diff --git a/.github/workflows/performance.yaml b/.github/workflows/performance.yaml
index 9244de9708..e61264f2b1 100644
--- a/.github/workflows/performance.yaml
+++ b/.github/workflows/performance.yaml
@@ -8,6 +8,10 @@ on:
   pull_request:
     branches:
       - main
+    # `labeled` lets applying the `performance-benchmark` label trigger a rerun
+    # without needing an empty commit; the job-level `if` filters out unrelated
+    # label changes so we don't burn CI on every label add.
+    types: [opened, synchronize, reopened, labeled]
 
 env:
   CI: true
@@ -16,6 +20,9 @@ env:
 jobs:
   historical-versions:
     runs-on: ubuntu-latest
+    # Skip the job when a `labeled` event fires for an unrelated label —
+    # opens/pushes/reopens always run; only the label firehose is filtered.
+    if: github.event.action != 'labeled' || github.event.label.name == 'performance-benchmark'
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
       - name: Install pnpm for Rebilly/api-definitions
@@ -33,8 +40,9 @@ jobs:
         run: npm i -g hyperfine
 
       - name: Add more versions to test
-        # Run only on the release branch (changeset-release/main):
-        if: github.head_ref == 'changeset-release/main'
+        # Runs on the release PR (changeset-release/main) automatically, or on
+        # any PR carrying the `performance-benchmark` label for opt-in deep-dive.
+        if: github.head_ref == 'changeset-release/main' || contains(github.event.pull_request.labels.*.name, 'performance-benchmark')
         run: |
           cd tests/performance/
           cat package.json | jq ".dependencies = $(cat package.json | jq ._enhancedDependencies)" > package.json
@@ -52,12 +60,16 @@ jobs:
         run: |
           cd tests/performance/
           npm run test # This command is generated and injected into package.json in the previous step.
-          cat benchmark_check.md
+          cat benchmark_bundle.md benchmark_lint.md benchmark_check-config.md
           npm run chart # Creates benchmark_chart.md with the performance bar chart.
 
       - name: Comment PR
+        # If the PR carries the `performance-benchmark` label, each push posts a
+        # fresh comment (so variance across reruns can be compared in-thread);
+        # otherwise the comment is overwritten in place so normal PRs keep a
+        # single result.
         if: ${{ github.event.pull_request.head.repo.full_name == github.repository }}
         uses: thollander/actions-comment-pull-request@24bffb9b452ba05a4f3f77933840a6a841d1b32b # v3.0.1
         with:
           file-path: tests/performance/benchmark_chart.md
-          comment-tag: historical-versions-comparison
+          comment-tag: ${{ contains(github.event.pull_request.labels.*.name, 'performance-benchmark') && format('historical-versions-rerun-{0}', github.run_id) || 'historical-versions-comparison' }}
diff --git a/tests/performance/.gitignore b/tests/performance/.gitignore
index 7d90fe0bc1..e59ba9b4a2 100644
--- a/tests/performance/.gitignore
+++ b/tests/performance/.gitignore
@@ -2,6 +2,10 @@ api-definitions
 node_modules
 package-lock.json
 test-command.txt
-benchmark_check.md
-benchmark_check.json
+benchmark_bundle.md
+benchmark_bundle.json
+benchmark_lint.md
+benchmark_lint.json
+benchmark_check-config.md
+benchmark_check-config.json
 benchmark_chart.md
diff --git a/tests/performance/chart.js b/tests/performance/chart.js
index 9cfcb8fb0c..f8daade056 100644
--- a/tests/performance/chart.js
+++ b/tests/performance/chart.js
@@ -1,39 +1,65 @@
 import fs from 'node:fs';
 
