || coveragePercent > 100) {
+            throw new IllegalArgumentException("coveragePercent must be between 1 and 100.");
+        }
+        if (coveragePercent == 100) {
+            return List.copyOf(rows);
+        }
+
+        final int selectedCount = Math.max(1, Math.round(rows.size() * coveragePercent / 100.0F));
+        final List<DictionaryRow> rankedRows = new ArrayList<DictionaryRow>(rows);
+        rankedRows.sort(Comparator.comparingLong(DictionaryRow::rank).thenComparingInt(DictionaryRow::lineNumber));
+
+        final Set<Integer> selectedLineNumbers = new HashSet<Integer>(selectedCount);
+        for (int index = 0; index < selectedCount; index++) {
+            selectedLineNumbers.add(rankedRows.get(index).lineNumber());
+        }
+
+        final List<DictionaryRow> selectedRows = new ArrayList<DictionaryRow>(selectedCount);
+        for (DictionaryRow row : rows) {
+            if (selectedLineNumbers.contains(row.lineNumber())) {
+                selectedRows.add(row);
+            }
+        }
+        return selectedRows;
+    }
+
+    private static List<DictionaryRow> readEnglishRows() throws IOException {
+        final String resourcePath = StemmerPatchTrieLoader.Language.US_UK.resourcePath();
+        final InputStream resource = StemmerPatchTrieLoader.class.getClassLoader().getResourceAsStream(resourcePath);
+        if (resource == null) {
+            throw new IllegalStateException("Missing bundled English dictionary resource " + resourcePath + ".");
+        }
+
+        final List<DictionaryRow> rows = new ArrayList<DictionaryRow>(400_000);
+        try (InputStream inputStream = resource;
+                GZIPInputStream gzipInputStream = new GZIPInputStream(inputStream);
+                InputStreamReader inputStreamReader = new InputStreamReader(gzipInputStream, StandardCharsets.UTF_8);
+                BufferedReader reader = new BufferedReader(inputStreamReader)) {
+            StemmerDictionaryParser.parse(reader, resourcePath, (stem, variants, lineNumber) -> {
+                rows.add(new DictionaryRow(lineNumber, stem, variants, rank(lineNumber, stem, variants)));
+            });
+        }
+        return rows;
+    }
+
+    private static long rank(final int lineNumber, final String stem, final String[] variants) {
+        long hash = 0xcbf29ce484222325L;
+        hash = mix(hash, lineNumber);
+        hash = mix(hash, stem);
+        for (String variant : variants) {
+            hash = mix(hash, variant);
+        }
+        return hash;
+    }
+
+    private static long mix(final long hash, final int value) {
+        long result = hash;
+        result ^= value & 0xFFL;
+        result *= 0x100000001b3L;
+        result ^= value >>> 8 & 0xFFL;
+        result *= 0x100000001b3L;
+        result ^= value >>> 16 & 0xFFL;
+        result *= 0x100000001b3L;
+        result ^= value >>> 24 & 0xFFL;
+        result *= 0x100000001b3L;
+        return result;
+    }
+
+    private static long mix(final long hash, final String value) {
+        long result = hash;
+        for (int index = 0; index < value.length(); index++) {
+            final char character = value.charAt(index);
+            result ^= character & 0xFFL;
+            result *= 0x100000001b3L;
+            result ^= character >>> 8;
+            result *= 0x100000001b3L;
+        }
+        result ^= 0xFFL;
+        result *= 0x100000001b3L;
+        return result;
+    }
+
+    /**
+     * One parsed dictionary row with deterministic selection rank.
+     *
+     * @param lineNumber source dictionary line number
+     * @param stem       canonical stem from the first column
+     * @param variants   normalized variants from following columns
+     * @param rank       deterministic selection rank
+     */
+    private record DictionaryRow(int lineNumber, String stem, String[] variants, long rank) {
+
+        /**
+         * Creates one immutable dictionary row snapshot.
+         *
+         * @param lineNumber source dictionary line number
+         * @param stem       canonical stem from the first column
+         * @param variants   normalized variants from following columns
+         * @param rank       deterministic selection rank
+         */
+        DictionaryRow {
+            Objects.requireNonNull(stem, "stem");
+            variants = variants.clone();
+        }
+
+        @Override
+        public String[] variants() {
+            return this.variants.clone();
+        }
+    }
+
+    /**
+     * Exact-root accounting result for one quality operation.
+     *
+     * @param correctMatches         exact-root matches for all tokens
+     * @param evaluatedTokens        evaluated token count
+     * @param changedCorrectMatches  exact-root matches for changed tokens
+     * @param changedEvaluatedTokens evaluated changed-token count
+     * @param rootPreservedMatches   exact-root matches for root-equal tokens
+     * @param rootEvaluatedTokens    evaluated root-equal token count
+     */
+    private record QualityCounts(int correctMatches, int evaluatedTokens, int changedCorrectMatches,
+            int changedEvaluatedTokens, int rootPreservedMatches, int rootEvaluatedTokens) {
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/EnglishStemmerComparisonBenchmark.java

@@ -32,141 +32,314 @@ package org.egothor.stemmer.benchmark;
 
 import java.io.IOException;
 import java.util.concurrent.TimeUnit;
+import java.util.logging.Logger;
+
+import org.apache.lucene.analysis.TokenStream;
+import org.apache.lucene.analysis.en.EnglishMinimalStemFilter;
+import org.apache.lucene.analysis.en.EnglishPossessiveFilter;
+import org.apache.lucene.analysis.en.KStemFilter;
+import org.apache.lucene.analysis.en.PorterStemFilter;
+import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
+import org.egothor.stemmer.benchmark.snowball.ext.englishStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.porterStemmer;
 
-import org.egothor.stemmer.FrequencyTrie;
-import org.egothor.stemmer.PatchCommandEncoder;
-import org.egothor.stemmer.ReductionMode;
-import org.egothor.stemmer.StemmerPatchTrieLoader;
 import org.openjdk.jmh.annotations.Benchmark;
 import org.openjdk.jmh.annotations.BenchmarkMode;
 import org.openjdk.jmh.annotations.Level;
 import org.openjdk.jmh.annotations.Measurement;
 import org.openjdk.jmh.annotations.Mode;
 import org.openjdk.jmh.annotations.OutputTimeUnit;
-import org.openjdk.jmh.annotations.Param;
 import org.openjdk.jmh.annotations.Scope;
 import org.openjdk.jmh.annotations.Setup;
 import org.openjdk.jmh.annotations.State;
 import org.openjdk.jmh.annotations.Warmup;
 import org.openjdk.jmh.infra.Blackhole;
-import org.tartarus.snowball.ext.englishStemmer;
-import org.tartarus.snowball.ext.porterStemmer;
+
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.StemmerDictionaryParser;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
 
 /**
- * Compares English stemming throughput across Radixor and Snowball stemmers.
+ * Compares English stemming throughput across Radixor and selected Java
+ * algorithm paths with a shared deterministic corpus.
  *
  * <p>
- * The benchmark processes the same deterministic token array with:
+ * The comparison uses one shared changed-token dictionary array for all methods:
  * </p>
  * <ul>
- * <li>Radixor using bundled
- * {@link StemmerPatchTrieLoader.Language#US_UK_PROFI}</li>
- * <li>Snowball original Porter stemmer</li>
- * <li>Snowball English stemmer, commonly referred to as Porter2</li>
+ * <li>Radixor direct dictionary lookup</li>
+ * <li>Snowball Porter</li>
+ * <li>Snowball English (Porter2)</li>
+ * <li>Lucene direct Porter API (generated copy)</li>
+ * <li>Lucene Porter, KStem, and EnglishMinimal token-filter paths</li>
+ * <li>Benchmark-only Paice/Husk Lancaster baseline</li>
  * </ul>
- *
- * <p>
- * This benchmark compares throughput on a shared workload. It does not imply
- * that the algorithms are linguistically equivalent.
- * </p>
  */
 @BenchmarkMode(Mode.AverageTime)
 @OutputTimeUnit(TimeUnit.NANOSECONDS)
-@Warmup(iterations = 3, time = 1)
-@Measurement(iterations = 5, time = 1)
+@Warmup(iterations = 3, time = 1, timeUnit = TimeUnit.SECONDS)
+@Measurement(iterations = 5, time = 1, timeUnit = TimeUnit.SECONDS)
 public class EnglishStemmerComparisonBenchmark {
 
     /**
-     * Shared benchmark data.
+     * Shared, parameterized benchmark corpus state.
      */
     @State(Scope.Benchmark)
     public static class SharedState {
 
         /**
-         * Number of generated lexical families.
-         */
-        @Param({ "1000", "5000" })
-        public int familyCount;
-
-        /**
-         * Token workload processed by all compared stemmers.
+         * Shared deterministic token corpus.
          */
         private String[] tokens;
 
         /**
-         * Radixor trie loaded from the bundled professional English dictionary.
+         * Radixor benchmark adapter for the US/UK benchmark corpus.
          */
-        private FrequencyTrie<String> radixorTrie;
+        private RadixorBenchmarkStemmer radixorStemmer;
 
         /**
-         * Initializes the shared benchmark state.
-         *
-         * @throws IOException if the bundled Radixor dictionary cannot be loaded
+         * Initializes shared corpus and trie state once per trial.
          */
         @Setup(Level.Trial)
-        public void setUp() throws IOException {
-            this.tokens = EnglishComparisonCorpus.createTokens(this.familyCount);
-            this.radixorTrie = StemmerPatchTrieLoader.load(StemmerPatchTrieLoader.Language.US_UK_PROFI, true,
-                    ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+        public void setUp() throws java.io.IOException {
+            Logger.getLogger(StemmerDictionaryParser.class.getName())
+                    .setLevel(java.util.logging.Level.OFF);
+            Logger.getLogger(StemmerDictionaryParser.class.getName()).setUseParentHandlers(false);
+            this.tokens = EnglishComparisonCorpus.createTokens();
+            this.radixorStemmer = new RadixorBenchmarkStemmer(StemmerPatchTrieLoader.loadCompiled(
+                    StemmerPatchTrieLoader.Language.US_UK, true,
+                    ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS));
         }
     }
 
     /**
-     * Per-thread reusable Snowball stemmers.
+     * Reusable direct stemmer instances.
      */
     @State(Scope.Thread)
-    public static class SnowballState {
+    public static class DirectStemmerState {
 
         /**
-         * Adapter for the original Porter stemmer.
+         * Snowball classic Porter.
          */
         private SnowballStemmerAdapter porterStemmer;
 
         /**
-         * Adapter for the Snowball English stemmer.
+         * Snowball English (Porter2) for legacy and dictionary comparison.
          */
-        private SnowballStemmerAdapter englishStemmer;
+        private SnowballStemmerAdapter englishPorterStemmer;
 
         /**
-         * Initializes reusable Snowball stemmers for the executing thread.
+         * Generated Lucene direct Porter implementation copy.
+         */
+        private LucenePorterStemmerCopied lucenePorter;
+
+        /**
+         * Benchmark-only Paice/Husk Lancaster implementation.
+         */
+        private PaiceHuskLancasterStemmer paiceHuskLancaster;
+
+        /**
+         * Apache OpenNLP Porter stemmer.
+         */
+        private opennlp.tools.stemmer.PorterStemmer openNlpPorterStemmer;
+
+        /**
+         * Initializes mutable stemmer instances reused by all benchmark calls.
          */
         @Setup(Level.Trial)
         public void setUp() {
             this.porterStemmer = new SnowballStemmerAdapter(porterStemmer::new);
-            this.englishStemmer = new SnowballStemmerAdapter(englishStemmer::new);
+            this.englishPorterStemmer = new SnowballStemmerAdapter(englishStemmer::new);
+            this.lucenePorter = new LucenePorterStemmerCopied();
+            this.paiceHuskLancaster = new PaiceHuskLancasterStemmer();
+            this.openNlpPorterStemmer = new opennlp.tools.stemmer.PorterStemmer();
+        }
+    }
+
+    /**
+     * Reusable Lucene token streams and filters.
+     */
+    @State(Scope.Thread)
+    public static class LuceneFilterState {
+
+        /**
+         * Reused Porter filter stream input.
+         */
+        private final EnglishStemmerComparisonTokenStream porterStemFilterInput;
+
+        /**
+         * Porter token filter for public API integration-path comparison.
+         */
+        private final PorterStemFilter porterStemFilter;
+
+        /**
+         * Porter filter attributes.
+         */
+        private final CharTermAttribute porterStemFilterTerm;
+
+        /**
+         * Reused KStem filter stream input.
+         */
+        private final EnglishStemmerComparisonTokenStream kStemFilterInput;
+
+        /**
+         * KStem token filter for a second Lucene English baseline.
+         */
+        private final KStemFilter kStemFilter;
+
+        /**
+         * KStem filter attributes.
+         */
+        private final CharTermAttribute kStemTerm;
+
+        /**
+         * Reused minimal stem filter stream input.
+         */
+        private final EnglishStemmerComparisonTokenStream englishMinimalStemFilterInput;
+
+        /**
+         * EnglishMinimal token filter.
+         */
+        private final EnglishMinimalStemFilter englishMinimalStemFilter;
+
+        /**
+         * EnglishMinimal filter attributes.
+         */
+        private final CharTermAttribute englishMinimalTerm;
+
+        /**
+         * Reused English possessive filter stream input.
+         */
+        private final EnglishStemmerComparisonTokenStream englishPossessiveFilterInput;
+
+        /**
+         * English possessive filter.
+         */
+        private final EnglishPossessiveFilter englishPossessiveFilter;
+
+        /**
+         * English possessive filter attributes.
+         */
+        private final CharTermAttribute englishPossessiveTerm;
+
+        /**
+         * Creates benchmark stream/filter state and attaches token attributes.
+         */
+        public LuceneFilterState() {
+            this.porterStemFilterInput = new EnglishStemmerComparisonTokenStream(new String[0]);
+            this.porterStemFilter = new PorterStemFilter(this.porterStemFilterInput);
+            this.porterStemFilterTerm = this.porterStemFilter.getAttribute(CharTermAttribute.class);
+
+            this.kStemFilterInput = new EnglishStemmerComparisonTokenStream(new String[0]);
+            this.kStemFilter = new KStemFilter(this.kStemFilterInput);
+            this.kStemTerm = this.kStemFilter.getAttribute(CharTermAttribute.class);
+
+            this.englishMinimalStemFilterInput = new EnglishStemmerComparisonTokenStream(new String[0]);
+            this.englishMinimalStemFilter = new EnglishMinimalStemFilter(this.englishMinimalStemFilterInput);
+            this.englishMinimalTerm = this.englishMinimalStemFilter.getAttribute(CharTermAttribute.class);
+
+            this.englishPossessiveFilterInput = new EnglishStemmerComparisonTokenStream(new String[0]);
+            this.englishPossessiveFilter = new EnglishPossessiveFilter(this.englishPossessiveFilterInput);
+            this.englishPossessiveTerm = this.englishPossessiveFilter.getAttribute(CharTermAttribute.class);
+        }
+
+        /**
+         * Rebinds the shared corpus and resets all streams for another measured
+         * operation.
+         *
+         * <p>
+         * The {@code String[]} to Lucene character-buffer conversion is deliberately
+         * performed every time so TokenFilter benchmarks include the cost of adapting
+         * the benchmark's canonical string corpus to Lucene's mutable token
+         * attributes.
+         * </p>
+         *
+         * @param tokens benchmark token corpus
+         */
+        void configure(final String[] tokens) throws IOException {
+            this.porterStemFilterInput.setTokens(tokens);
+            this.kStemFilterInput.setTokens(tokens);
+            this.englishMinimalStemFilterInput.setTokens(tokens);
+            this.englishPossessiveFilterInput.setTokens(tokens);
+
+            this.porterStemFilter.reset();
+            this.kStemFilter.reset();
+            this.englishMinimalStemFilter.reset();
+            this.englishPossessiveFilter.reset();
+        }
+
+        /**
+         * Reuses one mutable filter stream and returns all emitted tokens to blackhole.
+         *
+         * @param stream     benchmark token stream with configured filter
+         * @param term       token text attribute
+         * @param blackhole  sink
+         * @throws IOException on token stream failure
+         */
+        private static void consume(final TokenStream stream, final CharTermAttribute term, final Blackhole blackhole)
+                throws IOException {
+            while (stream.incrementToken()) {
+                blackhole.consume(term.toString());
+            }
+            stream.end();
+        }
+
+        /**
+         * Executes Porter filter over the shared corpus.
+         *
+         * @param blackhole sink
+         * @throws IOException if tokenization fails
+         */
+        void runPorterStemFilter(final Blackhole blackhole) throws IOException {
+            consume(this.porterStemFilter, this.porterStemFilterTerm, blackhole);
+        }
+
+        /**
+         * Executes KStem filter over the shared corpus.
+         *
+         * @param blackhole sink
+         * @throws IOException if tokenization fails
+         */
+        void runKStemFilter(final Blackhole blackhole) throws IOException {
+            consume(this.kStemFilter, this.kStemTerm, blackhole);
+        }
+
+        /**
+         * Executes English minimal filter over the shared corpus.
+         *
+         * @param blackhole sink
+         * @throws IOException if tokenization fails
+         */
+        void runEnglishMinimalStemFilter(final Blackhole blackhole) throws IOException {
+            consume(this.englishMinimalStemFilter, this.englishMinimalTerm, blackhole);
+        }
+
+        /**
+         * Executes English possessive filter over the shared corpus.
+         *
+         * @param blackhole sink
+         * @throws IOException if tokenization fails
+         */
+        void runEnglishPossessiveFilter(final Blackhole blackhole) throws IOException {
+            consume(this.englishPossessiveFilter, this.englishPossessiveTerm, blackhole);
         }
     }
 
     /**
      * Measures Radixor preferred-result stemming throughput.
      *
-     * @param sharedState shared benchmark data
-     * @param blackhole   sink preventing dead-code elimination
+     * <p>
+     * This path uses a single shared dictionary lookup and patch application.
+     * </p>
+     *
+     * @param sharedState shared corpus and trie
+     * @param blackhole   result sink
      */
     @Benchmark
     public void radixorUsUkProfiPreferredStem(final SharedState sharedState, final Blackhole blackhole) {
         final String[] tokens = sharedState.tokens;
-        final FrequencyTrie<String> trie = sharedState.radixorTrie;
-
-        for (String token : tokens) {
-            final String patch = trie.get(token);
-            final String stem = patch == null ? token : PatchCommandEncoder.apply(token, patch);
-            blackhole.consume(stem);
-        }
-    }
-
-    /**
-     * Measures Snowball original Porter stemming throughput.
-     *
-     * @param sharedState   shared benchmark data
-     * @param snowballState reusable Snowball stemmers
-     * @param blackhole     sink preventing dead-code elimination
-     */
-    @Benchmark
-    public void snowballOriginalPorter(final SharedState sharedState, final SnowballState snowballState,
-            final Blackhole blackhole) {
-        final String[] tokens = sharedState.tokens;
-        final SnowballStemmerAdapter stemmer = snowballState.porterStemmer;
+        final RadixorBenchmarkStemmer stemmer = sharedState.radixorStemmer;
 
         for (String token : tokens) {
             blackhole.consume(stemmer.stem(token));
@@ -174,25 +347,169 @@ public class EnglishStemmerComparisonBenchmark {
     }
 
     /**
-     * Measures Snowball English stemming throughput.
+     * Measures the canonical Snowball Porter stemming throughput used by the
+     * performance badge.
      *
      * <p>
-     * Snowball English is the newer English stemmer commonly referred to as
-     * Porter2.
+     * This uses Snowball classic Porter as a direct stemmer API call and includes
+     * no Lucene token stream integration overhead.
      * </p>
      *
-     * @param sharedState   shared benchmark data
-     * @param snowballState reusable Snowball stemmers
-     * @param blackhole     sink preventing dead-code elimination
+     * @param sharedState  shared corpus
+     * @param stemmerState reusable Snowball adapter state
+     * @param blackhole    result sink
      */
     @Benchmark
-    public void snowballEnglishPorter2(final SharedState sharedState, final SnowballState snowballState,
+    public void snowballOriginalPorter(final SharedState sharedState, final DirectStemmerState stemmerState,
             final Blackhole blackhole) {
         final String[] tokens = sharedState.tokens;
-        final SnowballStemmerAdapter stemmer = snowballState.englishStemmer;
+        final SnowballStemmerAdapter stemmer = stemmerState.porterStemmer;
 
         for (String token : tokens) {
             blackhole.consume(stemmer.stem(token));
         }
     }
+
+    /**
+     * Measures Snowball English (Porter2) direct API throughput.
+     *
+     * @param sharedState  shared corpus
+     * @param stemmerState reusable Snowball adapter state
+     * @param blackhole    result sink
+     */
+    @Benchmark
+    public void snowballEnglishPorter2(final SharedState sharedState, final DirectStemmerState stemmerState,
+            final Blackhole blackhole) {
+        final String[] tokens = sharedState.tokens;
+        final SnowballStemmerAdapter stemmer = stemmerState.englishPorterStemmer;
+
+        for (String token : tokens) {
+            blackhole.consume(stemmer.stem(token));
+        }
+    }
+
+    /**
+     * Measures Lucene generated Porter stemmer API throughput.
+     *
+     * <p>
+     * This path is a generated copy of Lucene&apos;s package-private PorterStemmer
+     * class, compiled into the JMH source set only.
+     * </p>
+     *
+     * @param sharedState  shared corpus
+     * @param stemmerState reusable Lucene copied API state
+     * @param blackhole    result sink
+     */
+    @Benchmark
+    public void lucenePorterStemmerCopied(final SharedState sharedState, final DirectStemmerState stemmerState,
+            final Blackhole blackhole) {
+        final String[] tokens = sharedState.tokens;
+        final LucenePorterStemmerCopied stemmer = stemmerState.lucenePorter;
+
+        for (String token : tokens) {
+            blackhole.consume(stemmer.stem(token));
+        }
+    }
+
+    /**
+     * Measures Lucene Porter token-filter integration throughput.
+     *
+     * <p>
+     * This includes stream, reusable token attributes, and filter overhead and is
+     * not equivalent to a direct API stemmer call.
+     * </p>
+     *
+     * @param sharedState shared corpus
+     * @param filterState reusable filter state
+     * @param blackhole    sink
+     * @throws IOException if token stream fails
+     */
+    @Benchmark
+    public void lucenePorterStemFilter(final SharedState sharedState, final LuceneFilterState filterState,
+            final Blackhole blackhole) throws IOException {
+        filterState.configure(sharedState.tokens);
+        filterState.runPorterStemFilter(blackhole);
+    }
+
+    /**
+     * Measures Lucene KStem integration-path throughput.
+     *
+     * @param sharedState shared corpus
+     * @param filterState reusable filter state
+     * @param blackhole    sink
+     * @throws IOException if token stream fails
+     */
+    @Benchmark
+    public void luceneKStemFilter(final SharedState sharedState, final LuceneFilterState filterState,
+            final Blackhole blackhole) throws IOException {
+        filterState.configure(sharedState.tokens);
+        filterState.runKStemFilter(blackhole);
+    }
+
+    /**
+     * Measures Lucene EnglishMinimal integration-path throughput.
+     *
+     * @param sharedState shared corpus
+     * @param filterState reusable filter state
+     * @param blackhole    sink
+     * @throws IOException if token stream fails
+     */
+    @Benchmark
+    public void luceneEnglishMinimalStemFilter(final SharedState sharedState, final LuceneFilterState filterState,
+            final Blackhole blackhole) throws IOException {
+        filterState.configure(sharedState.tokens);
+        filterState.runEnglishMinimalStemFilter(blackhole);
+    }
+
+    /**
+     * Measures benchmark-only Paice/Husk Lancaster throughput.
+     *
+     * @param sharedState  shared corpus
+     * @param stemmerState reusable Paice/Husk instance
+     * @param blackhole    sink
+     */
+    @Benchmark
+    public void paiceHuskLancaster(final SharedState sharedState, final DirectStemmerState stemmerState,
+            final Blackhole blackhole) {
+        final String[] tokens = sharedState.tokens;
+        final PaiceHuskLancasterStemmer stemmer = stemmerState.paiceHuskLancaster;
+
+        for (String token : tokens) {
+            blackhole.consume(stemmer.stem(token));
+        }
+    }
+
+    /**
+     * Measures Apache OpenNLP Porter stemming throughput.
+     *
+     * @param sharedState  shared corpus
+     * @param stemmerState reusable OpenNLP Porter instance
+     * @param blackhole    sink
+     */
+    @Benchmark
+    public void opennlpPorterStemmer(final SharedState sharedState, final DirectStemmerState stemmerState,
+            final Blackhole blackhole) {
+        final String[] tokens = sharedState.tokens;
+        final opennlp.tools.stemmer.PorterStemmer stemmer = stemmerState.openNlpPorterStemmer;
+
+        for (String token : tokens) {
+            blackhole.consume(stemmer.stem(token).toString());
+        }
+    }
+
+    /**
+     * Measures Lucene EnglishPossessiveFilter as a narrow possessive-removal
+     * baseline.
+     *
+     * @param sharedState shared corpus
+     * @param filterState reusable filter state
+     * @param blackhole   sink
+     * @throws IOException if token stream fails
+     */
+    @Benchmark
+    public void luceneEnglishPossessiveFilter(final SharedState sharedState, final LuceneFilterState filterState,
+            final Blackhole blackhole) throws IOException {
+        filterState.configure(sharedState.tokens);
+        filterState.runEnglishPossessiveFilter(blackhole);
+    }
 }

src/jmh/java/org/egothor/stemmer/benchmark/EnglishStemmerComparisonBenchmarkQuality.java

@@ -0,0 +1,258 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.IOException;
+import java.util.Objects;
+import java.util.concurrent.TimeUnit;
+
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+import org.egothor.stemmer.benchmark.snowball.ext.porterStemmer;
+import org.openjdk.jmh.annotations.AuxCounters;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Fork;
+import org.openjdk.jmh.annotations.Level;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.Setup;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.Warmup;
+import org.openjdk.jmh.infra.Blackhole;
+
+/**
+ * Emits exact-root agreement metrics for the canonical English badge pair.
+ *
+ * <p>
+ * This class is deliberately named so the existing focused include pattern for
+ * English stemmer comparison benchmarks includes it. The benchmark methods are
+ * separate from throughput methods so equality checks do not contaminate timing
+ * scores.
+ * </p>
+ */
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.NANOSECONDS)
+@Warmup(iterations = 0)
+@Measurement(iterations = 1, time = 1, timeUnit = TimeUnit.MILLISECONDS)
+@Fork(0)
+public class EnglishStemmerComparisonBenchmarkQuality {
+
+    /**
+     * Shared English quality corpus and stemmer state.
+     */
+    @State(Scope.Benchmark)
+    public static class SharedState {
+
+        /**
+         * Complete English resource-derived corpus.
+         */
+        private LanguageBenchmarkCorpus.Corpus corpus;
+
+        /**
+         * Compiled Radixor English trie.
+         */
+        private RadixorBenchmarkStemmer radixorStemmer;
+
+        /**
+         * Reusable Snowball Porter adapter.
+         */
+        private SnowballStemmerAdapter porterStemmer;
+
+        /**
+         * Initializes quality resources.
+         *
+         * @throws IOException if corpus or trie loading fails
+         */
+        @Setup(Level.Trial)
+        public void setUp() throws IOException {
+            this.corpus = LanguageBenchmarkCorpus.createFullCorpus(StemmerPatchTrieLoader.Language.US_UK);
+            this.radixorStemmer = new RadixorBenchmarkStemmer(StemmerPatchTrieLoader.loadCompiled(
+                    StemmerPatchTrieLoader.Language.US_UK, true,
+                    ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS));
+            this.porterStemmer = new SnowballStemmerAdapter(porterStemmer::new);
+        }
+    }
+
+    /**
+     * JMH auxiliary counters for exact-root agreement.
+     */
+    @State(Scope.Thread)
+    @AuxCounters(AuxCounters.Type.EVENTS)
+    public static class AccuracyCounters {
+
+        /**
+         * Number of exact-root matches.
+         */
+        public long correctMatches;
+
+        /**
+         * Number of evaluated tokens.
+         */
+        public long evaluatedTokens;
+
+        /**
+         * Number of exact-root matches where the input token differs from the
+         * expected root.
+         */
+        public long changedCorrectMatches;
+
+        /**
+         * Number of evaluated tokens where the input token differs from the expected
+         * root.
+         */
+        public long changedEvaluatedTokens;
+
+        /**
+         * Number of exact-root matches where the input token is already the expected
+         * root.
+         */
+        public long rootPreservedMatches;
+
+        /**
+         * Number of evaluated tokens where the input token is already the expected
+         * root.
+         */
+        public long rootEvaluatedTokens;
+
+        /**
+         * Resets counters before each measured iteration.
+         */
+        @Setup(Level.Iteration)
+        public void reset() {
+            this.correctMatches = 0L;
+            this.evaluatedTokens = 0L;
+            this.changedCorrectMatches = 0L;
+            this.changedEvaluatedTokens = 0L;
+            this.rootPreservedMatches = 0L;
+            this.rootEvaluatedTokens = 0L;
+        }
+    }
+
+    /**
+     * Evaluates exact-root agreement for the canonical Radixor badge method.
+     *
+     * @param sharedState shared English quality state
+     * @param counters JMH auxiliary counters
+     * @param blackhole result sink
+     * @return exact-root match count
+     */
+    @Benchmark
+    public int radixorUsUkProfiPreferredStemAccuracy(final SharedState sharedState,
+            final AccuracyCounters counters, final Blackhole blackhole) {
+        return evaluate(sharedState.corpus, sharedState.radixorStemmer::stem, counters, blackhole);
+    }
+
+    /**
+     * Evaluates exact-root agreement for the canonical Snowball Porter badge
+     * method.
+     *
+     * @param sharedState shared English quality state
+     * @param counters JMH auxiliary counters
+     * @param blackhole result sink
+     * @return exact-root match count
+     */
+    @Benchmark
+    public int snowballOriginalPorterAccuracy(final SharedState sharedState,
+            final AccuracyCounters counters, final Blackhole blackhole) {
+        return evaluate(sharedState.corpus, sharedState.porterStemmer::stem, counters, blackhole);
+    }
+
+    /**
+     * Evaluates one stemmer against the expected roots.
+     *
+     * @param corpus token/root corpus
+     * @param stemmer stemmer under evaluation
+     * @param counters JMH auxiliary counters
+     * @param blackhole result sink
+     * @return exact-root match count
+     */
+    private static int evaluate(final LanguageBenchmarkCorpus.Corpus corpus, final Stemmer stemmer,
+            final AccuracyCounters counters, final Blackhole blackhole) {
+        Objects.requireNonNull(corpus, "corpus");
+        Objects.requireNonNull(stemmer, "stemmer");
+
+        int correct = 0;
+        int changedCorrect = 0;
+        int changedEvaluated = 0;
+        int rootPreserved = 0;
+        int rootEvaluated = 0;
+        final String[] tokens = corpus.tokens();
+        final String[] expectedRoots = corpus.expectedRoots();
+        for (int index = 0; index < tokens.length; index++) {
+            final String token = tokens[index];
+            final String expectedRoot = expectedRoots[index];
+            final String actual = stemmer.stem(token);
+            blackhole.consume(actual);
+            final boolean exact = Objects.equals(expectedRoot, actual);
+            if (exact) {
+                correct++;
+            }
+            if (Objects.equals(token, expectedRoot)) {
+                rootEvaluated++;
+                if (exact) {
+                    rootPreserved++;
+                }
+            } else {
+                changedEvaluated++;
+                if (exact) {
+                    changedCorrect++;
+                }
+            }
+        }
+
+        counters.correctMatches += correct;
+        counters.evaluatedTokens += tokens.length;
+        counters.changedCorrectMatches += changedCorrect;
+        counters.changedEvaluatedTokens += changedEvaluated;
+        counters.rootPreservedMatches += rootPreserved;
+        counters.rootEvaluatedTokens += rootEvaluated;
+        return correct;
+    }
+
+    /**
+     * Direct stemmer function.
+     */
+    @FunctionalInterface
+    private interface Stemmer {
+
+        /**
+         * Produces one stem.
+         *
+         * @param token input token
+         * @return produced stem
+         */
+        String stem(String token);
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/EnglishStemmerComparisonTokenStream.java

@@ -0,0 +1,145 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ * 
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ * 
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ * 
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.IOException;
+
+import org.apache.lucene.analysis.TokenStream;
+import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
+import org.apache.lucene.analysis.tokenattributes.OffsetAttribute;
+import org.apache.lucene.analysis.tokenattributes.PositionIncrementAttribute;
+
+/**
+ * Reusable token stream driven by a deterministic token corpus.
+ *
+ * <p>
+ * The stream emits each token from a shared array and supports repeated
+ * {@link #reset()} + {@link #incrementToken()} cycles without per-token
+ * object allocation.
+ * </p>
+ */
+final class EnglishStemmerComparisonTokenStream extends TokenStream {
+
+    /**
+     * Current token text.
+     */
+    private final CharTermAttribute charTermAttribute;
+
+    /**
+     * Token offsets for benchmark stream compliance.
+     */
+    private final OffsetAttribute offsetAttribute;
+
+    /**
+     * Position increment attribute for benchmark stream compliance.
+     */
+    private final PositionIncrementAttribute positionIncrementAttribute;
+
+    /**
+     * Reusable token source.
+     */
+    private final BenchmarkTokenSequence tokenSequence;
+
+    /**
+     * Creates a deterministic token stream for benchmark reuse.
+     *
+     * @param tokens tokens emitted by the stream
+     */
+    EnglishStemmerComparisonTokenStream(final String[] tokens) {
+        this.tokenSequence = new BenchmarkTokenSequence(tokens);
+        this.charTermAttribute = addAttribute(CharTermAttribute.class);
+        this.offsetAttribute = addAttribute(OffsetAttribute.class);
+        this.positionIncrementAttribute = addAttribute(PositionIncrementAttribute.class);
+    }
+
+    /**
+     * Replaces the token corpus for this stream.
+     *
+     * @param tokens new token corpus
+     */
+    void setTokens(final String[] tokens) {
+        this.tokenSequence.setTokens(tokens);
+    }
+
+    /**
+     * Returns whether the stream is drained and ready to be exhausted.
+     *
+     * @return true if all configured tokens were consumed
+     */
+    boolean isDrained() {
+        return !this.tokenSequence.hasNext();
+    }
+
+    /**
+     * {@inheritDoc}
+     */
+    @Override
+    public boolean incrementToken() throws IOException {
+        if (!this.tokenSequence.advance()) {
+            return false;
+        }
+
+        clearAttributes();
+        final char[] token = this.tokenSequence.currentToken();
+        this.charTermAttribute.copyBuffer(token, 0, token.length);
+        this.positionIncrementAttribute.setPositionIncrement(1);
+        this.offsetAttribute.setOffset(this.tokenSequence.currentStartOffset(), this.tokenSequence.currentEndOffset());
+        return true;
+    }
+
+    /**
+     * {@inheritDoc}
+     */
+    @Override
+    public void reset() throws IOException {
+        super.reset();
+        this.tokenSequence.reset();
+    }
+
+    /**
+     * {@inheritDoc}
+     */
+    @Override
+    public void end() throws IOException {
+        super.end();
+        final int endOffset = this.tokenSequence.endOffset();
+        this.offsetAttribute.setOffset(endOffset, endOffset);
+    }
+
+    /**
+     * {@inheritDoc}
+     */
+    @Override
+    public void close() throws IOException {
+        super.close();
+        this.charTermAttribute.setEmpty();
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/FrequencyTrieLookupBenchmark.java

@@ -31,11 +31,13 @@
 package org.egothor.stemmer.benchmark;
 
 import java.io.IOException;
+import java.util.List;
 import java.util.concurrent.TimeUnit;
 import org.egothor.stemmer.FrequencyTrie;
 import org.egothor.stemmer.PatchCommandEncoder;
 import org.egothor.stemmer.ReductionMode;
 import org.egothor.stemmer.ReductionSettings;
+import org.egothor.stemmer.ValueCount;
 import org.openjdk.jmh.annotations.Benchmark;
 import org.openjdk.jmh.annotations.BenchmarkMode;
 import org.openjdk.jmh.annotations.Level;
@@ -63,6 +65,7 @@ import org.openjdk.jmh.infra.Blackhole;
 @OutputTimeUnit(TimeUnit.NANOSECONDS)
 @Warmup(iterations = 3, time = 1)
 @Measurement(iterations = 5, time = 1)
+@SuppressWarnings("deprecation")
 public class FrequencyTrieLookupBenchmark {
 
     /**
@@ -97,12 +100,45 @@ public class FrequencyTrieLookupBenchmark {
          */
         private String[] lookupKeys;
 
+        /**
+         * Lookup keys as normalized caller-owned character storage.
+         */
+        private char[][] lookupKeyCharacters;
+
         /**
          * Keys that are known to return multiple patch candidates from
          * {@code getAll()}.
          */
         private String[] ambiguousLookupKeys;
 
+        /**
+         * Ambiguous lookup keys as normalized caller-owned character storage.
+         */
+        private char[][] ambiguousLookupKeyCharacters;
+
+        /**
+         * Preferred patches aligned with {@link #lookupKeys}.
+         */
+        private String[] preferredPatches;
+
+        /**
+         * Reusable output buffer for patch application benchmarks.
+         */
+        private char[] outputBuffer;
+
+        /**
+         * Mutable field consumed by visitor sinks.
+         */
+        private int visitorAccumulator;
+
+        /**
+         * Sink used by visitor lookup benchmarks without per-invocation allocation.
+         */
+        private final FrequencyTrie.EntrySink<String> visitorSink = (value, count, rank) -> {
+            this.visitorAccumulator += value.length() + count + rank;
+            return true;
+        };
+
         /**
          * Initializes the benchmark state.
          *
@@ -116,6 +152,23 @@ public class FrequencyTrieLookupBenchmark {
             this.trie = BenchmarkCorpusSupport.compilePatchTrie(corpus.dictionaryText(), settings, true);
             this.lookupKeys = corpus.lookupKeys();
             this.ambiguousLookupKeys = corpus.ambiguousLookupKeys();
+            this.lookupKeyCharacters = toCharArrays(this.lookupKeys);
+            this.ambiguousLookupKeyCharacters = toCharArrays(this.ambiguousLookupKeys);
+            this.preferredPatches = new String[this.lookupKeys.length];
+            int maxKeyLength = 0;
+            for (int index = 0; index < this.lookupKeys.length; index++) {
+                this.preferredPatches[index] = this.trie.get(this.lookupKeys[index]);
+                maxKeyLength = Math.max(maxKeyLength, this.lookupKeys[index].length());
+            }
+            this.outputBuffer = new char[maxKeyLength + 32];
+        }
+
+        private static char[][] toCharArrays(final String[] values) {
+            final char[][] characters = new char[values.length][];
+            for (int index = 0; index < values.length; index++) {
+                characters[index] = values[index].toCharArray();
+            }
+            return characters;
         }
     }
 
@@ -155,6 +208,61 @@ public class FrequencyTrieLookupBenchmark {
         }
     }
 
+    /**
+     * Measures retrieval of all patch candidates through caller-owned normalized
+     * character storage and a visitor sink.
+     *
+     * @param state prepared lookup state
+     * @param blackhole sink preventing dead-code elimination
+     */
+    @Benchmark
+    public void lookupAllPatchesWithNormalizedCharVisitor(final LookupState state, final Blackhole blackhole) {
+        final char[][] keys = state.ambiguousLookupKeyCharacters;
+        for (char[] key : keys) {
+            final int count = state.trie.getAllNormalized(key, 0, key.length, state.visitorSink, Integer.MAX_VALUE);
+            if (count < 2) {
+                throw new IllegalStateException("Expected multiple patches for benchmark key.");
+            }
+        }
+        blackhole.consume(state.visitorAccumulator);
+    }
+
+    /**
+     * Measures counted candidate retrieval through the allocating entry API.
+     *
+     * @param state prepared lookup state
+     * @param blackhole sink preventing dead-code elimination
+     */
+    @Benchmark
+    public void lookupPatchEntries(final LookupState state, final Blackhole blackhole) {
+        final String[] keys = state.ambiguousLookupKeys;
+        for (String key : keys) {
+            final List<ValueCount<String>> entries = state.trie.getEntries(key);
+            if (entries.size() < 2) {
+                throw new IllegalStateException("Expected multiple entries for key " + key + '.');
+            }
+            blackhole.consume(entries);
+        }
+    }
+
+    /**
+     * Measures counted candidate retrieval through the visitor API.
+     *
+     * @param state prepared lookup state
+     * @param blackhole sink preventing dead-code elimination
+     */
+    @Benchmark
+    public void lookupPatchEntriesWithVisitor(final LookupState state, final Blackhole blackhole) {
+        final char[][] keys = state.ambiguousLookupKeyCharacters;
+        for (char[] key : keys) {
+            final int count = state.trie.getAllNormalized(key, 0, key.length, state.visitorSink, Integer.MAX_VALUE);
+            if (count < 2) {
+                throw new IllegalStateException("Expected multiple entries for benchmark key.");
+            }
+        }
+        blackhole.consume(state.visitorAccumulator);
+    }
+
     /**
      * Measures end-to-end preferred stemming from lookup plus patch application.
      *
@@ -170,6 +278,28 @@ public class FrequencyTrieLookupBenchmark {
         }
     }
 