-const content = fs.readFileSync('benchmark_check.json', 'utf8');
-const json = JSON.parse(content);
-const arr = json.results.map((r) => [
-  r.command.replace(/^node node_modules\/([^/]+)\/.*/, (_, cliVersion) => cliVersion),
-  r.mean,
-  r.stddev,
-]);
-const minMean = Math.min(...arr.map(([_, mean]) => mean));
-
-const constructBarForChart = (mean, min) => {
+const median = (xs) => {
+  const sorted = [...xs].sort((a, b) => a - b);
+  const mid = Math.floor(sorted.length / 2);
+  return sorted.length % 2 ? sorted[mid] : (sorted[mid - 1] + sorted[mid]) / 2;
+};
+
+const medianAbsoluteDeviation = (xs, centre) => median(xs.map((x) => Math.abs(x - centre)));
+
+const constructBarForChart = (value, min) => {
   if (min <= 0) return 'N/A';
-  const slownessRatio = mean / min;
-  const slownessFactor = slownessRatio - 1;
+  const slownessFactor = value / min - 1;
   const maxBarLength = 30;
-  const visualFactor = Math.min(1, slownessFactor);
-  const length = Math.floor(visualFactor * maxBarLength);
+  const length = Math.floor(Math.min(1, slownessFactor) * maxBarLength);
   return '▓' + '▓'.repeat(length);
 };
 
+const loadResults = (jsonPath) => {
+  const json = JSON.parse(fs.readFileSync(jsonPath, 'utf8'));
+  return new Map(
+    json.results.map((r) => {
+      const cliVersion = r.command.replace(/^node node_modules\/([^/]+)\/.*/, (_, v) => v);
+      return [cliVersion, { median: r.median, mad: medianAbsoluteDeviation(r.times, r.median) }];
+    })
+  );
+};
+
+const findFastest = (results) =>
+  [...results.values()].reduce((best, r) => (r.median < best.median ? r : best));
+
+const renderCell = (entry, fastest) => {
+  const bar = constructBarForChart(entry.median, fastest.median);
+  const factor = entry.median / fastest.median;
+  if (entry === fastest) {
+    return `${bar} ${factor.toFixed(2)}x (Fastest)`;
+  }
+  const relativeUnc =
+    factor * Math.sqrt((entry.mad / entry.median) ** 2 + (fastest.mad / fastest.median) ** 2);
+  return `${bar} ${factor.toFixed(2)}x ± ${relativeUnc.toFixed(2)}`;
+};
+
+const operations = [
+  { name: 'Bundle', file: 'benchmark_bundle.json' },
+  { name: 'Lint', file: 'benchmark_lint.json' },
+  { name: 'Check Config', file: 'benchmark_check-config.json' },
+];
+
+const columns = operations.map(({ name, file }) => {
+  const data = loadResults(file);
+  return { name, data, fastest: findFastest(data) };
+});
+const versions = [...columns[0].data.keys()];
+
 const output = [
-  '| CLI Version | Mean Time ± Std Dev (s) | Relative Performance (Lower is Faster) |',
-  '|---|---|---|',
-  ...arr.map(([cliVersion, mean, stddev]) => {
-    const bar = constructBarForChart(mean, minMean);
-    const meanFormatted = mean.toFixed(3);
-    const stddevFormatted = stddev.toFixed(3);
-    const relativeSpeedFactor = (mean / minMean).toFixed(2);
-    const factorSuffix = mean === minMean ? 'x (Fastest)' : 'x';
-
-    const timeWithStddev = `${meanFormatted}s ± ${stddevFormatted}s`;
-    const performanceDisplay = `${bar} ${relativeSpeedFactor}${factorSuffix}`;
-
-    return `| ${cliVersion} | ${timeWithStddev} | ${performanceDisplay} |`;
-  }),
+  '## Performance Benchmark',
+  '',
+  `| CLI Version | ${columns.map((c) => c.name).join(' | ')} |`,
+  `|---|${columns.map(() => '---').join('|')}|`,
+  ...versions.map(
+    (v) => `| ${v} | ${columns.map((c) => renderCell(c.data.get(v), c.fastest)).join(' | ')} |`
+  ),
 ].join('\n');
 
 process.stdout.write(output);
diff --git a/tests/performance/make-test-command.sh b/tests/performance/make-test-command.sh
index 4b1d96cfd5..2b5a42c34b 100644
--- a/tests/performance/make-test-command.sh
+++ b/tests/performance/make-test-command.sh
@@ -7,7 +7,11 @@ git clone https://github.com/Rebilly/api-definitions.git
 cd api-definitions && pnpm install && cd ..
 