+    /**
+     * Measures patch application into caller-owned output storage.
+     *
+     * @param state prepared lookup state
+     * @param blackhole sink preventing dead-code elimination
+     */
+    @Benchmark
+    public void applyPreferredPatchToBuffer(final LookupState state, final Blackhole blackhole) {
+        final String[] keys = state.lookupKeys;
+        final String[] patches = state.preferredPatches;
+        final char[] output = state.outputBuffer;
+        for (int index = 0; index < keys.length; index++) {
+            final int length = PatchCommandEncoder.applyTo(keys[index], patches[index],
+                    state.trie.traversalDirection(), output, 0, output.length);
+            if (length == PatchCommandEncoder.APPLY_INSUFFICIENT_CAPACITY) {
+                throw new IllegalStateException("Output buffer too small for key " + keys[index] + '.');
+            }
+            blackhole.consume(length);
+            blackhole.consume(output[0]);
+        }
+    }
+
     /**
      * Measures end-to-end full candidate stemming from {@code getAll()} plus
      * patch application.

src/jmh/java/org/egothor/stemmer/benchmark/GermanGoldstandardStemmerComparisonBenchmark.java

@@ -0,0 +1,579 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.BufferedReader;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.InputStreamReader;
+import java.nio.charset.StandardCharsets;
+import java.util.ArrayList;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.Objects;
+import java.util.function.Function;
+
+import org.apache.lucene.analysis.LowerCaseFilter;
+import org.apache.lucene.analysis.TokenStream;
+import org.apache.lucene.analysis.de.GermanLightStemFilter;
+import org.apache.lucene.analysis.de.GermanMinimalStemFilter;
+import org.apache.lucene.analysis.de.GermanNormalizationFilter;
+import org.apache.lucene.analysis.de.GermanStemFilter;
+import org.apache.lucene.analysis.snowball.SnowballFilter;
+import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
+import org.apache.lucene.analysis.tokenattributes.PositionIncrementAttribute;
+import org.openjdk.jmh.annotations.AuxCounters;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Fork;
+import org.openjdk.jmh.annotations.Level;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+import org.openjdk.jmh.annotations.Param;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.Setup;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.Warmup;
+import org.openjdk.jmh.infra.Blackhole;
+
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+/**
+ * German-only stemmer comparison on CISTEM gold standards.
+ *
+ * <p>
+ * Each benchmark operation is fed by one cluster file. The same candidate set is
+ * evaluated twice, once per file, to produce one precision/recall/f-measure
+ * table for each gold standard.
+ * </p>
+ */
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(java.util.concurrent.TimeUnit.NANOSECONDS)
+@Warmup(iterations = 3, time = 1, timeUnit = java.util.concurrent.TimeUnit.SECONDS)
+@Measurement(iterations = 5, time = 1, timeUnit = java.util.concurrent.TimeUnit.SECONDS)
+@Fork(1)
+public class GermanGoldstandardStemmerComparisonBenchmark {
+
+    /**
+     * Shared German benchmark state for one dataset and one candidate.
+     */
+    @State(Scope.Benchmark)
+    public static class SharedState {
+
+        /**
+         * Gold standard dataset.
+         */
+        @Param({"goldstandard1.txt", "goldstandard2.txt"})
+        public String goldStandardFileName;
+
+        /**
+         * Candidate stemmer.
+         */
+        @Param({
+                "GERMAN_RADIXOR",
+                "GERMAN_LUCENE_GERMAN_STEM_FILTER",
+                "GERMAN_LUCENE_GERMAN_LIGHT_STEM_FILTER",
+                "GERMAN_LUCENE_GERMAN_MINIMAL_STEM_FILTER",
+                "GERMAN_CISTEM",
+                "SNOWBALL_GERMAN_DIRECT",
+                "SNOWBALL_GERMAN_LUCENE_FILTER"
+        })
+        public String candidateName;
+
+        /**
+         * Parsed gold standard corpus.
+         */
+        private GermanGoldstandardCorpus corpus;
+
+        /**
+         * Gold standard words flattened by cluster order.
+         */
+        private String[] allTokens;
+
+        /**
+         * Candidate evaluator.
+         */
+        private GoldstandardStemmer stemmer;
+
+        /**
+         * Initializes one candidate on one gold standard corpus.
+         *
+         * @throws IOException when the corpus cannot be loaded
+         */
+        @Setup(Level.Trial)
+        public void setUp() throws IOException {
+            this.corpus = loadCorpus(this.goldStandardFileName);
+            this.allTokens = flattenCorpusTokens(this.corpus);
+            this.stemmer = GermanCandidate.valueOf(this.candidateName).createEvaluator();
+        }
+    }
+
+    /**
+     * JMH auxiliary counters for CISTEM-style cluster accounting.
+     */
+    @State(Scope.Thread)
+    @AuxCounters(AuxCounters.Type.EVENTS)
+    public static class GoldstandardQualityCounters {
+
+        /**
+         * True positives across clusters.
+         */
+        public long truePositives;
+
+        /**
+         * False positives across clusters.
+         */
+        public long falsePositives;
+
+        /**
+         * False negatives across clusters.
+         */
+        public long falseNegatives;
+
+        /**
+         * Evaluated clusters.
+         */
+        public long evaluatedClusters;
+
+        /**
+         * Evaluated tokens.
+         */
+        public long evaluatedTokens;
+
+        /**
+         * Resets counters before each measured iteration.
+         */
+        @Setup(Level.Iteration)
+        public void reset() {
+            this.truePositives = 0L;
+            this.falsePositives = 0L;
+            this.falseNegatives = 0L;
+            this.evaluatedClusters = 0L;
+            this.evaluatedTokens = 0L;
+        }
+    }
+
+    /**
+     * Evaluates CISTEM-style precision, recall, and F1-relevant counts.
+     *
+     * @param state shared benchmark state
+     * @param counters quality counters
+     * @param blackhole result sink
+     * @return evaluated token count for this operation
+     * @throws IOException if token filtering cannot run
+     */
+    @Benchmark
+    @Warmup(iterations = 0)
+    @Measurement(iterations = 1, time = 1, timeUnit = java.util.concurrent.TimeUnit.MILLISECONDS)
+    @Fork(0)
+    public long cistemStyleQuality(final SharedState state, final GoldstandardQualityCounters counters,
+            final Blackhole blackhole) throws IOException {
+        final GoldstandardResult result = evaluateCistemStyle(state.corpus, state.allTokens, state.stemmer, blackhole);
+        counters.truePositives += result.truePositives();
+        counters.falsePositives += result.falsePositives();
+        counters.falseNegatives += result.falseNegatives();
+        counters.evaluatedClusters += result.evaluatedClusters();
+        counters.evaluatedTokens += result.evaluatedTokens();
+        return result.evaluatedTokens();
+    }
+
+    /**
+     * Benchmarks candidate throughput over the selected gold standard.
+     *
+     * @param state shared benchmark state
+     * @param blackhole result sink
+     * @throws IOException if token filtering cannot run
+     */
+    @Benchmark
+    public void cistemStyleSpeed(final SharedState state, final Blackhole blackhole) throws IOException {
+        state.stemmer.stem(state.allTokens, blackhole);
+    }
+
+    /**
+     * Named German candidates used for the CISTEM gold-standard comparison.
+     */
+    private enum GermanCandidate {
+        GERMAN_RADIXOR,
+        GERMAN_LUCENE_GERMAN_STEM_FILTER,
+        GERMAN_LUCENE_GERMAN_LIGHT_STEM_FILTER,
+        GERMAN_LUCENE_GERMAN_MINIMAL_STEM_FILTER,
+        GERMAN_CISTEM,
+        SNOWBALL_GERMAN_DIRECT,
+        SNOWBALL_GERMAN_LUCENE_FILTER;
+
+        /**
+         * Creates a candidate evaluator.
+         *
+         * @return stemmer evaluator
+         * @throws IOException if trie resources cannot be loaded
+         */
+        GoldstandardStemmer createEvaluator() throws IOException {
+            return switch (this) {
+                case GERMAN_RADIXOR -> direct(createGermanRadixorStemmer());
+                case GERMAN_LUCENE_GERMAN_STEM_FILTER ->
+                    tokenFilter(input -> new GermanStemFilter(lowercase(input)));
+                case GERMAN_LUCENE_GERMAN_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new GermanLightStemFilter(germanNormalize(input)));
+                case GERMAN_LUCENE_GERMAN_MINIMAL_STEM_FILTER ->
+                    tokenFilter(input -> new GermanMinimalStemFilter(germanNormalize(input)));
+                case GERMAN_CISTEM -> direct(Cistem::stem);
+                case SNOWBALL_GERMAN_DIRECT -> direct(SnowballLanguageCase.GERMAN.createDirectStemmer()::stem);
+                case SNOWBALL_GERMAN_LUCENE_FILTER ->
+                    tokenFilter(input -> new SnowballFilter(new LowerCaseFilter(input),
+                            SnowballLanguageCase.GERMAN.luceneSnowballName()));
+            };
+        }
+    }
+
+    /**
+     * Evaluates one full corpus through CISTEM-style cluster scoring.
+     *
+     * <p>
+     * For each cluster, the most frequent predicted stem is considered the
+     * cluster main stem. TP are cluster words mapped to this stem, FN are
+     * words mapped elsewhere inside the same cluster, and FP are words from
+     * other clusters mapped to the same main stem.
+     * </p>
+     *
+     * @param corpus parsed gold standard corpus
+     * @param allTokens flattened token sequence
+     * @param stemmer candidate stemmer
+     * @param blackhole result sink
+     * @return aggregated TP/FP/FN counters and token metrics
+     * @throws IOException when token filtering cannot run
+     */
+    private static GoldstandardResult evaluateCistemStyle(final GermanGoldstandardCorpus corpus,
+            final String[] allTokens, final GoldstandardStemmer stemmer, final Blackhole blackhole) throws IOException {
+        final String[] predicted = stemmer.stem(allTokens, blackhole);
+        final Map<String, Integer> globalPredictions = new LinkedHashMap<>();
+        for (int index = 0; index < allTokens.length; index++) {
+            final String prediction = normalizePrediction(predicted[index], allTokens[index]);
+            globalPredictions.put(prediction, globalPredictions.getOrDefault(prediction, 0) + 1);
+        }
+
+        long truePositives = 0L;
+        long falsePositives = 0L;
+        long falseNegatives = 0L;
+        int tokenOffset = 0;
+        for (final String[] cluster : corpus.clusters()) {
+            if (cluster.length == 0) {
+                continue;
+            }
+
+            final Map<String, Integer> localPredictions = new LinkedHashMap<>();
+            for (int index = 0; index < cluster.length; index++) {
+                final int tokenIndex = tokenOffset + index;
+                final String word = allTokens[tokenIndex];
+                final String prediction = normalizePrediction(predicted[tokenIndex], word);
+                localPredictions.put(prediction, localPredictions.getOrDefault(prediction, 0) + 1);
+            }
+
+            final String mainStem = mostFrequent(localPredictions);
+            final int predictedAsMain = localPredictions.get(mainStem);
+            final int clusterSize = cluster.length;
+            final int falseNegative = clusterSize - predictedAsMain;
+            final int falsePositive = globalPredictions.get(mainStem) - predictedAsMain;
+
+            truePositives += predictedAsMain;
+            falseNegatives += falseNegative;
+            falsePositives += falsePositive;
+            tokenOffset += clusterSize;
+        }
+
+        return new GoldstandardResult(truePositives, falsePositives, falseNegatives, corpus.clusters().length,
+                allTokens.length);
+    }
+
+    /**
+     * Creates a direct evaluator.
+     *
+     * @param stemmer direct word stemmer
+     * @return evaluator
+     */
+    private static GoldstandardStemmer direct(final Stemmer stemmer) {
+        Objects.requireNonNull(stemmer, "stemmer");
+        return (tokens, blackhole) -> {
+            final String[] outputs = new String[tokens.length];
+            for (int index = 0; index < tokens.length; index++) {
+                final String output = stemmer.stem(tokens[index]);
+                outputs[index] = output;
+                blackhole.consume(output);
+            }
+            return outputs;
+        };
+    }
+
+    /**
+     * Creates a TokenFilter evaluator.
+     *
+     * @param factory filter stream factory
+     * @return evaluator
+     */
+    private static GoldstandardStemmer tokenFilter(final Function<TokenStream, TokenStream> factory) {
+        Objects.requireNonNull(factory, "factory");
+        return (tokens, blackhole) -> firstTokenFilterOutputs(tokens, factory, blackhole);
+    }
+
+    /**
+     * Loads and parses one gold standard file from generated JMH resources.
+     *
+     * @param resourceName gold standard file name
+     * @return parsed corpus
+     * @throws IOException if reading fails
+     */
+    private static GermanGoldstandardCorpus loadCorpus(final String resourceName) throws IOException {
+        final ClassLoader classLoader = GermanGoldstandardStemmerComparisonBenchmark.class.getClassLoader();
+        final InputStream resourceStream = classLoader.getResourceAsStream(resourceName);
+        if (resourceStream == null) {
+            throw new IllegalStateException("Missing generated CISTEM gold standard resource: " + resourceName
+                    + ". Run the Gradle JMH resource preparation task to download benchmark-only inputs.");
+        }
+        try (InputStream input = resourceStream) {
+            return parseCorpus(new BufferedReader(new InputStreamReader(input, StandardCharsets.UTF_8)));
+        }
+    }
+
+    /**
+     * Parses CISTEM gold standard format into clustered candidates.
+     *
+     * @param reader UTF-8 reader
+     * @return parsed corpus
+     * @throws IOException if input cannot be read
+     */
+    private static GermanGoldstandardCorpus parseCorpus(final BufferedReader reader) throws IOException {
+        final List<String[]> clusters = new ArrayList<>();
+        String line = reader.readLine();
+        while (line != null) {
+            final String trimmed = line.trim();
+            if (!trimmed.isEmpty()) {
+                final String[] words = trimmed.split("\\s+");
+                if (words.length > 0) {
+                    clusters.add(words);
+                }
+            }
+            line = reader.readLine();
+        }
+        return new GermanGoldstandardCorpus(clusters.toArray(String[][]::new));
+    }
+
+    /**
+     * Flattens the corpus in deterministic cluster order.
+     *
+     * @param corpus corpus to flatten
+     * @return flattened token array
+     */
+    private static String[] flattenCorpusTokens(final GermanGoldstandardCorpus corpus) {
+        int total = 0;
+        for (final String[] cluster : corpus.clusters()) {
+            total += cluster.length;
+        }
+        final String[] tokens = new String[total];
+        int index = 0;
+        for (final String[] cluster : corpus.clusters()) {
+            System.arraycopy(cluster, 0, tokens, index, cluster.length);
+            index += cluster.length;
+        }
+        return tokens;
+    }
+
+    /**
+     * Returns the most frequent key; insertion order is preserved on ties.
+     *
+     * @param frequencies predicted stem frequencies
+     * @return most frequent stem
+     */
+    private static String mostFrequent(final Map<String, Integer> frequencies) {
+        String best = null;
+        int bestCount = -1;
+        for (final Map.Entry<String, Integer> entry : frequencies.entrySet()) {
+            if (entry.getValue() > bestCount) {
+                best = entry.getKey();
+                bestCount = entry.getValue();
+            }
+        }
+        return best;
+    }
+
+    /**
+     * Normalizes a null/empty prediction using the input token as fallback.
+     *
+     * @param prediction stemmed token
+     * @param fallback fallback token
+     * @return safe prediction
+     */
+    private static String normalizePrediction(final String prediction, final String fallback) {
+        if (prediction == null || prediction.isEmpty()) {
+            return fallback;
+        }
+        return prediction;
+    }
+
+    /**
+     * Applies one TokenFilter to all input tokens and returns the first emitted term
+     * for each input token.
+     *
+     * @param tokens input token corpus
+     * @param factory TokenFilter factory
+     * @param blackhole result sink
+     * @return first emitted term per input token
+     * @throws IOException if token streaming fails
+     */
+    private static String[] firstTokenFilterOutputs(final String[] tokens, final Function<TokenStream, TokenStream> factory,
+            final Blackhole blackhole) throws IOException {
+        final String[] outputs = new String[tokens.length];
+        final BenchmarkTokenStream input = new BenchmarkTokenStream(tokens);
+        final TokenStream output = factory.apply(input);
+        final CharTermAttribute termAttribute = output.addAttribute(CharTermAttribute.class);
+        final PositionIncrementAttribute positionAttribute = output.addAttribute(PositionIncrementAttribute.class);
+
+        int inputIndex = -1;
+        boolean recordedForPosition = false;
+        output.reset();
+        while (output.incrementToken()) {
+            final int positionIncrement = positionAttribute.getPositionIncrement();
+            if (positionIncrement > 0) {
+                inputIndex += positionIncrement;
+                recordedForPosition = false;
+            }
+            if (inputIndex >= 0 && inputIndex < outputs.length && !recordedForPosition) {
+                outputs[inputIndex] = termAttribute.toString();
+                blackhole.consume(termAttribute);
+                recordedForPosition = true;
+            }
+        }
+        output.end();
+        output.close();
+
+        for (int index = 0; index < outputs.length; index++) {
+            if (outputs[index] == null) {
+                outputs[index] = tokens[index];
+            }
+        }
+        return outputs;
+    }
+
+    /**
+     * Creates a direct Radixor evaluator using the contracted dictionary trie.
+     *
+     * @return direct Radixor stemmer
+     * @throws IOException if the trie cannot be loaded
+     */
+    private static Stemmer createGermanRadixorStemmer() throws IOException {
+        return new RadixorBenchmarkStemmer(StemmerPatchTrieLoader.loadCompiled(
+                StemmerPatchTrieLoader.Language.DE_DE, true,
+                ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS))::stem;
+    }
+
+    /**
+     * Adds Lucene lower-case normalization.
+     *
+     * @param input token stream
+     * @return normalized token stream
+     */
+    private static TokenStream lowercase(final TokenStream input) {
+        return new LowerCaseFilter(input);
+    }
+
+    /**
+     * Adds Lucene German normalization for light and minimal filters.
+     *
+     * @param input token stream
+     * @return normalized token stream
+     */
+    private static TokenStream germanNormalize(final TokenStream input) {
+        return new GermanNormalizationFilter(lowercase(input));
+    }
+
+    /**
+     * Direct or filter stemmer adapter used by this benchmark.
+     */
+    @FunctionalInterface
+    private interface GoldstandardStemmer {
+
+        /**
+         * Runs one complete token list.
+         *
+         * @param tokens input tokens
+         * @param blackhole result sink
+         * @return per-token outputs
+         * @throws IOException if filter processing fails
+         */
+        String[] stem(String[] tokens, Blackhole blackhole) throws IOException;
+    }
+
+    /**
+     * Deterministic direct word stem function.
+     */
+    @FunctionalInterface
+    private interface Stemmer {
+
+        /**
+         * Stems one token.
+         *
+         * @param token input token
+         * @return stemmed token
+         */
+        String stem(String token);
+    }
+
+    /**
+     * Immutable parsed CISTEM gold standard corpus.
+     */
+    private static final class GermanGoldstandardCorpus {
+
+        private final String[][] clusters;
+
+        GermanGoldstandardCorpus(final String[][] clusters) {
+            this.clusters = clusters;
+        }
+
+        String[][] clusters() {
+            return this.clusters;
+        }
+    }
+
+    /**
+     * Aggregated quality result for one benchmark operation.
+     *
+     * @param truePositives true positives
+     * @param falsePositives false positives
+     * @param falseNegatives false negatives
+     * @param evaluatedClusters evaluated clusters
+     * @param evaluatedTokens evaluated tokens
+     */
+    private record GoldstandardResult(long truePositives, long falsePositives, long falseNegatives,
+            long evaluatedClusters, long evaluatedTokens) {
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/HunspellStemmerComparisonBenchmark.java

@@ -0,0 +1,315 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.IOException;
+import java.io.InputStream;
+import java.text.ParseException;
+import java.util.List;
+import java.util.Locale;
+import java.util.concurrent.TimeUnit;
+
+import org.apache.lucene.analysis.LowerCaseFilter;
+import org.apache.lucene.analysis.TokenStream;
+import org.apache.lucene.analysis.hunspell.Dictionary;
+import org.apache.lucene.analysis.hunspell.HunspellStemFilter;
+import org.apache.lucene.analysis.hunspell.SortingStrategy;
+import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Level;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+import org.openjdk.jmh.annotations.Param;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.Setup;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.Warmup;
+import org.openjdk.jmh.infra.Blackhole;
+
+/**
+ * Compares Radixor with Lucene's Hunspell integration over selected
+ * benchmark-only Hunspell dictionaries.
+ */
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.NANOSECONDS)
+@Warmup(iterations = 3, time = 1, timeUnit = TimeUnit.SECONDS)
+@Measurement(iterations = 5, time = 1, timeUnit = TimeUnit.SECONDS)
+public class HunspellStemmerComparisonBenchmark {
+
+    /**
+     * Parameterized benchmark case.
+     */
+    @State(Scope.Benchmark)
+    public static class SharedState {
+
+        /**
+         * Selected language case.
+         */
+        @Param({ "ENGLISH", "CZECH", "GERMAN", "SPANISH", "FRENCH", "DUTCH", "POLISH", "UKRAINIAN" })
+        public String languageCaseName;
+
+        /**
+         * Selected language case descriptor.
+         */
+        private HunspellLanguageCase languageCase;
+
+        /**
+         * Shared deterministic changed-token corpus.
+         */
+        private String[] tokens;
+
+        /**
+         * Radixor benchmark adapter.
+         */
+        private RadixorBenchmarkStemmer radixorStemmer;
+
+        /**
+         * Initializes the selected language corpus and Radixor stemmer.
+         *
+         * @throws IOException if the Radixor corpus or trie cannot be loaded
+         */
+        @Setup(Level.Trial)
+        public void setUp() throws IOException {
+            this.languageCase = HunspellLanguageCase.valueOf(this.languageCaseName);
+            this.tokens = LanguageBenchmarkCorpus.createTokens(this.languageCase.radixorLanguage());
+            final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadCompiled(
+                    this.languageCase.radixorLanguage(), true,
+                    ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
+            this.radixorStemmer = new RadixorBenchmarkStemmer(trie);
+        }
+    }
+
+    /**
+     * Reusable Hunspell filter state.
+     */
+    @State(Scope.Thread)
+    public static class HunspellState {
+
+        /**
+         * Reusable benchmark input stream.
+         */
+        private BenchmarkTokenStream input;
+
+        /**
+         * Reusable Hunspell filter output stream.
+         */
+        private TokenStream output;
+
+        /**
+         * Output term attribute.
+         */
+        private CharTermAttribute termAttribute;
+
+        /**
+         * Initializes the Hunspell dictionary and filter for the selected language.
+         *
+         * @param sharedState selected language state
+         * @throws IOException    if dictionary resources cannot be read
+         * @throws ParseException if the Hunspell dictionary cannot be parsed
+         */
+        @Setup(Level.Trial)
+        public void setUp(final SharedState sharedState) throws IOException, ParseException {
+            this.input = new BenchmarkTokenStream(new String[0]);
+            final Dictionary dictionary = loadDictionary(sharedState.languageCase);
+            this.output = new HunspellStemFilter(new LowerCaseFilter(this.input), dictionary, true);
+            this.termAttribute = this.output.addAttribute(CharTermAttribute.class);
+        }
+
+        /**
+         * Runs Hunspell over one token corpus.
+         *
+         * @param tokens    token corpus
+         * @param blackhole result sink
+         * @throws IOException if Lucene token streaming fails
+         */
+        private void run(final String[] tokens, final Blackhole blackhole) throws IOException {
+            this.input.setTokens(tokens);
+            this.output.reset();
+            while (this.output.incrementToken()) {
+                blackhole.consume(this.termAttribute.toString());
+            }
+            this.output.end();
+        }
+    }
+
+    /**
+     * Runs Radixor direct lookup and patch application.
+     *
+     * @param sharedState selected language state
+     * @param blackhole   result sink
+     */
+    @Benchmark
+    public void radixor(final SharedState sharedState, final Blackhole blackhole) {
+        final String[] tokens = sharedState.tokens;
+        final RadixorBenchmarkStemmer stemmer = sharedState.radixorStemmer;
+        for (String token : tokens) {
+            blackhole.consume(stemmer.stem(token));
+        }
+    }
+
+    /**
+     * Runs Lucene HunspellStemFilter over the selected language corpus.
+     *
+     * @param sharedState selected language state
+     * @param hunspellState reusable Hunspell state
+     * @param blackhole result sink
+     * @throws IOException if Lucene token streaming fails
+     */
+    @Benchmark
+    public void luceneHunspellStemFilter(final SharedState sharedState, final HunspellState hunspellState,
+            final Blackhole blackhole) throws IOException {
+        hunspellState.run(sharedState.tokens, blackhole);
+    }
+
+    /**
+     * Loads a benchmark-only Hunspell dictionary from generated JMH resources.
+     *
+     * @param languageCase selected language case
+     * @return parsed Hunspell dictionary
+     * @throws IOException    if dictionary resources cannot be read
+     * @throws ParseException if the Hunspell dictionary cannot be parsed
+     */
+    private static Dictionary loadDictionary(final HunspellLanguageCase languageCase) throws IOException,
+            ParseException {
+        final ClassLoader classLoader = HunspellStemmerComparisonBenchmark.class.getClassLoader();
+        final String basePath = "hunspell/" + languageCase.hunspellResourceCode() + "/index.";
+        try (InputStream affixStream = openRequiredResource(classLoader, basePath + "aff");
+                InputStream dictionaryStream = openRequiredResource(classLoader, basePath + "dic")) {
+            return new Dictionary(affixStream, List.of(dictionaryStream), true, SortingStrategy.inMemory());
+        }
+    }
+
+    /**
+     * Opens a classpath resource or fails with a descriptive exception.
+     *
+     * @param classLoader class loader
+     * @param path resource path
+     * @return resource stream
+     */
+    private static InputStream openRequiredResource(final ClassLoader classLoader, final String path) {
+        final InputStream stream = classLoader.getResourceAsStream(path);
+        if (stream == null) {
+            throw new IllegalStateException("Missing benchmark-only Hunspell resource: " + path);
+        }
+        return stream;
+    }
+
+    /**
+     * Benchmark language mapping.
+     */
+    private enum HunspellLanguageCase {
+
+        /**
+         * English Hunspell dictionary over the Radixor English corpus.
+         */
+        ENGLISH("en", StemmerPatchTrieLoader.Language.US_UK),
+
+        /**
+         * Czech Hunspell dictionary over the Radixor Czech corpus.
+         */
+        CZECH("cs", StemmerPatchTrieLoader.Language.CS_CZ),
+
+        /**
+         * German Hunspell dictionary over the Radixor German corpus.
+         */
+        GERMAN("de", StemmerPatchTrieLoader.Language.DE_DE),
+
+        /**
+         * Spanish Hunspell dictionary over the Radixor Spanish corpus.
+         */
+        SPANISH("es", StemmerPatchTrieLoader.Language.ES_ES),
+
+        /**
+         * French Hunspell dictionary over the Radixor French corpus.
+         */
+        FRENCH("fr", StemmerPatchTrieLoader.Language.FR_FR),
+
+        /**
+         * Dutch Hunspell dictionary over the Radixor Dutch corpus.
+         */
+        DUTCH("nl", StemmerPatchTrieLoader.Language.NL_NL),
+
+        /**
+         * Polish Hunspell dictionary over the Radixor Polish corpus.
+         */
+        POLISH("pl", StemmerPatchTrieLoader.Language.PL_PL),
+
+        /**
+         * Ukrainian Hunspell dictionary over the Radixor Ukrainian corpus.
+         */
+        UKRAINIAN("uk", StemmerPatchTrieLoader.Language.UK_UA);
+
+        /**
+         * wooorm/dictionaries resource code.
+         */
+        private final String hunspellResourceCode;
+
+        /**
+         * Matching Radixor language.
+         */
+        private final StemmerPatchTrieLoader.Language radixorLanguage;
+
+        /**
+         * Creates a language mapping.
+         *
+         * @param hunspellResourceCode Hunspell resource code
+         * @param radixorLanguage      Radixor language
+         */
+        HunspellLanguageCase(final String hunspellResourceCode, final StemmerPatchTrieLoader.Language radixorLanguage) {
+            this.hunspellResourceCode = hunspellResourceCode.toLowerCase(Locale.ROOT);
+            this.radixorLanguage = radixorLanguage;
+        }
+
+        /**
+         * Returns the Hunspell resource code.
+         *
+         * @return resource code
+         */
+        String hunspellResourceCode() {
+            return this.hunspellResourceCode;
+        }
+
+        /**
+         * Returns the matching Radixor language.
+         *
+         * @return Radixor language
+         */
+        StemmerPatchTrieLoader.Language radixorLanguage() {
+            return this.radixorLanguage;
+        }
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/HunspellStemmerComparisonBenchmarkQuality.java

@@ -0,0 +1,436 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.IOException;
+import java.io.InputStream;
+import java.text.ParseException;
+import java.util.List;
+import java.util.Locale;
+import java.util.Objects;
+import java.util.concurrent.TimeUnit;
+
+import org.apache.lucene.analysis.LowerCaseFilter;
+import org.apache.lucene.analysis.TokenStream;
+import org.apache.lucene.analysis.hunspell.Dictionary;
+import org.apache.lucene.analysis.hunspell.HunspellStemFilter;
+import org.apache.lucene.analysis.hunspell.SortingStrategy;
+import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
+import org.apache.lucene.analysis.tokenattributes.PositionIncrementAttribute;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+import org.openjdk.jmh.annotations.AuxCounters;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Fork;
+import org.openjdk.jmh.annotations.Level;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+import org.openjdk.jmh.annotations.Param;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.Setup;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.Warmup;
+import org.openjdk.jmh.infra.Blackhole;
+
+/**
+ * Emits exact-root agreement metrics for the benchmark-only Hunspell comparisons.
+ *
+ * <p>
+ * This class mirrors the existing Hunspell throughput setup but adds
+ * quality-style accuracy counters for every Hunspell language dictionary used
+ * in benchmark-only throughput comparisons.
+ * </p>
+ */
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.NANOSECONDS)
+@Warmup(iterations = 0)
+@Measurement(iterations = 1, time = 1, timeUnit = TimeUnit.MILLISECONDS)
+@Fork(0)
+public class HunspellStemmerComparisonBenchmarkQuality {
+
+    /**
+     * Shared quality corpus and Hunspell dictionary for a selected language.
+     */
+    @State(Scope.Benchmark)
+    public static class SharedState {
+
+        /**
+         * Selected language case.
+         */
+        @Param({ "ENGLISH", "CZECH", "GERMAN", "SPANISH", "FRENCH", "DUTCH", "POLISH", "UKRAINIAN" })
+        public String languageCaseName;
+
+        /**
+         * Selected language descriptor.
+         */
+        private HunspellLanguageCase languageCase;
+
+        /**
+         * Complete language dictionary corpus and expected roots.
+         */
+        private LanguageBenchmarkCorpus.Corpus corpus;
+
+        /**
+         * Parsed benchmark-only Hunspell dictionary.
+         */
+        private Dictionary dictionary;
+
+        /**
+         * Initializes quality resources.
+         *
+         * @throws IOException    if corpus or dictionary loading fails
+         * @throws ParseException if the Hunspell dictionary cannot be parsed
+         */
+        @Setup(Level.Trial)
+        public void setUp() throws IOException, ParseException {
+            this.languageCase = HunspellLanguageCase.valueOf(this.languageCaseName);
+            this.corpus = LanguageBenchmarkCorpus.createFullCorpus(this.languageCase.radixorLanguage());
+            this.dictionary = loadDictionary(this.languageCase);
+        }
+    }
+
+    /**
+     * JMH auxiliary counters for exact-root agreement.
+     */
+    @State(Scope.Thread)
+    @AuxCounters(AuxCounters.Type.EVENTS)
+    public static class AccuracyCounters {
+
+        /**
+         * Number of exact-root matches.
+         */
+        public long correctMatches;
+
+        /**
+         * Number of evaluated tokens.
+         */
+        public long evaluatedTokens;
+
+        /**
+         * Number of exact-root matches where the input token differs from the
+         * expected root.
+         */
+        public long changedCorrectMatches;
+
+        /**
+         * Number of evaluated tokens where the input token differs from the expected
+         * root.
+         */
+        public long changedEvaluatedTokens;
+
+        /**
+         * Number of exact-root matches where the input token is already the expected
+         * root.
+         */
+        public long rootPreservedMatches;
+
+        /**
+         * Number of evaluated tokens where the input token is already the expected
+         * root.
+         */
+        public long rootEvaluatedTokens;
+
+        /**
+         * Resets counters before each measured iteration.
+         */
+        @Setup(Level.Iteration)
+        public void reset() {
+            this.correctMatches = 0L;
+            this.evaluatedTokens = 0L;
+            this.changedCorrectMatches = 0L;
+            this.changedEvaluatedTokens = 0L;
+            this.rootPreservedMatches = 0L;
+            this.rootEvaluatedTokens = 0L;
+        }
+    }
+
+    /**
+     * Evaluates exact-root agreement for the selected Hunspell dictionary.
+     *
+     * @param sharedState shared quality state
+     * @param counters    JMH auxiliary counters
+     * @param blackhole   result sink
+     * @return exact-root match count
+     * @throws IOException if Lucene token streaming fails
+     */
+    @Benchmark
+    public int luceneHunspellStemFilterAccuracy(final SharedState sharedState, final AccuracyCounters counters,
+            final Blackhole blackhole) throws IOException {
+        final String[] actualStems = firstHunspellOutputs(sharedState.corpus.tokens(), sharedState.dictionary,
+                blackhole);
+        final String[] tokens = sharedState.corpus.tokens();
+        final String[] expectedRoots = sharedState.corpus.expectedRoots();
+
+        int correct = 0;
+        int changedCorrect = 0;
+        int changedEvaluated = 0;
+        int rootPreserved = 0;
+        int rootEvaluated = 0;
+        for (int index = 0; index < actualStems.length; index++) {
+            final String token = tokens[index];
+            final String expectedRoot = expectedRoots[index];
+            final boolean exact = Objects.equals(expectedRoot, actualStems[index]);
+            if (exact) {
+                correct++;
+            }
+            if (Objects.equals(token, expectedRoot)) {
+                rootEvaluated++;
+                if (exact) {
+                    rootPreserved++;
+                }
+            } else {
+                changedEvaluated++;
+                if (exact) {
+                    changedCorrect++;
+                }
+            }
+        }
+
+        counters.correctMatches += correct;
+        counters.evaluatedTokens += actualStems.length;
+        counters.changedCorrectMatches += changedCorrect;
+        counters.changedEvaluatedTokens += changedEvaluated;
+        counters.rootPreservedMatches += rootPreserved;
+        counters.rootEvaluatedTokens += rootEvaluated;
+        return correct;
+    }
+
+    /**
+     * Extracts the first emitted Hunspell stem for each input token.
+     *
+     * @param tokens     token corpus
+     * @param dictionary Hunspell dictionary
+     * @param blackhole  result sink
+     * @return first emitted term per input token
+     * @throws IOException if Lucene streaming fails
+     */
+    private static String[] firstHunspellOutputs(final String[] tokens, final Dictionary dictionary,
+            final Blackhole blackhole) throws IOException {
+        final String[] outputs = new String[tokens.length];
+        final BenchmarkTokenStream input = new BenchmarkTokenStream(tokens);
+        final TokenStream output = new HunspellStemFilter(new LowerCaseFilter(input), dictionary, true);
+        final CharTermAttribute termAttribute = output.addAttribute(CharTermAttribute.class);
+        final PositionIncrementAttribute positionAttribute = output.addAttribute(PositionIncrementAttribute.class);
+        int inputIndex = -1;
+        boolean recordedForPosition = false;
+
+        output.reset();
+        while (output.incrementToken()) {
+            final int positionIncrement = positionAttribute.getPositionIncrement();
+            if (positionIncrement > 0) {
+                inputIndex += positionIncrement;
+                recordedForPosition = false;
+            }
+            if (inputIndex >= 0 && inputIndex < outputs.length && !recordedForPosition) {
+                outputs[inputIndex] = termAttribute.toString();
+                recordedForPosition = true;
+            }
+            if (blackhole != null) {
+                blackhole.consume(termAttribute);
+            }
+        }
+        output.end();
+        output.close();
+
+        for (int index = 0; index < outputs.length; index++) {
+            if (outputs[index] == null) {
+                outputs[index] = tokens[index];
+            }
+        }
+        return outputs;
+    }
+
+    /**
+     * Loads a benchmark-only Hunspell dictionary from generated resources.
+     *
+     * @param languageCase selected language case
+     * @return parsed dictionary
+     * @throws IOException    if dictionary resources cannot be read
+     * @throws ParseException if dictionary parsing fails
+     */
+    private static Dictionary loadDictionary(final HunspellLanguageCase languageCase) throws IOException,
+            ParseException {
+        final ClassLoader classLoader = HunspellStemmerComparisonBenchmarkQuality.class.getClassLoader();
+        final String basePath = "hunspell/" + languageCase.hunspellResourceCode() + "/index.";
+        try (InputStream affixStream = openRequiredResource(classLoader, basePath + "aff");
+                InputStream dictionaryStream = openRequiredResource(classLoader, basePath + "dic")) {
+            return new Dictionary(affixStream, List.of(dictionaryStream), true, SortingStrategy.inMemory());
+        }
+    }
+
+    /**
+     * Stems one analytical batch through the exact Hunspell quality-benchmark path.
+     *
+     * @param languageCase declared Hunspell language case
+     * @param tokens original dictionary forms
+     * @return first Hunspell output per input form
+     * @throws IOException if dictionary parsing or token streaming fails
+     */
+    static String[] stemForQuality(final HunspellLanguageCase languageCase, final String[] tokens) throws IOException {
+        try {
+            return firstHunspellOutputs(tokens, loadDictionary(languageCase), null);
+        } catch (ParseException exception) {
+            throw new IOException("Unable to parse the JMH Hunspell dictionary for " + languageCase + ".", exception);
+        }
+    }
+
+    /** Returns all distinct Hunspell stems per token through the quality-benchmark dictionary. */
+    static List<List<String>> stemCandidatesForQuality(final HunspellLanguageCase languageCase,
+            final String[] tokens) throws IOException {
+        try {
+            final Dictionary dictionary = loadDictionary(languageCase);
+            final List<java.util.LinkedHashSet<String>> candidates = new java.util.ArrayList<>(tokens.length);
+            for (int index = 0; index < tokens.length; index++) { candidates.add(new java.util.LinkedHashSet<>()); }
+            final BenchmarkTokenStream input = new BenchmarkTokenStream(tokens);
+            final TokenStream output = new HunspellStemFilter(new LowerCaseFilter(input), dictionary, true);
+            final CharTermAttribute term = output.addAttribute(CharTermAttribute.class);
+            final PositionIncrementAttribute position = output.addAttribute(PositionIncrementAttribute.class);
+            int inputIndex = -1;
+            output.reset();
+            while (output.incrementToken()) {
+                if (position.getPositionIncrement() > 0) { inputIndex += position.getPositionIncrement(); }
+                if (inputIndex >= 0 && inputIndex < candidates.size()) { candidates.get(inputIndex).add(term.toString()); }
+            }
+            output.end();
+            output.close();
+            final String[] primary = firstHunspellOutputs(tokens, dictionary, null);
+            final List<List<String>> result = new java.util.ArrayList<>(tokens.length);
+            for (int index = 0; index < tokens.length; index++) {
+                candidates.get(index).add(primary[index]);
+                result.add(List.copyOf(candidates.get(index)));
+            }
+            return List.copyOf(result);
+        } catch (ParseException exception) {
+            throw new IOException("Unable to parse the JMH Hunspell dictionary for " + languageCase + ".", exception);
+        }
+    }
+
+    /**
+     * Opens a required classpath resource.
+     *
+     * @param classLoader class loader
+     * @param path        resource path
+     * @return resource stream
+     */
+    private static InputStream openRequiredResource(final ClassLoader classLoader, final String path) {
+        final InputStream stream = classLoader.getResourceAsStream(path);
+        if (stream == null) {
+            throw new IllegalStateException("Missing benchmark-only Hunspell resource: " + path);
+        }
+        return stream;
+    }
+
+    /**
+     * Benchmark language mapping.
+     */
+    enum HunspellLanguageCase {
+
+        /**
+         * English Hunspell dictionary over the Radixor English corpus.
+         */
+        ENGLISH("en", StemmerPatchTrieLoader.Language.US_UK),
+
+        /**
+         * Czech Hunspell dictionary over the Radixor Czech corpus.
+         */
+        CZECH("cs", StemmerPatchTrieLoader.Language.CS_CZ),
+
+        /**
+         * German Hunspell dictionary over the Radixor German corpus.
+         */
+        GERMAN("de", StemmerPatchTrieLoader.Language.DE_DE),
+
+        /**
+         * Spanish Hunspell dictionary over the Radixor Spanish corpus.
+         */
+        SPANISH("es", StemmerPatchTrieLoader.Language.ES_ES),
+
+        /**
+         * French Hunspell dictionary over the Radixor French corpus.
+         */
+        FRENCH("fr", StemmerPatchTrieLoader.Language.FR_FR),
+
+        /**
+         * Dutch Hunspell dictionary over the Radixor Dutch corpus.
+         */
+        DUTCH("nl", StemmerPatchTrieLoader.Language.NL_NL),
+
+        /**
+         * Polish Hunspell dictionary over the Radixor Polish corpus.
+         */
+        POLISH("pl", StemmerPatchTrieLoader.Language.PL_PL),
+
+        /**
+         * Ukrainian Hunspell dictionary over the Radixor Ukrainian corpus.
+         */
+        UKRAINIAN("uk", StemmerPatchTrieLoader.Language.UK_UA);
+
+        /**
+         * Wooorm/dictionaries resource code.
+         */
+        private final String hunspellResourceCode;
+
+        /**
+         * Matching Radixor language.
+         */
+        private final StemmerPatchTrieLoader.Language radixorLanguage;
+
+        /**
+         * Creates a language mapping.
+         *
+         * @param hunspellResourceCode Hunspell resource code
+         * @param radixorLanguage      Radixor language
+         */
+        HunspellLanguageCase(final String hunspellResourceCode, final StemmerPatchTrieLoader.Language radixorLanguage) {
+            this.hunspellResourceCode = hunspellResourceCode.toLowerCase(Locale.ROOT);
+            this.radixorLanguage = radixorLanguage;
+        }
+
+        /**
+         * Returns the Hunspell resource code.
+         *
+         * @return resource code
+         */
+        String hunspellResourceCode() {
+            return this.hunspellResourceCode;
+        }
+
+        /**
+         * Returns the matching Radixor language.
+         *
+         * @return Radixor language
+         */
+        StemmerPatchTrieLoader.Language radixorLanguage() {
+            return this.radixorLanguage;
+        }
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/LanguageBenchmarkCorpus.java