 # Store the command into a text file:
-echo REDOCLY_SUPPRESS_UPDATE_NOTICE=true hyperfine --warmup 1 $(cat package.json | jq '.dependencies' | jq 'keys' | jq 'map("'\''node node_modules/" + . + "/bin/cli.js bundle all@latest --config=api-definitions/redocly.yaml'\''")' | jq 'join(" ")' | xargs) --export-markdown benchmark_check.md --export-json benchmark_check.json > test-command.txt
+build_cmds() {
+  jq -r --arg suffix "$1" '.dependencies | keys | map("'\''node node_modules/" + . + "/bin/cli.js " + $suffix + "'\''") | join(" ")' package.json
+}
+
+echo "REDOCLY_SUPPRESS_UPDATE_NOTICE=true hyperfine --warmup 2 $(build_cmds 'bundle all@latest --config=api-definitions/redocly.yaml') --export-markdown benchmark_bundle.md --export-json benchmark_bundle.json && REDOCLY_SUPPRESS_UPDATE_NOTICE=true hyperfine --warmup 2 $(build_cmds 'lint all@latest --config=api-definitions/redocly.yaml --generate-ignore-file') --export-markdown benchmark_lint.md --export-json benchmark_lint.json && REDOCLY_SUPPRESS_UPDATE_NOTICE=true hyperfine --warmup 2 $(build_cmds 'check-config --config=api-definitions/redocly.yaml --lint-config=warn') --export-markdown benchmark_check-config.md --export-json benchmark_check-config.json" > test-command.txt
 
 # Put the command in the test section of the package.json:
 cat package.json | jq ".scripts.test = \"$(cat test-command.txt)\"" > package.json
diff --git a/tests/performance/package.json b/tests/performance/package.json
index 0f5d76d56e..c22bb87bdd 100644
--- a/tests/performance/package.json
+++ b/tests/performance/package.json
@@ -13,20 +13,12 @@
   },
   "_enhancedDependencies": {
     "cli-2.0.0": "npm:@redocly/cli@2.0.0",
-    "cli-2.03.1": "npm:@redocly/cli@2.3.1",
-    "cli-2.08.0": "npm:@redocly/cli@2.8.0",
     "cli-2.11.1": "npm:@redocly/cli@2.11.1",
-    "cli-2.12.0": "npm:@redocly/cli@2.12.0",
-    "cli-2.12.2": "npm:@redocly/cli@2.12.2",
     "cli-2.13.0": "npm:@redocly/cli@2.13.0",
-    "cli-2.14.1": "npm:@redocly/cli@2.14.1",
     "cli-2.14.2": "npm:@redocly/cli@2.14.2",
-    "cli-2.19.2": "npm:@redocly/cli@2.19.2",
-    "cli-2.24.1": "npm:@redocly/cli@2.24.1",
-    "cli-2.25.0": "npm:@redocly/cli@2.25.0",
-    "cli-2.25.4": "npm:@redocly/cli@2.25.4",
     "cli-2.27.0": "npm:@redocly/cli@2.27.0",
     "cli-2.30.2": "npm:@redocly/cli@2.30.2",
+    "cli-2.31.0": "npm:@redocly/cli@2.31.0",
     "cli-latest": "npm:@redocly/cli@latest",
     "cli-next": "file:../../redocly-cli.tgz"
   },