@@ -0,0 +1,445 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.BufferedReader;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.InputStreamReader;
+import java.nio.charset.StandardCharsets;
+import java.util.ArrayList;
+import java.util.EnumMap;
+import java.util.List;
+import java.util.Map;
+import java.util.Locale;
+import java.util.Objects;
+import java.util.zip.GZIPInputStream;
+
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+
+/**
+ * Builds deterministic language-specific benchmark corpora from bundled
+ * Radixor dictionary resources.
+ *
+ * <p>
+ * Corpus construction is setup work only. It is intentionally based on the same
+ * resource that backs the Radixor benchmark path so every competitor for a
+ * language consumes the same changed-token timing workload, while quality
+ * benchmarks can still use the complete dictionary workload.
+ * </p>
+ */
+final class LanguageBenchmarkCorpus {
+
+    /**
+     * Minimum token count for timing benchmark operations.
+     */
+    static final int MINIMUM_TIMING_TOKEN_COUNT = 5_000;
+
+    /**
+     * Shared timing corpora keyed by bundled Radixor language.
+     */
+    private static final Map<StemmerPatchTrieLoader.Language, Corpus> TIMING_CORPORA =
+            new EnumMap<>(StemmerPatchTrieLoader.Language.class);
+
+    /**
+     * Shared changed-token timing corpora keyed by bundled Radixor language.
+     */
+    private static final Map<StemmerPatchTrieLoader.Language, Corpus> CHANGED_TIMING_CORPORA =
+            new EnumMap<>(StemmerPatchTrieLoader.Language.class);
+
+    /**
+     * Shared complete corpora keyed by bundled Radixor language.
+     */
+    private static final Map<StemmerPatchTrieLoader.Language, Corpus> FULL_CORPORA =
+            new EnumMap<>(StemmerPatchTrieLoader.Language.class);
+
+    /**
+     * Utility class.
+     */
+    private LanguageBenchmarkCorpus() {
+        throw new AssertionError("No instances.");
+    }
+
+    /**
+     * Creates a deterministic changed-token timing corpus from a bundled language
+     * dictionary.
+     *
+     * <p>
+     * Only token/root pairs where the token differs from the expected root are
+     * included. Smaller changed-token resources are repeated in stable order until
+     * the timing corpus reaches 5,000 tokens.
+     * </p>
+     *
+     * @param language bundled Radixor language
+     * @return token array containing changed-token dictionary entries, repeated
+     *         only when the changed-token resource is smaller than 5,000 tokens
+     * @throws IOException if the resource cannot be read
+     */
+    static String[] createTokens(final StemmerPatchTrieLoader.Language language) throws IOException {
+        return createChangedCorpus(language).tokens();
+    }
+
+    /**
+     * Creates a deterministic changed-token timing corpus from a bundled language
+     * dictionary.
+     *
+     * @param language bundled Radixor language
+     * @return changed-token corpus with expected roots
+     * @throws IOException if the resource cannot be read
+     */
+    static Corpus createChangedCorpus(final StemmerPatchTrieLoader.Language language) throws IOException {
+        return cachedChangedCorpus(language);
+    }
+
+    /**
+     * Creates a deterministic full-dictionary timing corpus and expected root
+     * array from a bundled language dictionary.
+     *
+     * @param language bundled Radixor language
+     * @return token corpus with expected roots
+     * @throws IOException if the resource cannot be read
+     */
+    static Corpus createCorpus(final StemmerPatchTrieLoader.Language language) throws IOException {
+        return cachedCorpus(TIMING_CORPORA, language, true);
+    }
+
+    /**
+     * Creates a deterministic full-dictionary timing corpus and expected root
+     * array from a bundled language dictionary.
+     *
+     * <p>
+     * The complete dictionary token sequence is used when it contains at least
+     * {@code minimumTokenCount} tokens. Smaller resources are repeated in stable
+     * order until the minimum is reached.
+     * </p>
+     *
+     * @param language          bundled Radixor language
+     * @param minimumTokenCount minimum token count for timing
+     * @return token corpus with expected roots
+     * @throws IOException if the resource cannot be read
+     */
+    static Corpus createCorpus(final StemmerPatchTrieLoader.Language language, final int minimumTokenCount)
+            throws IOException {
+        Objects.requireNonNull(language, "language");
+        if (minimumTokenCount < 1) {
+            throw new IllegalArgumentException("minimumTokenCount must be at least 1.");
+        }
+        if (minimumTokenCount == MINIMUM_TIMING_TOKEN_COUNT) {
+            return createCorpus(language);
+        }
+
+        return buildTimingCorpus(language, minimumTokenCount);
+    }
+
+    /**
+     * Creates or returns the shared complete corpus for a bundled language.
+     *
+     * @param language bundled Radixor language
+     * @return complete token corpus with expected roots
+     * @throws IOException if the resource cannot be read
+     */
+    static Corpus createFullCorpus(final StemmerPatchTrieLoader.Language language) throws IOException {
+        return cachedCorpus(FULL_CORPORA, language, false);
+    }
+
+    /**
+     * Returns a cached corpus, creating it once per JVM when necessary.
+     *
+     * @param cache corpus cache
+     * @param language bundled Radixor language
+     * @param timing whether the timing-minimum corpus should be built
+     * @return cached corpus instance
+     * @throws IOException if the resource cannot be read
+     */
+    private static Corpus cachedCorpus(final Map<StemmerPatchTrieLoader.Language, Corpus> cache,
+            final StemmerPatchTrieLoader.Language language, final boolean timing) throws IOException {
+        Objects.requireNonNull(cache, "cache");
+        Objects.requireNonNull(language, "language");
+
+        synchronized (LanguageBenchmarkCorpus.class) {
+            final Corpus existing = cache.get(language);
+            if (existing != null) {
+                return existing;
+            }
+
+            final Corpus created = timing ? buildTimingCorpus(language, MINIMUM_TIMING_TOKEN_COUNT)
+                    : buildFullCorpus(language);
+            cache.put(language, created);
+            return created;
+        }
+    }
+
+    /**
+     * Returns a cached changed-token timing corpus, creating it once per JVM when
+     * necessary.
+     *
+     * @param language bundled Radixor language
+     * @return changed-token timing corpus
+     * @throws IOException if the resource cannot be read
+     */
+    private static Corpus cachedChangedCorpus(final StemmerPatchTrieLoader.Language language) throws IOException {
+        Objects.requireNonNull(language, "language");
+
+        synchronized (LanguageBenchmarkCorpus.class) {
+            final Corpus existing = CHANGED_TIMING_CORPORA.get(language);
+            if (existing != null) {
+                return existing;
+            }
+
+            final Corpus created = buildChangedTimingCorpus(language, MINIMUM_TIMING_TOKEN_COUNT);
+            CHANGED_TIMING_CORPORA.put(language, created);
+            return created;
+        }
+    }
+
+    /**
+     * Builds a deterministic timing corpus from a bundled language dictionary.
+     *
+     * @param language bundled Radixor language
+     * @param minimumTokenCount minimum token count for timing
+     * @return token corpus with expected roots
+     * @throws IOException if the resource cannot be read
+     */
+    private static Corpus buildTimingCorpus(final StemmerPatchTrieLoader.Language language, final int minimumTokenCount)
+            throws IOException {
+        Objects.requireNonNull(language, "language");
+        if (minimumTokenCount < 1) {
+            throw new IllegalArgumentException("minimumTokenCount must be at least 1.");
+        }
+
+        final List<Entry> candidates = readCandidates(language, Integer.MAX_VALUE);
+        if (candidates.isEmpty()) {
+            throw new IllegalStateException("No benchmark corpus tokens were available for " + language + ".");
+        }
+
+        final int timingTokenCount = Math.max(candidates.size(), minimumTokenCount);
+        final String[] tokens = new String[timingTokenCount];
+        final String[] expectedRoots = new String[timingTokenCount];
+        for (int index = 0; index < tokens.length; index++) {
+            final Entry entry = candidates.get(index % candidates.size());
+            tokens[index] = entry.token();
+            expectedRoots[index] = entry.root();
+        }
+        return new Corpus(tokens, expectedRoots);
+    }
+
+    /**
+     * Builds a deterministic changed-token timing corpus from a bundled language
+     * dictionary.
+     *
+     * @param language bundled Radixor language
+     * @param minimumTokenCount minimum token count for timing
+     * @return changed-token corpus with expected roots
+     * @throws IOException if the resource cannot be read
+     */
+    private static Corpus buildChangedTimingCorpus(final StemmerPatchTrieLoader.Language language,
+            final int minimumTokenCount) throws IOException {
+        Objects.requireNonNull(language, "language");
+        if (minimumTokenCount < 1) {
+            throw new IllegalArgumentException("minimumTokenCount must be at least 1.");
+        }
+
+        final List<Entry> allCandidates = readCandidates(language, Integer.MAX_VALUE);
+        final List<Entry> changedCandidates = new ArrayList<>(allCandidates.size());
+        for (Entry entry : allCandidates) {
+            if (!Objects.equals(entry.token(), entry.root())) {
+                changedCandidates.add(entry);
+            }
+        }
+        if (changedCandidates.isEmpty()) {
+            throw new IllegalStateException("No changed-token benchmark corpus tokens were available for "
+                    + language + ".");
+        }
+
+        final int timingTokenCount = Math.max(changedCandidates.size(), minimumTokenCount);
+        final String[] tokens = new String[timingTokenCount];
+        final String[] expectedRoots = new String[timingTokenCount];
+        for (int index = 0; index < tokens.length; index++) {
+            final Entry entry = changedCandidates.get(index % changedCandidates.size());
+            tokens[index] = entry.token();
+            expectedRoots[index] = entry.root();
+        }
+        return new Corpus(tokens, expectedRoots);
+    }
+
+    /**
+     * Creates a complete deterministic token corpus and expected root array from a
+     * bundled language dictionary.
+     *
+     * <p>
+     * This method is intended for exact-root quality accounting. It includes all
+     * single-token fields available in the dictionary resource and does not repeat
+     * small dictionaries to the timing minimum.
+     * </p>
+     *
+     * @param language bundled Radixor language
+     * @return complete token corpus with expected roots
+     * @throws IOException if the resource cannot be read
+     */
+    private static Corpus buildFullCorpus(final StemmerPatchTrieLoader.Language language) throws IOException {
+        Objects.requireNonNull(language, "language");
+
+        final List<Entry> candidates = readCandidates(language, Integer.MAX_VALUE);
+        if (candidates.isEmpty()) {
+            throw new IllegalStateException("No benchmark corpus tokens were available for " + language + ".");
+        }
+
+        final String[] tokens = new String[candidates.size()];
+        final String[] expectedRoots = new String[candidates.size()];
+        for (int index = 0; index < tokens.length; index++) {
+            final Entry entry = candidates.get(index);
+            tokens[index] = entry.token();
+            expectedRoots[index] = entry.root();
+        }
+        return new Corpus(tokens, expectedRoots);
+    }
+
+    /**
+     * Reads token candidates from a bundled compressed dictionary.
+     *
+     * @param language bundled Radixor language
+     * @param maximumTokenCount maximum token count to read
+     * @return deterministic candidate list
+     * @throws IOException if the resource cannot be read
+     */
+    private static List<Entry> readCandidates(final StemmerPatchTrieLoader.Language language, final int maximumTokenCount)
+            throws IOException {
+        final String resourcePath = language.resourcePath();
+        final InputStream resource = StemmerPatchTrieLoader.class.getClassLoader().getResourceAsStream(resourcePath);
+        if (resource == null) {
+            throw new IllegalStateException("Missing bundled benchmark resource " + resourcePath + ".");
+        }
+
+        final List<Entry> candidates = new ArrayList<>(MINIMUM_TIMING_TOKEN_COUNT);
+        try (InputStream inputStream = resource;
+                GZIPInputStream gzipInputStream = new GZIPInputStream(inputStream);
+                InputStreamReader inputStreamReader = new InputStreamReader(gzipInputStream, StandardCharsets.UTF_8);
+                BufferedReader reader = new BufferedReader(inputStreamReader)) {
+            String line = reader.readLine();
+            while (line != null && candidates.size() < maximumTokenCount) {
+                collectLineCandidates(line, candidates, maximumTokenCount);
+                line = reader.readLine();
+            }
+        }
+        return candidates;
+    }
+
+    /**
+     * Collects lower-case token candidates from one dictionary line.
+     *
+     * @param line       dictionary line
+     * @param candidates mutable candidate list
+     * @param maximumTokenCount maximum token count to read
+     */
+    private static void collectLineCandidates(final String line, final List<Entry> candidates,
+            final int maximumTokenCount) {
+        if (line == null || line.isBlank() || line.startsWith("#") || line.startsWith("//")) {
+            return;
+        }
+
+        final String[] fields = line.split("\t");
+        if (fields.length == 0) {
+            return;
+        }
+
+        final String root = normalizeToken(fields[0]);
+        if (root.isEmpty() || containsWhitespace(root)) {
+            return;
+        }
+
+        for (String field : fields) {
+            if (candidates.size() >= maximumTokenCount) {
+                return;
+            }
+            final String token = normalizeToken(field);
+            if (!token.isEmpty() && !containsWhitespace(token)) {
+                candidates.add(new Entry(token, root));
+            }
+        }
+    }
+
+    /**
+     * Normalizes dictionary token text for deterministic benchmark lookup.
+     *
+     * @param token dictionary token field
+     * @return normalized token
+     */
+    private static String normalizeToken(final String token) {
+        return token.trim().toLowerCase(Locale.ROOT);
+    }
+
+    /**
+     * Returns whether a token contains Unicode whitespace.
+     *
+     * @param token token candidate
+     * @return {@code true} when whitespace is present
+     */
+    private static boolean containsWhitespace(final String token) {
+        for (int index = 0; index < token.length(); index++) {
+            if (Character.isWhitespace(token.charAt(index))) {
+                return true;
+            }
+        }
+        return false;
+    }
+
+    /**
+     * Immutable token corpus with expected roots.
+     *
+     * @param tokens        benchmark token corpus
+     * @param expectedRoots expected root for each token
+     */
+    record Corpus(String[] tokens, String[] expectedRoots) {
+
+        /**
+         * Creates corpus data.
+         *
+         * @param tokens        benchmark token corpus
+         * @param expectedRoots expected root for each token
+         */
+        Corpus {
+            Objects.requireNonNull(tokens, "tokens");
+            Objects.requireNonNull(expectedRoots, "expectedRoots");
+            if (tokens.length != expectedRoots.length) {
+                throw new IllegalArgumentException("tokens and expectedRoots must have the same length.");
+            }
+        }
+    }
+
+    /**
+     * Immutable dictionary-derived token/root entry.
+     *
+     * @param token token form
+     * @param root  expected root
+     */
+    private record Entry(String token, String root) {
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/QualityStemmerMatrix.java

@@ -0,0 +1,133 @@
+package org.egothor.stemmer.benchmark;
+
+import java.io.IOException;
+import java.util.Arrays;
+import java.util.List;
+import java.util.Objects;
+import java.util.ArrayList;
+import java.util.EnumSet;
+
+import org.egothor.stemmer.StemmerPatchTrieLoader.Language;
+
+/** Authoritative analytical view of the candidate matrix defined by the JMH quality benchmark. */
+public final class QualityStemmerMatrix {
+    /** Utility class. */
+    private QualityStemmerMatrix() {
+        throw new AssertionError("No instances.");
+    }
+
+    /**
+     * Returns every currently registered JMH quality candidate in declaration order.
+     * The returned list is immutable and is derived directly from the benchmark enum.
+     *
+     * @return complete immutable candidate list
+     */
+    public static List<Candidate> candidates() {
+        final List<Candidate> candidates = new ArrayList<>();
+        Arrays.stream(StemmerComparisonBenchmarkQuality.QualityCandidate.values())
+                .map(candidate -> new Candidate(candidate.name(), candidate.radixorLanguage(),
+                        () -> adapt(candidate.createStemmer())))
+                .forEach(candidates::add);
+        final EnumSet<Language> registeredRadixorLanguages = candidates.stream()
+                .filter(candidate -> candidate.name().endsWith("_RADIXOR"))
+                .map(Candidate::language).collect(() -> EnumSet.noneOf(Language.class), EnumSet::add, EnumSet::addAll);
+        Arrays.stream(Language.values()).filter(language -> !registeredRadixorLanguages.contains(language))
+                .map(language -> new Candidate(language.name() + "_RADIXOR", language,
+                        () -> adapt(StemmerComparisonBenchmarkQuality.createRadixorQualityStemmer(language))))
+                .forEach(candidates::add);
+        Arrays.stream(HunspellStemmerComparisonBenchmarkQuality.HunspellLanguageCase.values())
+                .map(languageCase -> new Candidate("HUNSPELL_" + languageCase.name() + "_LUCENE_FILTER",
+                        languageCase.radixorLanguage(),
+                        () -> new BatchStemmer() {
+                            /** {@inheritDoc} */
+                            @Override public String[] stem(final String[] forms) throws IOException {
+                                return HunspellStemmerComparisonBenchmarkQuality.stemForQuality(languageCase, forms);
+                            }
+                            /** {@inheritDoc} */
+                            @Override public List<List<String>> stemCandidates(final String[] forms) throws IOException {
+                                return HunspellStemmerComparisonBenchmarkQuality.stemCandidatesForQuality(languageCase, forms);
+                            }
+                            /** {@inheritDoc} */
+                            @Override public boolean supportsMultipleOutputs() { return true; }
+                        }))
+                .forEach(candidates::add);
+        return List.copyOf(candidates);
+    }
+
+    /** Adapts one authoritative general-matrix stemmer without changing capability semantics. */
+    private static BatchStemmer adapt(final StemmerComparisonBenchmarkQuality.CandidateStemmer stemmer) {
+        return new BatchStemmer() {
+            /** {@inheritDoc} */
+            @Override public String[] stem(final String[] forms) throws IOException { return stemmer.stem(forms); }
+            /** {@inheritDoc} */
+            @Override public List<List<String>> stemCandidates(final String[] forms) throws IOException {
+                return stemmer.stemCandidates(forms);
+            }
+            /** {@inheritDoc} */
+            @Override public boolean supportsMultipleOutputs() { return stemmer.supportsMultipleOutputs(); }
+        };
+    }
+
+    /** One JMH candidate and its authoritative dictionary-language mapping. */
+    public static final class Candidate {
+        private final String name;
+        private final Language language;
+        private final StemmerFactory factory;
+
+        /** Creates an immutable facade over one benchmark candidate. */
+        private Candidate(final String name, final Language language,
+                final StemmerFactory factory) {
+            this.name = Objects.requireNonNull(name, "name");
+            this.language = Objects.requireNonNull(language, "language");
+            this.factory = Objects.requireNonNull(factory, "factory");
+        }
+
+        /** @return stable JMH candidate name */
+        public String name() {
+            return this.name;
+        }
+
+        /** @return registered Radixor gold-standard dictionary language */
+        public Language language() {
+            return this.language;
+        }
+
+        /**
+         * Creates a scenario-confined adapter using exactly the JMH factory and preprocessing path.
+         *
+         * @return sequential batch stemmer
+         * @throws IOException if benchmark-only resources cannot be loaded
+         */
+        public BatchStemmer createStemmer() throws IOException {
+            return this.factory.create();
+        }
+    }
+
+    /** Internal checked factory shared by the JMH quality registries. */
+    @FunctionalInterface
+    private interface StemmerFactory {
+        /** @return a scenario-confined adapter @throws IOException if resources fail */
+        BatchStemmer create() throws IOException;
+    }
+
+    /** Sequential, scenario-confined batch stemmer contract. */
+    @FunctionalInterface
+    public interface BatchStemmer {
+        /**
+         * Stems all supplied forms in order.
+         *
+         * @param forms input forms, never {@code null}
+         * @return one non-null output per form
+         * @throws IOException when the JMH adapter fails
+         */
+        String[] stem(String[] forms) throws IOException;
+
+        /** Returns complete candidate sets; single-output adapters return singleton sets. */
+        default List<List<String>> stemCandidates(final String[] forms) throws IOException {
+            return Arrays.stream(stem(forms)).map(List::of).toList();
+        }
+
+        /** @return whether this adapter exposes genuine alternative outputs */
+        default boolean supportsMultipleOutputs() { return false; }
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/RadixorBenchmarkStemmer.java

@@ -0,0 +1,103 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.util.LinkedHashSet;
+import java.util.List;
+import java.util.Objects;
+import java.util.Set;
+
+import org.egothor.stemmer.CompiledPatchCommand;
+import org.egothor.stemmer.FrequencyTrie;
+
+/**
+ * Benchmark-only Radixor stemmer adapter for the canonical preferred-result
+ * path over normalized benchmark tokens.
+ *
+ * <p>
+ * The benchmark corpus is normalized during setup, so this adapter uses
+ * {@link FrequencyTrie#getNormalizedString(String)} to avoid measuring
+ * redundant lookup-time normalization. Patch commands are applied with the
+ * traversal direction persisted in the trie metadata.
+ * </p>
+ *
+ * <p>
+ * Instances are mutable and intended for one JMH worker thread.
+ * </p>
+ */
+final class RadixorBenchmarkStemmer {
+
+    /**
+     * Compiled Radixor patch trie with decoded patch-command values.
+     */
+    private final FrequencyTrie<CompiledPatchCommand> trie;
+
+    /**
+     * Creates a benchmark stemmer around one compiled Radixor trie.
+     *
+     * @param trie compiled Radixor patch trie
+     */
+    RadixorBenchmarkStemmer(final FrequencyTrie<CompiledPatchCommand> trie) {
+        this.trie = Objects.requireNonNull(trie, "trie");
+    }
+
+    /**
+     * Stems one benchmark token through the canonical trie lookup API.
+     *
+     * @param token input token
+     * @return Radixor stem or the input token when no patch is stored
+     */
+    String stem(final String token) {
+        final CompiledPatchCommand patch = this.trie.getNormalizedString(token);
+        if (patch == null || patch.preservesAllSources()) {
+            return token;
+        }
+        return patch.apply(token);
+    }
+
+    /**
+     * Returns every distinct candidate stem from the ranked {@code getAll} path,
+     * always including the deterministic primary output.
+     *
+     * @param token original input token
+     * @return immutable candidate list in deterministic ranked order
+     */
+    List<String> stemAll(final String token) {
+        final String primary = stem(token);
+        final Set<String> candidates = new LinkedHashSet<>();
+        candidates.add(primary);
+        final CompiledPatchCommand[] patches = this.trie.getAll(token);
+        for (CompiledPatchCommand patch : patches) {
+            candidates.add(patch.preservesAllSources() ? token : patch.apply(token));
+        }
+        return List.copyOf(candidates);
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/SnowballLanguageCase.java

@@ -0,0 +1,196 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+import org.egothor.stemmer.benchmark.snowball.ext.danishStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.dutchStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.finnishStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.frenchStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.germanStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.hungarianStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.italianStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.norwegianStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.portugueseStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.russianStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.spanishStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.swedishStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.yiddishStemmer;
+
+/**
+ * Maps Radixor dictionary languages to matching official Snowball algorithms.
+ */
+enum SnowballLanguageCase {
+
+    /**
+     * Danish Snowball stemming over the Radixor Danish dictionary.
+     */
+    DANISH("Danish", StemmerPatchTrieLoader.Language.DA_DK, danishStemmer::new, "Danish"),
+
+    /**
+     * Dutch Snowball stemming over the Radixor Dutch dictionary.
+     */
+    DUTCH("Dutch", StemmerPatchTrieLoader.Language.NL_NL, dutchStemmer::new, "Dutch"),
+
+    /**
+     * Finnish Snowball stemming over the Radixor Finnish dictionary.
+     */
+    FINNISH("Finnish", StemmerPatchTrieLoader.Language.FI_FI, finnishStemmer::new, "Finnish"),
+
+    /**
+     * French Snowball stemming over the Radixor French dictionary.
+     */
+    FRENCH("French", StemmerPatchTrieLoader.Language.FR_FR, frenchStemmer::new, "French"),
+
+    /**
+     * German Snowball stemming over the Radixor German dictionary.
+     */
+    GERMAN("German", StemmerPatchTrieLoader.Language.DE_DE, germanStemmer::new, "German"),
+
+    /**
+     * Hungarian Snowball stemming over the Radixor Hungarian dictionary.
+     */
+    HUNGARIAN("Hungarian", StemmerPatchTrieLoader.Language.HU_HU, hungarianStemmer::new, "Hungarian"),
+
+    /**
+     * Italian Snowball stemming over the Radixor Italian dictionary.
+     */
+    ITALIAN("Italian", StemmerPatchTrieLoader.Language.IT_IT, italianStemmer::new, "Italian"),
+
+    /**
+     * Norwegian Snowball stemming over the Radixor Bokmal dictionary.
+     */
+    NORWEGIAN_BOKMAL("Norwegian Bokmal", StemmerPatchTrieLoader.Language.NB_NO, norwegianStemmer::new,
+            "Norwegian"),
+
+    /**
+     * Norwegian Snowball stemming over the Radixor Nynorsk dictionary.
+     */
+    NORWEGIAN_NYNORSK("Norwegian Nynorsk", StemmerPatchTrieLoader.Language.NN_NO, norwegianStemmer::new,
+            "Norwegian"),
+
+    /**
+     * Portuguese Snowball stemming over the Radixor Portuguese dictionary.
+     */
+    PORTUGUESE("Portuguese", StemmerPatchTrieLoader.Language.PT_PT, portugueseStemmer::new, "Portuguese"),
+
+    /**
+     * Russian Snowball stemming over the Radixor Russian dictionary.
+     */
+    RUSSIAN("Russian", StemmerPatchTrieLoader.Language.RU_RU, russianStemmer::new, "Russian"),
+
+    /**
+     * Spanish Snowball stemming over the Radixor Spanish dictionary.
+     */
+    SPANISH("Spanish", StemmerPatchTrieLoader.Language.ES_ES, spanishStemmer::new, "Spanish"),
+
+    /**
+     * Swedish Snowball stemming over the Radixor Swedish dictionary.
+     */
+    SWEDISH("Swedish", StemmerPatchTrieLoader.Language.SV_SE, swedishStemmer::new, "Swedish"),
+
+    /**
+     * Yiddish Snowball stemming over the Radixor Yiddish dictionary.
+     */
+    YIDDISH("Yiddish", StemmerPatchTrieLoader.Language.YI, yiddishStemmer::new, "Yiddish");
+
+    /**
+     * Human-readable language name.
+     */
+    private final String displayLanguage;
+
+    /**
+     * Matching Radixor language resource.
+     */
+    private final StemmerPatchTrieLoader.Language radixorLanguage;
+
+    /**
+     * Factory for the isolated benchmark-only Snowball implementation.
+     */
+    private final SnowballStemmerAdapter.Factory directFactory;
+
+    /**
+     * Lucene SnowballFilter algorithm name.
+     */
+    private final String luceneSnowballName;
+
+    /**
+     * Creates a language case.
+     *
+     * @param displayLanguage   human-readable language name
+     * @param radixorLanguage   matching Radixor language resource
+     * @param directFactory     direct Snowball stemmer factory
+     * @param luceneSnowballName Lucene SnowballFilter algorithm name
+     */
+    SnowballLanguageCase(final String displayLanguage, final StemmerPatchTrieLoader.Language radixorLanguage,
+            final SnowballStemmerAdapter.Factory directFactory, final String luceneSnowballName) {
+        this.displayLanguage = displayLanguage;
+        this.radixorLanguage = radixorLanguage;
+        this.directFactory = directFactory;
+        this.luceneSnowballName = luceneSnowballName;
+    }
+
+    /**
+     * Returns the human-readable language name.
+     *
+     * @return display language
+     */
+    String displayLanguage() {
+        return this.displayLanguage;
+    }
+
+    /**
+     * Returns the matching Radixor dictionary language.
+     *
+     * @return Radixor language
+     */
+    StemmerPatchTrieLoader.Language radixorLanguage() {
+        return this.radixorLanguage;
+    }
+
+    /**
+     * Creates a direct Snowball stemmer adapter.
+     *
+     * @return direct Snowball adapter
+     */
+    SnowballStemmerAdapter createDirectStemmer() {
+        return new SnowballStemmerAdapter(this.directFactory);
+    }
+
+    /**
+     * Returns the Lucene SnowballFilter algorithm name.
+     *
+     * @return Lucene SnowballFilter algorithm name
+     */
+    String luceneSnowballName() {
+        return this.luceneSnowballName;
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/SnowballLanguageStemmerComparisonBenchmark.java

@@ -0,0 +1,246 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.IOException;
+import java.util.concurrent.TimeUnit;
+
+import org.apache.lucene.analysis.LowerCaseFilter;
+import org.apache.lucene.analysis.TokenStream;
+import org.apache.lucene.analysis.snowball.SnowballFilter;
+import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Level;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+import org.openjdk.jmh.annotations.Param;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.Setup;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.Warmup;
+import org.openjdk.jmh.infra.Blackhole;
+
+/**
+ * Compares Radixor with official Snowball algorithms for every Radixor language
+ * that has a matching Snowball Java stemmer.
+ *
+ * <p>
+ * Each benchmark operation processes the same changed-token Radixor
+ * dictionary-derived language corpus, repeated only when the changed-token
+ * resource contains fewer than 5,000 token fields. The direct Snowball method
+ * measures the isolated benchmark-only Snowball source. The Lucene
+ * SnowballFilter method measures Lucene's TokenStream integration path,
+ * including lower-case normalization and token attribute overhead.
+ * </p>
+ */
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.NANOSECONDS)
+@Warmup(iterations = 3, time = 1, timeUnit = TimeUnit.SECONDS)
+@Measurement(iterations = 5, time = 1, timeUnit = TimeUnit.SECONDS)
+public class SnowballLanguageStemmerComparisonBenchmark {
+
+    /**
+     * Shared language corpus and Radixor trie state.
+     */
+    @State(Scope.Benchmark)
+    public static class SharedState {
+
+        /**
+         * Language/algorithm case under comparison.
+         */
+        @Param({ "DANISH", "DUTCH", "FINNISH", "FRENCH", "GERMAN", "HUNGARIAN", "ITALIAN",
+                "NORWEGIAN_BOKMAL", "NORWEGIAN_NYNORSK", "PORTUGUESE", "RUSSIAN", "SPANISH", "SWEDISH",
+                "YIDDISH" })
+        public String languageCaseName;
+
+        /**
+         * Resolved language/algorithm case.
+         */
+        private SnowballLanguageCase languageCase;
+
+        /**
+         * Shared deterministic changed-token dictionary corpus.
+         */
+        private String[] tokens;
+
+        /**
+         * Compiled Radixor trie for the selected language.
+         */
+        private RadixorBenchmarkStemmer radixorStemmer;
+
+        /**
+         * Initializes shared language resources before measurement.
+         *
+         * @throws IOException if the corpus or trie cannot be loaded
+         */
+        @Setup(Level.Trial)
+        public void setUp() throws IOException {
+            this.languageCase = SnowballLanguageCase.valueOf(this.languageCaseName);
+            this.tokens = LanguageBenchmarkCorpus.createTokens(this.languageCase.radixorLanguage());
+            this.radixorStemmer = new RadixorBenchmarkStemmer(StemmerPatchTrieLoader.loadCompiled(
+                    this.languageCase.radixorLanguage(), true,
+                    ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS));
+        }
+    }
+
+    /**
+     * Per-thread direct Snowball state.
+     */
+    @State(Scope.Thread)
+    public static class DirectState {
+
+        /**
+         * Reusable direct Snowball adapter.
+         */
+        private SnowballStemmerAdapter snowballStemmer;
+
+        /**
+         * Initializes direct Snowball state for the selected language.
+         *
+         * @param sharedState selected language state
+         */
+        @Setup(Level.Trial)
+        public void setUp(final SharedState sharedState) {
+            this.snowballStemmer = sharedState.languageCase.createDirectStemmer();
+        }
+    }
+
+    /**
+     * Per-thread Lucene SnowballFilter state.
+     */
+    @State(Scope.Thread)
+    public static class LuceneSnowballState {
+
+        /**
+         * Reusable benchmark input stream.
+         */
+        private BenchmarkTokenStream input;
+
+        /**
+         * Reusable Lucene SnowballFilter output stream.
+         */
+        private TokenStream output;
+
+        /**
+         * Reusable term attribute.
+         */
+        private CharTermAttribute termAttribute;
+
+        /**
+         * Initializes Lucene SnowballFilter state for the selected language.
+         *
+         * @param sharedState selected language state
+         */
+        @Setup(Level.Trial)
+        public void setUp(final SharedState sharedState) {
+            this.input = new BenchmarkTokenStream(new String[0]);
+            final TokenStream normalizedInput = new LowerCaseFilter(this.input);
+            this.output = new SnowballFilter(normalizedInput, sharedState.languageCase.luceneSnowballName());
+            this.termAttribute = this.output.addAttribute(CharTermAttribute.class);
+        }
+
+        /**
+         * Runs the reusable Lucene SnowballFilter over one corpus.
+         *
+         * <p>
+         * The benchmark intentionally rebinds the {@code String[]} corpus on every
+         * measured operation so the adaptation cost from the canonical string input to
+         * Lucene's mutable character attributes is included.
+         * </p>
+         *
+         * @param tokens    token corpus
+         * @param blackhole result sink
+         * @throws IOException if Lucene token streaming fails
+         */
+        void run(final String[] tokens, final Blackhole blackhole) throws IOException {
+            this.input.setTokens(tokens);
+            this.output.reset();
+            while (this.output.incrementToken()) {
+                blackhole.consume(this.termAttribute.toString());
+            }
+            this.output.end();
+        }
+    }
+
+    /**
+     * Runs Radixor over the selected Snowball-language corpus.
+     *
+     * @param sharedState shared benchmark state
+     * @param blackhole   result sink
+     */
+    @Benchmark
+    public void radixor(final SharedState sharedState, final Blackhole blackhole) {
+        final String[] tokens = sharedState.tokens;
+        final RadixorBenchmarkStemmer stemmer = sharedState.radixorStemmer;
+
+        for (String token : tokens) {
+            blackhole.consume(stemmer.stem(token));
+        }
+    }
+
+    /**
+     * Runs the official Snowball direct Java implementation over the selected
+     * language corpus.
+     *
+     * @param sharedState shared benchmark state
+     * @param directState reusable direct Snowball state
+     * @param blackhole   result sink
+     */
+    @Benchmark
+    public void snowballDirect(final SharedState sharedState, final DirectState directState,
+            final Blackhole blackhole) {
+        final String[] tokens = sharedState.tokens;
+        final SnowballStemmerAdapter stemmer = directState.snowballStemmer;
+
+        for (String token : tokens) {
+            blackhole.consume(stemmer.stem(token));
+        }
+    }
+
+    /**
+     * Runs Lucene SnowballFilter over the selected language corpus.
+     *
+     * @param sharedState shared benchmark state
+     * @param luceneState reusable Lucene Snowball state
+     * @param blackhole   result sink
+     * @throws IOException if Lucene token streaming fails
+     */
+    @Benchmark
+    public void luceneSnowballFilter(final SharedState sharedState, final LuceneSnowballState luceneState,
+            final Blackhole blackhole) throws IOException {
+        luceneState.run(sharedState.tokens, blackhole);
+    }
+}

src/jmh/java/org/egothor/stemmer/benchmark/SnowballStemmerAdapter.java

@@ -32,7 +32,7 @@ package org.egothor.stemmer.benchmark;
 
 import java.util.Objects;
 
-import org.tartarus.snowball.SnowballStemmer;
+import org.egothor.stemmer.benchmark.snowball.SnowballStemmer;
 
 /**
  * Small adapter around a Snowball stemmer instance used by benchmarks.

src/jmh/java/org/egothor/stemmer/benchmark/StemmerComparisonBenchmarkQuality.java

@@ -0,0 +1,923 @@
+/*******************************************************************************
+ * Copyright (C) 2026, Leo Galambos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+package org.egothor.stemmer.benchmark;
+
+import java.io.IOException;
+import java.net.URL;
+import java.util.List;
+import java.util.Objects;
+import java.util.concurrent.TimeUnit;
+import java.util.function.Function;
+
+import org.apache.lucene.analysis.LowerCaseFilter;
+import org.apache.lucene.analysis.TokenStream;
+import org.apache.lucene.analysis.ar.ArabicNormalizationFilter;
+import org.apache.lucene.analysis.core.DecimalDigitFilter;
+import org.apache.lucene.analysis.cz.CzechStemFilter;
+import org.apache.lucene.analysis.de.GermanLightStemFilter;
+import org.apache.lucene.analysis.de.GermanMinimalStemFilter;
+import org.apache.lucene.analysis.de.GermanNormalizationFilter;
+import org.apache.lucene.analysis.de.GermanStemFilter;
+import org.apache.lucene.analysis.en.EnglishMinimalStemFilter;
+import org.apache.lucene.analysis.en.EnglishPossessiveFilter;
+import org.apache.lucene.analysis.en.KStemFilter;
+import org.apache.lucene.analysis.en.PorterStemFilter;
+import org.apache.lucene.analysis.es.SpanishLightStemFilter;
+import org.apache.lucene.analysis.es.SpanishMinimalStemFilter;
+import org.apache.lucene.analysis.es.SpanishPluralStemFilter;
+import org.apache.lucene.analysis.fa.PersianNormalizationFilter;
+import org.apache.lucene.analysis.fa.PersianStemFilter;
+import org.apache.lucene.analysis.fi.FinnishLightStemFilter;
+import org.apache.lucene.analysis.fr.FrenchLightStemFilter;
+import org.apache.lucene.analysis.fr.FrenchMinimalStemFilter;
+import org.apache.lucene.analysis.hu.HungarianLightStemFilter;
+import org.apache.lucene.analysis.it.ItalianLightStemFilter;
+import org.apache.lucene.analysis.morfologik.MorfologikFilter;
+import org.apache.lucene.analysis.no.NorwegianLightStemFilter;
+import org.apache.lucene.analysis.no.NorwegianMinimalStemFilter;
+import org.apache.lucene.analysis.pl.PolishAnalyzer;
+import org.apache.lucene.analysis.pt.PortugueseLightStemFilter;
+import org.apache.lucene.analysis.pt.PortugueseMinimalStemFilter;
+import org.apache.lucene.analysis.pt.PortugueseStemFilter;
+import org.apache.lucene.analysis.ru.RussianLightStemFilter;
+import org.apache.lucene.analysis.snowball.SnowballFilter;
+import org.apache.lucene.analysis.stempel.StempelFilter;
+import org.apache.lucene.analysis.stempel.StempelStemmer;
+import org.apache.lucene.analysis.sv.SwedishLightStemFilter;
+import org.apache.lucene.analysis.sv.SwedishMinimalStemFilter;
+import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
+import org.apache.lucene.analysis.tokenattributes.PositionIncrementAttribute;
+import org.egothor.stemmer.FrequencyTrie;
+import org.egothor.stemmer.ReductionMode;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
+import org.egothor.stemmer.benchmark.snowball.ext.englishStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.porterStemmer;
+import org.openjdk.jmh.annotations.AuxCounters;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Fork;
+import org.openjdk.jmh.annotations.Level;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+import org.openjdk.jmh.annotations.Param;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.Setup;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.Warmup;
+import org.openjdk.jmh.infra.Blackhole;
+
+import morfologik.stemming.Dictionary;
+import morfologik.stemming.DictionaryLookup;
+import morfologik.stemming.WordData;
+
+/**
+ * Emits exact-root agreement metrics through standard JMH result files.
+ *
+ * <p>
+ * This benchmark is a quality pass, not a throughput competitor. Each operation
+ * evaluates one stemmer against the complete Radixor dictionary resource for
+ * the matching language. The useful outputs are the JMH auxiliary counters
+ * {@code correctMatches}, {@code evaluatedTokens},
+ * {@code changedCorrectMatches}, {@code changedEvaluatedTokens},
+ * {@code rootPreservedMatches}, and {@code rootEvaluatedTokens}.
+ * </p>
+ */
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.NANOSECONDS)
+@Warmup(iterations = 0)
+@Measurement(iterations = 1, time = 1, timeUnit = TimeUnit.MILLISECONDS)
+@Fork(0)
+public class StemmerComparisonBenchmarkQuality {
+
+    /**
+     * Shared quality state for one candidate stemmer.
+     */
+    @State(Scope.Benchmark)
+    public static class QualityState {
+
+        /**
+         * Candidate stemmer whose exact-root agreement is measured.
+         */
+        @Param({
+                "ENGLISH_RADIXOR",
+                "ENGLISH_SNOWBALL_ORIGINAL_PORTER",
+                "ENGLISH_SNOWBALL_PORTER2",
+                "ENGLISH_LUCENE_PORTER_COPIED",
+                "ENGLISH_LUCENE_PORTER_FILTER",
+                "ENGLISH_LUCENE_KSTEM_FILTER",
+                "ENGLISH_LUCENE_MINIMAL_FILTER",
+                "ENGLISH_LUCENE_POSSESSIVE_FILTER",
+                "ENGLISH_PAICE_HUSK_LANCASTER",
+                "ENGLISH_OPENNLP_PORTER",
+                "CZECH_RADIXOR",
+                "CZECH_LUCENE_CZECH_STEM_FILTER",
+                "GERMAN_RADIXOR",
+                "GERMAN_LUCENE_GERMAN_STEM_FILTER",
+                "GERMAN_LUCENE_GERMAN_LIGHT_STEM_FILTER",
+                "GERMAN_LUCENE_GERMAN_MINIMAL_STEM_FILTER",
+                "GERMAN_CISTEM",
+                "SPANISH_RADIXOR",
+                "SPANISH_LUCENE_SPANISH_LIGHT_STEM_FILTER",
+                "SPANISH_LUCENE_SPANISH_MINIMAL_STEM_FILTER",
+                "SPANISH_LUCENE_SPANISH_PLURAL_STEM_FILTER",
+                "PERSIAN_RADIXOR",
+                "PERSIAN_LUCENE_PERSIAN_STEM_FILTER",
+                "FINNISH_RADIXOR",
+                "FINNISH_LUCENE_FINNISH_LIGHT_STEM_FILTER",
+                "FRENCH_RADIXOR",
+                "FRENCH_LUCENE_FRENCH_LIGHT_STEM_FILTER",
+                "FRENCH_LUCENE_FRENCH_MINIMAL_STEM_FILTER",
+                "HUNGARIAN_RADIXOR",
+                "HUNGARIAN_LUCENE_HUNGARIAN_LIGHT_STEM_FILTER",
+                "ITALIAN_RADIXOR",
+                "ITALIAN_LUCENE_ITALIAN_LIGHT_STEM_FILTER",
+                "NORWEGIAN_BOKMAL_RADIXOR",
+                "NORWEGIAN_BOKMAL_LUCENE_NORWEGIAN_LIGHT_STEM_FILTER",
+                "NORWEGIAN_BOKMAL_LUCENE_NORWEGIAN_MINIMAL_STEM_FILTER",
+                "POLISH_RADIXOR",
+                "POLISH_LUCENE_STEMPEL_DIRECT",
+                "POLISH_LUCENE_STEMPEL_FILTER",
+                "POLISH_LUCENE_MORFOLOGIK_FILTER",
+                "PORTUGUESE_RADIXOR",
+                "PORTUGUESE_LUCENE_PORTUGUESE_STEM_FILTER",
+                "PORTUGUESE_LUCENE_PORTUGUESE_LIGHT_STEM_FILTER",
+                "PORTUGUESE_LUCENE_PORTUGUESE_MINIMAL_STEM_FILTER",
+                "RUSSIAN_RADIXOR",
+                "RUSSIAN_LUCENE_RUSSIAN_LIGHT_STEM_FILTER",
+                "SWEDISH_RADIXOR",
+                "SWEDISH_LUCENE_SWEDISH_LIGHT_STEM_FILTER",
+                "SWEDISH_LUCENE_SWEDISH_MINIMAL_STEM_FILTER",
+                "UKRAINIAN_RADIXOR",
+                "UKRAINIAN_MORFOLOGIK_DIRECT",
+                "UKRAINIAN_LUCENE_MORFOLOGIK_FILTER",
+                "SNOWBALL_DANISH_DIRECT",
+                "SNOWBALL_DANISH_LUCENE_FILTER",
+                "SNOWBALL_DUTCH_DIRECT",
+                "SNOWBALL_DUTCH_LUCENE_FILTER",
+                "SNOWBALL_FINNISH_DIRECT",
+                "SNOWBALL_FINNISH_LUCENE_FILTER",
+                "SNOWBALL_FRENCH_DIRECT",
+                "SNOWBALL_FRENCH_LUCENE_FILTER",
+                "SNOWBALL_GERMAN_DIRECT",
+                "SNOWBALL_GERMAN_LUCENE_FILTER",
+                "SNOWBALL_HUNGARIAN_DIRECT",
+                "SNOWBALL_HUNGARIAN_LUCENE_FILTER",
+                "SNOWBALL_ITALIAN_DIRECT",
+                "SNOWBALL_ITALIAN_LUCENE_FILTER",
+                "SNOWBALL_NORWEGIAN_BOKMAL_DIRECT",
+                "SNOWBALL_NORWEGIAN_BOKMAL_LUCENE_FILTER",
+                "SNOWBALL_NORWEGIAN_NYNORSK_DIRECT",
+                "SNOWBALL_NORWEGIAN_NYNORSK_LUCENE_FILTER",
+                "SNOWBALL_PORTUGUESE_DIRECT",
+                "SNOWBALL_PORTUGUESE_LUCENE_FILTER",
+                "SNOWBALL_RUSSIAN_DIRECT",
+                "SNOWBALL_RUSSIAN_LUCENE_FILTER",
+                "SNOWBALL_SPANISH_DIRECT",
+                "SNOWBALL_SPANISH_LUCENE_FILTER",
+                "SNOWBALL_SWEDISH_DIRECT",
+                "SNOWBALL_SWEDISH_LUCENE_FILTER",
+                "SNOWBALL_YIDDISH_DIRECT",
+                "SNOWBALL_YIDDISH_LUCENE_FILTER"
+        })
+        public String candidateName;
+
+        /**
+         * Full dictionary corpus for the selected language.
+         */
+        private LanguageBenchmarkCorpus.Corpus corpus;
+
+        /**
+         * Candidate evaluator.
+         */
+        private QualityEvaluator evaluator;
+
+        /**
+         * Initializes corpus and evaluator before measurement.
+         *
+         * @throws IOException if dictionary or stemmer resources cannot be loaded
+         */
+        @Setup(Level.Trial)
+        public void setUp() throws IOException {
+            final QualityCandidate candidate = QualityCandidate.valueOf(this.candidateName);
+            this.corpus = LanguageBenchmarkCorpus.createFullCorpus(candidate.radixorLanguage());
+            this.evaluator = candidate.createEvaluator();
+        }
+    }
+
+    /**
+     * JMH auxiliary counters for exact-root agreement.
+     */
+    @State(Scope.Thread)
+    @AuxCounters(AuxCounters.Type.EVENTS)
+    public static class QualityCounters {
+
+        /**
+         * Number of outputs equal to the dictionary root.
+         */
+        public long correctMatches;
+
+        /**
+         * Number of evaluated input tokens.
+         */
+        public long evaluatedTokens;
+
+        /**
+         * Number of exact-root matches where the input token differs from the
+         * expected root.
+         */
+        public long changedCorrectMatches;
+
+        /**
+         * Number of evaluated tokens where the input token differs from the expected
+         * root.
+         */
+        public long changedEvaluatedTokens;
+
+        /**
+         * Number of exact-root matches where the input token is already the expected
+         * root.
+         */
+        public long rootPreservedMatches;
+
+        /**
+         * Number of evaluated tokens where the input token is already the expected
+         * root.
+         */
+        public long rootEvaluatedTokens;
+
+        /**
+         * Resets counters before each measured iteration.
+         */
+        @Setup(Level.Iteration)
+        public void reset() {
+            this.correctMatches = 0L;
+            this.evaluatedTokens = 0L;
+            this.changedCorrectMatches = 0L;
+            this.changedEvaluatedTokens = 0L;
+            this.rootPreservedMatches = 0L;
+            this.rootEvaluatedTokens = 0L;
+        }
+    }
+
+    /**
+     * Runs exact-root agreement over the full dictionary corpus.
+     *
+     * @param state quality state
+     * @param counters auxiliary JMH counters
+     * @param blackhole result sink
+     * @return exact-root match count for this operation
+     * @throws IOException if Lucene streaming fails
+     */
+    @Benchmark
+    public int exactRootAgreement(final QualityState state, final QualityCounters counters, final Blackhole blackhole)
+            throws IOException {
+        final QualityResult result = state.evaluator.evaluate(state.corpus, blackhole);
+        counters.correctMatches += result.correctMatches();
+        counters.evaluatedTokens += result.evaluatedTokens();
+        counters.changedCorrectMatches += result.changedCorrectMatches();
+        counters.changedEvaluatedTokens += result.changedEvaluatedTokens();
+        counters.rootPreservedMatches += result.rootPreservedMatches();
+        counters.rootEvaluatedTokens += result.rootEvaluatedTokens();
+        return result.correctMatches();
+    }
+
+    /**
+     * Candidate stemmers that can be evaluated against a Radixor resource.
+     */
+    enum QualityCandidate {
+        ENGLISH_RADIXOR(StemmerPatchTrieLoader.Language.US_UK),
+        ENGLISH_SNOWBALL_ORIGINAL_PORTER(StemmerPatchTrieLoader.Language.US_UK),
+        ENGLISH_SNOWBALL_PORTER2(StemmerPatchTrieLoader.Language.US_UK),
+        ENGLISH_LUCENE_PORTER_COPIED(StemmerPatchTrieLoader.Language.US_UK),
+        ENGLISH_LUCENE_PORTER_FILTER(StemmerPatchTrieLoader.Language.US_UK),
+        ENGLISH_LUCENE_KSTEM_FILTER(StemmerPatchTrieLoader.Language.US_UK),
+        ENGLISH_LUCENE_MINIMAL_FILTER(StemmerPatchTrieLoader.Language.US_UK),
+        ENGLISH_LUCENE_POSSESSIVE_FILTER(StemmerPatchTrieLoader.Language.US_UK),
+        ENGLISH_PAICE_HUSK_LANCASTER(StemmerPatchTrieLoader.Language.US_UK),
+        ENGLISH_OPENNLP_PORTER(StemmerPatchTrieLoader.Language.US_UK),
+        CZECH_RADIXOR(StemmerPatchTrieLoader.Language.CS_CZ),
+        CZECH_LUCENE_CZECH_STEM_FILTER(StemmerPatchTrieLoader.Language.CS_CZ),
+        GERMAN_RADIXOR(StemmerPatchTrieLoader.Language.DE_DE),
+        GERMAN_LUCENE_GERMAN_STEM_FILTER(StemmerPatchTrieLoader.Language.DE_DE),
+        GERMAN_LUCENE_GERMAN_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.DE_DE),
+        GERMAN_LUCENE_GERMAN_MINIMAL_STEM_FILTER(StemmerPatchTrieLoader.Language.DE_DE),
+        GERMAN_CISTEM(StemmerPatchTrieLoader.Language.DE_DE),
+        SPANISH_RADIXOR(StemmerPatchTrieLoader.Language.ES_ES),
+        SPANISH_LUCENE_SPANISH_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.ES_ES),
+        SPANISH_LUCENE_SPANISH_MINIMAL_STEM_FILTER(StemmerPatchTrieLoader.Language.ES_ES),
+        SPANISH_LUCENE_SPANISH_PLURAL_STEM_FILTER(StemmerPatchTrieLoader.Language.ES_ES),
+        PERSIAN_RADIXOR(StemmerPatchTrieLoader.Language.FA_IR),
+        PERSIAN_LUCENE_PERSIAN_STEM_FILTER(StemmerPatchTrieLoader.Language.FA_IR),
+        FINNISH_RADIXOR(StemmerPatchTrieLoader.Language.FI_FI),
+        FINNISH_LUCENE_FINNISH_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.FI_FI),
+        FRENCH_RADIXOR(StemmerPatchTrieLoader.Language.FR_FR),
+        FRENCH_LUCENE_FRENCH_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.FR_FR),
+        FRENCH_LUCENE_FRENCH_MINIMAL_STEM_FILTER(StemmerPatchTrieLoader.Language.FR_FR),
+        HUNGARIAN_RADIXOR(StemmerPatchTrieLoader.Language.HU_HU),
+        HUNGARIAN_LUCENE_HUNGARIAN_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.HU_HU),
+        ITALIAN_RADIXOR(StemmerPatchTrieLoader.Language.IT_IT),
+        ITALIAN_LUCENE_ITALIAN_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.IT_IT),
+        NORWEGIAN_BOKMAL_RADIXOR(StemmerPatchTrieLoader.Language.NB_NO),
+        NORWEGIAN_BOKMAL_LUCENE_NORWEGIAN_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.NB_NO),
+        NORWEGIAN_BOKMAL_LUCENE_NORWEGIAN_MINIMAL_STEM_FILTER(StemmerPatchTrieLoader.Language.NB_NO),
+        POLISH_RADIXOR(StemmerPatchTrieLoader.Language.PL_PL),
+        POLISH_LUCENE_STEMPEL_DIRECT(StemmerPatchTrieLoader.Language.PL_PL),
+        POLISH_LUCENE_STEMPEL_FILTER(StemmerPatchTrieLoader.Language.PL_PL),
+        POLISH_LUCENE_MORFOLOGIK_FILTER(StemmerPatchTrieLoader.Language.PL_PL),
+        PORTUGUESE_RADIXOR(StemmerPatchTrieLoader.Language.PT_PT),
+        PORTUGUESE_LUCENE_PORTUGUESE_STEM_FILTER(StemmerPatchTrieLoader.Language.PT_PT),
+        PORTUGUESE_LUCENE_PORTUGUESE_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.PT_PT),
+        PORTUGUESE_LUCENE_PORTUGUESE_MINIMAL_STEM_FILTER(StemmerPatchTrieLoader.Language.PT_PT),
+        RUSSIAN_RADIXOR(StemmerPatchTrieLoader.Language.RU_RU),
+        RUSSIAN_LUCENE_RUSSIAN_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.RU_RU),
+        SWEDISH_RADIXOR(StemmerPatchTrieLoader.Language.SV_SE),
+        SWEDISH_LUCENE_SWEDISH_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.SV_SE),
+        SWEDISH_LUCENE_SWEDISH_MINIMAL_STEM_FILTER(StemmerPatchTrieLoader.Language.SV_SE),
+        UKRAINIAN_RADIXOR(StemmerPatchTrieLoader.Language.UK_UA),
+        UKRAINIAN_MORFOLOGIK_DIRECT(StemmerPatchTrieLoader.Language.UK_UA),
+        UKRAINIAN_LUCENE_MORFOLOGIK_FILTER(StemmerPatchTrieLoader.Language.UK_UA),
+        SNOWBALL_DANISH_DIRECT(StemmerPatchTrieLoader.Language.DA_DK, SnowballLanguageCase.DANISH),
+        SNOWBALL_DANISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.DA_DK, SnowballLanguageCase.DANISH),
+        SNOWBALL_DUTCH_DIRECT(StemmerPatchTrieLoader.Language.NL_NL, SnowballLanguageCase.DUTCH),
+        SNOWBALL_DUTCH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.NL_NL, SnowballLanguageCase.DUTCH),
+        SNOWBALL_FINNISH_DIRECT(StemmerPatchTrieLoader.Language.FI_FI, SnowballLanguageCase.FINNISH),
+        SNOWBALL_FINNISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.FI_FI, SnowballLanguageCase.FINNISH),
+        SNOWBALL_FRENCH_DIRECT(StemmerPatchTrieLoader.Language.FR_FR, SnowballLanguageCase.FRENCH),
+        SNOWBALL_FRENCH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.FR_FR, SnowballLanguageCase.FRENCH),
+        SNOWBALL_GERMAN_DIRECT(StemmerPatchTrieLoader.Language.DE_DE, SnowballLanguageCase.GERMAN),
+        SNOWBALL_GERMAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.DE_DE, SnowballLanguageCase.GERMAN),
+        SNOWBALL_HUNGARIAN_DIRECT(StemmerPatchTrieLoader.Language.HU_HU, SnowballLanguageCase.HUNGARIAN),
+        SNOWBALL_HUNGARIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.HU_HU, SnowballLanguageCase.HUNGARIAN),
+        SNOWBALL_ITALIAN_DIRECT(StemmerPatchTrieLoader.Language.IT_IT, SnowballLanguageCase.ITALIAN),
+        SNOWBALL_ITALIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.IT_IT, SnowballLanguageCase.ITALIAN),
+        SNOWBALL_NORWEGIAN_BOKMAL_DIRECT(StemmerPatchTrieLoader.Language.NB_NO,
+                SnowballLanguageCase.NORWEGIAN_BOKMAL),
+        SNOWBALL_NORWEGIAN_BOKMAL_LUCENE_FILTER(StemmerPatchTrieLoader.Language.NB_NO,
+                SnowballLanguageCase.NORWEGIAN_BOKMAL),
+        SNOWBALL_NORWEGIAN_NYNORSK_DIRECT(StemmerPatchTrieLoader.Language.NN_NO,
+                SnowballLanguageCase.NORWEGIAN_NYNORSK),
+        SNOWBALL_NORWEGIAN_NYNORSK_LUCENE_FILTER(StemmerPatchTrieLoader.Language.NN_NO,
+                SnowballLanguageCase.NORWEGIAN_NYNORSK),
+        SNOWBALL_PORTUGUESE_DIRECT(StemmerPatchTrieLoader.Language.PT_PT, SnowballLanguageCase.PORTUGUESE),
+        SNOWBALL_PORTUGUESE_LUCENE_FILTER(StemmerPatchTrieLoader.Language.PT_PT, SnowballLanguageCase.PORTUGUESE),
+        SNOWBALL_RUSSIAN_DIRECT(StemmerPatchTrieLoader.Language.RU_RU, SnowballLanguageCase.RUSSIAN),
+        SNOWBALL_RUSSIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.RU_RU, SnowballLanguageCase.RUSSIAN),
+        SNOWBALL_SPANISH_DIRECT(StemmerPatchTrieLoader.Language.ES_ES, SnowballLanguageCase.SPANISH),
+        SNOWBALL_SPANISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.ES_ES, SnowballLanguageCase.SPANISH),
+        SNOWBALL_SWEDISH_DIRECT(StemmerPatchTrieLoader.Language.SV_SE, SnowballLanguageCase.SWEDISH),
+        SNOWBALL_SWEDISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.SV_SE, SnowballLanguageCase.SWEDISH),
+        SNOWBALL_YIDDISH_DIRECT(StemmerPatchTrieLoader.Language.YI, SnowballLanguageCase.YIDDISH),
+        SNOWBALL_YIDDISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.YI, SnowballLanguageCase.YIDDISH);
+
+        /**
+         * Radixor dictionary language used as truth.
+         */
+        private final StemmerPatchTrieLoader.Language radixorLanguage;
+
+        /**
+         * Optional Snowball language mapping.
+         */
+        private final SnowballLanguageCase snowballLanguageCase;
+
+        /**
+         * Creates a candidate.
+         *
+         * @param radixorLanguage Radixor dictionary language
+         */
+        QualityCandidate(final StemmerPatchTrieLoader.Language radixorLanguage) {
+            this(radixorLanguage, null);
+        }
+
+        /**
+         * Creates a candidate.
+         *
+         * @param radixorLanguage Radixor dictionary language
+         * @param snowballLanguageCase matching Snowball case
+         */
+        QualityCandidate(final StemmerPatchTrieLoader.Language radixorLanguage,
+                final SnowballLanguageCase snowballLanguageCase) {
+            this.radixorLanguage = radixorLanguage;
+            this.snowballLanguageCase = snowballLanguageCase;
+        }
+
+        /**
+         * Returns the Radixor dictionary language.
+         *
+         * @return Radixor language
+         */
+        StemmerPatchTrieLoader.Language radixorLanguage() {
+            return this.radixorLanguage;
+        }
+
+        /**
+         * Creates the evaluator for this candidate.
+         *
+         * @return quality evaluator
+         * @throws IOException if stemmer resources cannot be loaded
+         */
+        CandidateStemmer createStemmer() throws IOException {
+            if (name().endsWith("_RADIXOR")) {
+                return radixor(createRadixorStemmer(this.radixorLanguage));
+            }
+            if (name().endsWith("_DIRECT") && this.snowballLanguageCase != null) {
+                return direct(this.snowballLanguageCase.createDirectStemmer()::stem);
+            }
+            if (name().endsWith("_LUCENE_FILTER") && this.snowballLanguageCase != null) {
+                return tokenFilter(input -> new SnowballFilter(new LowerCaseFilter(input),
+                        this.snowballLanguageCase.luceneSnowballName()));
+            }
+
+            return switch (this) {
+                case ENGLISH_SNOWBALL_ORIGINAL_PORTER -> direct(new SnowballStemmerAdapter(porterStemmer::new)::stem);
+                case ENGLISH_SNOWBALL_PORTER2 -> direct(new SnowballStemmerAdapter(englishStemmer::new)::stem);
+                case ENGLISH_LUCENE_PORTER_COPIED -> direct(new LucenePorterStemmerCopied()::stem);
+                case ENGLISH_LUCENE_PORTER_FILTER -> tokenFilter(PorterStemFilter::new);
+                case ENGLISH_LUCENE_KSTEM_FILTER -> tokenFilter(KStemFilter::new);
+                case ENGLISH_LUCENE_MINIMAL_FILTER -> tokenFilter(EnglishMinimalStemFilter::new);
+                case ENGLISH_LUCENE_POSSESSIVE_FILTER -> tokenFilter(EnglishPossessiveFilter::new);
+                case ENGLISH_PAICE_HUSK_LANCASTER -> direct(new PaiceHuskLancasterStemmer()::stem);
+                case ENGLISH_OPENNLP_PORTER -> {
+                    final opennlp.tools.stemmer.PorterStemmer stemmer =
+                            new opennlp.tools.stemmer.PorterStemmer();
+                    yield direct(token -> stemmer.stem(token).toString());
+                }
+                case CZECH_LUCENE_CZECH_STEM_FILTER -> tokenFilter(input -> new CzechStemFilter(lowercase(input)));
+                case GERMAN_LUCENE_GERMAN_STEM_FILTER -> tokenFilter(input -> new GermanStemFilter(lowercase(input)));
+                case GERMAN_LUCENE_GERMAN_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new GermanLightStemFilter(germanNormalize(input)));
+                case GERMAN_LUCENE_GERMAN_MINIMAL_STEM_FILTER ->
+                    tokenFilter(input -> new GermanMinimalStemFilter(germanNormalize(input)));
+                case GERMAN_CISTEM -> direct(createGermanCistemStemmer());
+                case SPANISH_LUCENE_SPANISH_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new SpanishLightStemFilter(lowercase(input)));
+                case SPANISH_LUCENE_SPANISH_MINIMAL_STEM_FILTER ->
+                    tokenFilter(input -> new SpanishMinimalStemFilter(lowercase(input)));
+                case SPANISH_LUCENE_SPANISH_PLURAL_STEM_FILTER ->
+                    tokenFilter(input -> new SpanishPluralStemFilter(lowercase(input)));
+                case PERSIAN_LUCENE_PERSIAN_STEM_FILTER ->
+                    tokenFilter(input -> new PersianStemFilter(persianNormalize(input)));
+                case FINNISH_LUCENE_FINNISH_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new FinnishLightStemFilter(lowercase(input)));
+                case FRENCH_LUCENE_FRENCH_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new FrenchLightStemFilter(lowercase(input)));
+                case FRENCH_LUCENE_FRENCH_MINIMAL_STEM_FILTER ->
+                    tokenFilter(input -> new FrenchMinimalStemFilter(lowercase(input)));
+                case HUNGARIAN_LUCENE_HUNGARIAN_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new HungarianLightStemFilter(lowercase(input)));
+                case ITALIAN_LUCENE_ITALIAN_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new ItalianLightStemFilter(lowercase(input)));
+                case NORWEGIAN_BOKMAL_LUCENE_NORWEGIAN_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new NorwegianLightStemFilter(lowercase(input)));
+                case NORWEGIAN_BOKMAL_LUCENE_NORWEGIAN_MINIMAL_STEM_FILTER ->
+                    tokenFilter(input -> new NorwegianMinimalStemFilter(lowercase(input)));
+                case POLISH_LUCENE_STEMPEL_DIRECT -> {
+                    final StempelStemmer stemmer = new StempelStemmer(PolishAnalyzer.getDefaultTable());
+                    yield direct(token -> {
+                        final StringBuilder stem = stemmer.stem(token);
+                        return stem == null ? token : stem.toString();
+                    });
+                }
+                case POLISH_LUCENE_STEMPEL_FILTER ->
+                    tokenFilter(input -> new StempelFilter(input, new StempelStemmer(PolishAnalyzer.getDefaultTable())));
+                case POLISH_LUCENE_MORFOLOGIK_FILTER -> tokenFilter(MorfologikFilter::new, true);
+                case PORTUGUESE_LUCENE_PORTUGUESE_STEM_FILTER ->
+                    tokenFilter(input -> new PortugueseStemFilter(lowercase(input)));
+                case PORTUGUESE_LUCENE_PORTUGUESE_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new PortugueseLightStemFilter(lowercase(input)));
+                case PORTUGUESE_LUCENE_PORTUGUESE_MINIMAL_STEM_FILTER ->
+                    tokenFilter(input -> new PortugueseMinimalStemFilter(lowercase(input)));
+                case RUSSIAN_LUCENE_RUSSIAN_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new RussianLightStemFilter(lowercase(input)));
+                case SWEDISH_LUCENE_SWEDISH_LIGHT_STEM_FILTER ->
+                    tokenFilter(input -> new SwedishLightStemFilter(lowercase(input)));
+                case SWEDISH_LUCENE_SWEDISH_MINIMAL_STEM_FILTER ->
+                    tokenFilter(input -> new SwedishMinimalStemFilter(lowercase(input)));
+                case UKRAINIAN_MORFOLOGIK_DIRECT -> {
+                    final DictionaryLookup lookup = new DictionaryLookup(loadUkrainianMorfologikDictionary());
+                    yield morphologik(lookup);
+                }
+                case UKRAINIAN_LUCENE_MORFOLOGIK_FILTER -> {
+                    final Dictionary dictionary = loadUkrainianMorfologikDictionary();
+                    yield tokenFilter(input -> new MorfologikFilter(input, dictionary), true);
+                }
+                default -> throw new IllegalStateException("No evaluator for " + this + ".");
+            };
+        }
+
+        /** Creates the exact-root evaluator used by the JMH quality benchmark. */
+        QualityEvaluator createEvaluator() throws IOException {
+            return exactRootEvaluator(createStemmer());
+        }
+    }
+
+    /**
+     * Creates a CISTEM stemmer adapter.
+     *
+     * @return German stem function
+     */
+    private static Stemmer createGermanCistemStemmer() {
+        return Cistem::stem;
+    }
+
+    /**
+     * Direct stemmer function.
+     */
+    @FunctionalInterface
+    private interface Stemmer {
+
+        /**
+         * Produces one stem.
+         *
+         * @param token input token
+         * @return produced stem
+         */
+        String stem(String token);
+    }
+
+    /**
+     * Quality evaluator for one candidate.
+     */
+    @FunctionalInterface
+    private interface QualityEvaluator {
+
+        /**
+         * Evaluates exact-root agreement for one corpus.
+         *
+         * @param corpus token/root corpus
+         * @param blackhole result sink
+         * @return exact-root match count
+         * @throws IOException if Lucene streaming fails
+         */
+        QualityResult evaluate(LanguageBenchmarkCorpus.Corpus corpus, Blackhole blackhole) throws IOException;
+    }
+
+    /** Stateful candidate adapter confined to one sequential evaluation scenario. */
+    @FunctionalInterface
+    interface CandidateStemmer {
+
+        /**
+         * Stems a deterministic batch through the authoritative JMH invocation path.
+         *
+         * @param tokens input tokens, never {@code null}
+         * @return one non-null output for every input token
+         * @throws IOException if a token-stream implementation fails
+         */
+        String[] stem(String[] tokens) throws IOException;
+
+        /**
+         * Returns complete distinct candidate sets, each containing its primary output.
+         * Single-output adapters expose singleton lists.
+         *
+         * @param tokens input tokens
+         * @return immutable candidate list for every token
+         * @throws IOException if adapter processing fails
+         */
+        default List<List<String>> stemCandidates(final String[] tokens) throws IOException {
+            final String[] primary = stem(tokens);
+            return java.util.Arrays.stream(primary).map(List::of).toList();
+        }
+
+        /** @return whether the adapter exposes genuine alternative outputs */
+        default boolean supportsMultipleOutputs() {
+            return false;
+        }
+    }
+
+    /** Creates the candidate-capable Radixor adapter backed by ranked {@code getAll}. */
+    private static CandidateStemmer radixor(final RadixorBenchmarkStemmer stemmer) {
+        return new CandidateStemmer() {
+            /** {@inheritDoc} */
+            @Override public String[] stem(final String[] tokens) {
+                final String[] outputs = new String[tokens.length];
+                for (int index = 0; index < tokens.length; index++) { outputs[index] = stemmer.stem(tokens[index]); }
+                return outputs;
+            }
+            /** {@inheritDoc} */
+            @Override public List<List<String>> stemCandidates(final String[] tokens) {
+                return java.util.Arrays.stream(tokens).map(stemmer::stemAll).toList();
+            }
+            /** {@inheritDoc} */
+            @Override public boolean supportsMultipleOutputs() { return true; }
+        };
+    }
+
+    /**
+     * Creates the authoritative multi-output Radixor adapter for a validated dictionary language.
+     *
+     * @param language bundled Radixor language
+     * @return scenario-confined adapter using the JMH invocation path
+     * @throws IOException if the compiled dictionary cannot be loaded
+     */
+    static CandidateStemmer createRadixorQualityStemmer(final StemmerPatchTrieLoader.Language language)
+            throws IOException {
+        return radixor(createRadixorStemmer(language));
+    }
+
+    /**
+     * Exact-root agreement counters for one quality operation.
+     *
+     * @param correctMatches total exact-root matches
+     * @param evaluatedTokens total evaluated tokens
+     * @param changedCorrectMatches exact-root matches where token differs from root
+     * @param changedEvaluatedTokens evaluated tokens where token differs from root
+     * @param rootPreservedMatches exact-root matches where token already equals root
+     * @param rootEvaluatedTokens evaluated tokens where token already equals root
+     */
+    private record QualityResult(int correctMatches, int evaluatedTokens, int changedCorrectMatches,
+            int changedEvaluatedTokens, int rootPreservedMatches, int rootEvaluatedTokens) {
+    }
+
+    /**
+     * Creates a direct candidate adapter.
+     *
+     * @param stemmer direct stemmer
+     * @return sequential batch adapter
+     */
+    private static CandidateStemmer direct(final Stemmer stemmer) {
+        Objects.requireNonNull(stemmer, "stemmer");
+        return tokens -> {
+            final String[] outputs = new String[tokens.length];
+            for (int index = 0; index < tokens.length; index++) {
+                outputs[index] = stemmer.stem(tokens[index]);
+            }
+            return outputs;
+        };
+    }
+
+    /**
+     * Creates a TokenFilter candidate adapter.
+     *
+     * @param factory token stream factory
+     * @return sequential batch adapter
+     */
+    private static CandidateStemmer tokenFilter(final Function<TokenStream, TokenStream> factory) {
+        Objects.requireNonNull(factory, "factory");
+        return tokens -> firstTokenFilterOutputs(tokens, factory, null);
+    }
+
+    /** Creates a TokenFilter adapter that preserves all terms emitted per position. */
+    private static CandidateStemmer tokenFilter(final Function<TokenStream, TokenStream> factory,
+            final boolean multipleOutputs) {
+        if (!multipleOutputs) { return tokenFilter(factory); }
+        return new CandidateStemmer() {
+            /** {@inheritDoc} */
+            @Override public String[] stem(final String[] tokens) throws IOException {
+                return firstTokenFilterOutputs(tokens, factory, null);
+            }
+            /** {@inheritDoc} */
+            @Override public List<List<String>> stemCandidates(final String[] tokens) throws IOException {
+                return allTokenFilterOutputs(tokens, factory);
+            }
+            /** {@inheritDoc} */
+            @Override public boolean supportsMultipleOutputs() { return true; }
+        };
+    }
+
+    /** Creates a multi-analysis Morphologik direct adapter. */
+    private static CandidateStemmer morphologik(final DictionaryLookup lookup) {
+        return new CandidateStemmer() {
+            /** {@inheritDoc} */
+            @Override public String[] stem(final String[] tokens) {
+                final String[] outputs = new String[tokens.length];
+                for (int index = 0; index < tokens.length; index++) { outputs[index] = firstMorfologikStem(tokens[index], lookup); }
+                return outputs;
+            }
+            /** {@inheritDoc} */
+            @Override public List<List<String>> stemCandidates(final String[] tokens) {
+                return java.util.Arrays.stream(tokens).map(token -> allMorfologikStems(token, lookup)).toList();
+            }
+            /** {@inheritDoc} */
+            @Override public boolean supportsMultipleOutputs() { return true; }
+        };
+    }
+
+    /**
+     * Creates exact-root accounting around an authoritative candidate adapter.
+     *
+     * @param stemmer candidate adapter
+     * @return JMH exact-root evaluator
+     */
+    private static QualityEvaluator exactRootEvaluator(final CandidateStemmer stemmer) {
+        Objects.requireNonNull(stemmer, "stemmer");
+        return (corpus, blackhole) -> {
+            final String[] actualStems = stemmer.stem(corpus.tokens());
+            final String[] expectedRoots = corpus.expectedRoots();
+            final String[] tokens = corpus.tokens();
+            int correct = 0;
+            int changedCorrect = 0;
+            int changedEvaluated = 0;
+            int rootPreserved = 0;
+            int rootEvaluated = 0;
+            for (int index = 0; index < actualStems.length; index++) {
+                final String token = tokens[index];
+                final String expectedRoot = expectedRoots[index];
+                blackhole.consume(actualStems[index]);
+                final boolean exact = Objects.equals(expectedRoot, actualStems[index]);
+                if (exact) {
+                    correct++;
+                }
+                if (Objects.equals(token, expectedRoot)) {
+                    rootEvaluated++;
+                    if (exact) {
+                        rootPreserved++;
+                    }
+                } else {
+                    changedEvaluated++;
+                    if (exact) {
+                        changedCorrect++;
+                    }
+                }
+            }
+            return new QualityResult(correct, actualStems.length, changedCorrect, changedEvaluated, rootPreserved,
+                    rootEvaluated);
+        };
+    }
+
+    /**
+     * Creates a direct Radixor stemmer.
+     *
+     * @param language Radixor dictionary language
+     * @return direct stemmer
+     * @throws IOException if the trie cannot be loaded
+     */
+    private static RadixorBenchmarkStemmer createRadixorStemmer(final StemmerPatchTrieLoader.Language language) throws IOException {
+        return new RadixorBenchmarkStemmer(StemmerPatchTrieLoader.loadCompiled(
+                language, true, ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS));
+    }
+
+    /**
+     * Loads the benchmark-only Ukrainian Morfologik dictionary.
+     *
+     * @return Ukrainian Morfologik dictionary
+     * @throws IOException if the dictionary cannot be loaded
+     */
+    private static Dictionary loadUkrainianMorfologikDictionary() throws IOException {
+        final URL dictionaryUrl = StemmerComparisonBenchmarkQuality.class.getClassLoader()
+                .getResource("ua/net/nlp/ukrainian.dict");
+        if (dictionaryUrl == null) {
+            throw new IllegalStateException("Missing Ukrainian Morfologik dictionary resource.");
+        }
+        return Dictionary.read(dictionaryUrl);
+    }
+
+    /**
+     * Returns the first Morfologik stem for one token.
+     *
+     * @param token input token
+     * @param lookup dictionary lookup
+     * @return first Morfologik stem, or the input token when no analysis exists
+     */
+    private static String firstMorfologikStem(final String token, final DictionaryLookup lookup) {
+        final List<WordData> analyses = lookup.lookup(token);
+        if (analyses.isEmpty()) {
+            return token;
+        }
+        return analyses.get(0).getStem().toString();
+    }
+
+    /** Returns all distinct Morphologik lemma strings and always includes the primary output. */
+    private static List<String> allMorfologikStems(final String token, final DictionaryLookup lookup) {
+        final java.util.LinkedHashSet<String> stems = new java.util.LinkedHashSet<>();
+        stems.add(firstMorfologikStem(token, lookup));
+        for (WordData analysis : lookup.lookup(token)) { stems.add(analysis.getStem().toString()); }
+        return List.copyOf(stems);
+    }
+
+    /**
+     * Extracts the first emitted term for each input token from a TokenFilter
+     * pipeline.
+     *
+     * @param tokens token corpus
+     * @param factory token stream factory
+     * @param blackhole result sink
+     * @return first emitted term per input token
+     * @throws IOException if Lucene streaming fails
+     */
+    private static String[] firstTokenFilterOutputs(final String[] tokens, final Function<TokenStream, TokenStream> factory,
+            final Blackhole blackhole) throws IOException {
+        final String[] outputs = new String[tokens.length];
+        final BenchmarkTokenStream input = new BenchmarkTokenStream(tokens);
+        final TokenStream output = factory.apply(input);
+        final CharTermAttribute termAttribute = output.addAttribute(CharTermAttribute.class);
+        final PositionIncrementAttribute positionAttribute = output.addAttribute(PositionIncrementAttribute.class);
+        int inputIndex = -1;
+        boolean recordedForPosition = false;
+
+        output.reset();
+        while (output.incrementToken()) {
+            final int positionIncrement = positionAttribute.getPositionIncrement();
+            if (positionIncrement > 0) {
+                inputIndex += positionIncrement;
+                recordedForPosition = false;
+            }
+            if (inputIndex >= 0 && inputIndex < outputs.length && !recordedForPosition) {
+                outputs[inputIndex] = termAttribute.toString();
+                recordedForPosition = true;
+            }
+            if (blackhole != null) {
+                blackhole.consume(termAttribute);
+            }
+        }
+        output.end();
+        output.close();
+
+        for (int index = 0; index < outputs.length; index++) {
+            if (outputs[index] == null) {
+                outputs[index] = tokens[index];
+            }
+        }
+        return outputs;
+    }
+
+    /**
+     * Extracts every distinct emitted term for each input position and includes the
+     * deterministic primary output even when a filter omits it.
+     */
+    private static List<List<String>> allTokenFilterOutputs(final String[] tokens,
+            final Function<TokenStream, TokenStream> factory) throws IOException {
+        final List<java.util.LinkedHashSet<String>> candidates = new java.util.ArrayList<>(tokens.length);
+        for (int index = 0; index < tokens.length; index++) { candidates.add(new java.util.LinkedHashSet<>()); }
+        final BenchmarkTokenStream input = new BenchmarkTokenStream(tokens);
+        final TokenStream output = factory.apply(input);
+        final CharTermAttribute term = output.addAttribute(CharTermAttribute.class);
+        final PositionIncrementAttribute position = output.addAttribute(PositionIncrementAttribute.class);
+        int inputIndex = -1;
+        output.reset();
+        while (output.incrementToken()) {
+            if (position.getPositionIncrement() > 0) { inputIndex += position.getPositionIncrement(); }
+            if (inputIndex >= 0 && inputIndex < candidates.size()) { candidates.get(inputIndex).add(term.toString()); }
+        }
+        output.end();
+        output.close();
+        final String[] primary = firstTokenFilterOutputs(tokens, factory, null);
+        final List<List<String>> result = new java.util.ArrayList<>(tokens.length);
+        for (int index = 0; index < tokens.length; index++) {
+            candidates.get(index).add(primary[index]);
+            result.add(List.copyOf(candidates.get(index)));
+        }
+        return List.copyOf(result);
+    }
+
+    /**
+     * Adds Lucene lower-case normalization.
+     *
+     * @param input input token stream
+     * @return normalized stream
+     */
+    private static TokenStream lowercase(final TokenStream input) {
+        return new LowerCaseFilter(input);
+    }
+
+    /**
+     * Adds Lucene German normalization.
+     *
+     * @param input input token stream
+     * @return normalized stream
+     */
+    private static TokenStream germanNormalize(final TokenStream input) {
+        return new GermanNormalizationFilter(lowercase(input));
+    }
+
+    /**
+     * Adds Lucene Persian normalization.
+     *
+     * @param input input token stream
+     * @return normalized stream
+     */
+    private static TokenStream persianNormalize(final TokenStream input) {
+        TokenStream result = lowercase(input);
+        result = new DecimalDigitFilter(result);
+        result = new ArabicNormalizationFilter(result);
+        result = new PersianNormalizationFilter(result);
+        return result;
+    }
+}