diff --git a/.github/workflows/actions-audit.yaml b/.github/workflows/actions-audit.yaml
index e3efc27..123e78f 100644
--- a/.github/workflows/actions-audit.yaml
+++ b/.github/workflows/actions-audit.yaml
@@ -48,6 +48,11 @@ jobs:
         run: go install github.com/rhysd/actionlint/cmd/actionlint@v1.7.7
 
       - name: Run actionlint
+        env:
+          # Gate on shellcheck warning+ only. The info/style nits (SC2012 "use
+          # find not ls", SC2035 "use ./*glob*") in the canonical pr.yaml are
+          # not worth failing every PR over; warnings and errors still gate.
+          SHELLCHECK_OPTS: --severity=warning
         run: actionlint -color
 
   zizmor:
@@ -76,9 +81,9 @@ jobs:
           GH_TOKEN: ${{ github.token }}
         # Report-only for now: don't fail the job on findings, upload them to
         # Code Scanning instead. Remove `|| true` to turn this into a gate.
-        run: zizmor --format sarif .github/workflows/ > zizmor.sarif || true
+        run: zizmor --config .zizmor.yml --format sarif .github/workflows/ > zizmor.sarif || true
 
       - name: Upload zizmor SARIF
-        uses: github/codeql-action/upload-sarif@e0647621c2984b5ed2f768cb892365bf2a616ad1 # v4
+        uses: github/codeql-action/upload-sarif@e4fba868fa4b1b91e1fdab776edc8cfbe6e9fb81 # v4
         with:
           sarif_file: zizmor.sarif
diff --git a/.github/workflows/codeql.yaml b/.github/workflows/codeql.yaml
index 29a43fc..7309a7b 100644
--- a/.github/workflows/codeql.yaml
+++ b/.github/workflows/codeql.yaml
@@ -77,7 +77,7 @@ jobs:
 
     - name: Initialize CodeQL
       if: steps.check-csharp.outputs.has-csharp == 'true'
-      uses: github/codeql-action/init@e0647621c2984b5ed2f768cb892365bf2a616ad1 # v4
+      uses: github/codeql-action/init@e4fba868fa4b1b91e1fdab776edc8cfbe6e9fb81 # v4
       with:
         languages: ${{ matrix.language }}
         # security-extended adds the broader security query pack on top of the
@@ -159,7 +159,7 @@ jobs:
     - name: Perform CodeQL Analysis
       id: perform-codeql-analysis
       if: steps.check-csharp.outputs.has-csharp == 'true'
-      uses: github/codeql-action/analyze@e0647621c2984b5ed2f768cb892365bf2a616ad1 # v4
+      uses: github/codeql-action/analyze@e4fba868fa4b1b91e1fdab776edc8cfbe6e9fb81 # v4
       with:
         category: "/language:${{matrix.language}}"
 
diff --git a/.github/workflows/perf-regression.yaml b/.github/workflows/perf-regression.yaml
new file mode 100644
index 0000000..52c0a10
--- /dev/null
+++ b/.github/workflows/perf-regression.yaml
@@ -0,0 +1,87 @@
+name: Perf Regression
+
+# Per-PR performance-regression detection (#144).
+#
+# Runs the BenchmarkDotNet suite on the PR and compares it against the baseline
+# published to gh-pages (dev/bench) by benchmarks.yaml. If a benchmark regresses
+# beyond the alert threshold, github-action-benchmark comments on the PR.
+#
+# Report-only for now: BenchmarkDotNet on shared CI runners is noisy, so this
+# does NOT fail the build (`fail-on-alert: false`) and does NOT update the
+# baseline (`auto-push: false`). Flip `fail-on-alert` to true — with a threshold
+# tuned to the observed noise — to turn it into a gate.
+
+on:
+  pull_request:
+    paths:
+      - 'src/**'
+      - 'benchmarks/**'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: perf-regression-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  benchmark:
+    name: Benchmark vs base
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    permissions:
+      contents: read
+      pull-requests: write   # post the regression alert comment
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+        with:
+          persist-credentials: false
+
+      - name: Setup .NET
+        uses: actions/setup-dotnet@a98b56852c35b8e3190ac28c8c2271da59106c68 # v6
+        with:
+          dotnet-version: '10.0.x'
+
+      - name: Restore
+        run: dotnet restore benchmarks/Wolfgang.Etl.TestKit.Benchmarks/Wolfgang.Etl.TestKit.Benchmarks.csproj
+
+      - name: Build (Release)
+        run: dotnet build -c Release --no-restore benchmarks/Wolfgang.Etl.TestKit.Benchmarks/Wolfgang.Etl.TestKit.Benchmarks.csproj
+
+      - name: Run benchmarks
+        working-directory: benchmarks/Wolfgang.Etl.TestKit.Benchmarks
+        run: dotnet run -c Release --no-build -- --filter "*" --job short --memory --exporters json
+
+      - name: Merge per-class BDN reports
+        id: locate
+        working-directory: benchmarks/Wolfgang.Etl.TestKit.Benchmarks
+        run: |
+          shopt -s nullglob
+          reports=(BenchmarkDotNet.Artifacts/results/*-report-full-compressed.json)
+          if [ ${#reports[@]} -eq 0 ]; then
+            echo "::error::No BenchmarkDotNet JSON report found"
+            exit 1
+          fi
+          jq -s '{
+            Title: "BenchmarkDotNet combined report",
+            HostEnvironmentInfo: .[0].HostEnvironmentInfo,
+            Benchmarks: [.[].Benchmarks[]]
+          }' "${reports[@]}" > benchmarks-result.json
+          echo "report=benchmarks/Wolfgang.Etl.TestKit.Benchmarks/benchmarks-result.json" >> "$GITHUB_OUTPUT"
+
+      - name: Compare against gh-pages baseline
+        uses: benchmark-action/github-action-benchmark@52576c92bccf6ac60c8223ec7eb2565637cae9ba  # v1.22.1
+        with:
+          name: BenchmarkDotNet
+          tool: 'benchmarkdotnet'
+          output-file-path: ${{ steps.locate.outputs.report }}
+          gh-pages-branch: gh-pages
+          benchmark-data-dir-path: dev/bench
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          alert-threshold: '200%'
+          comment-on-alert: true
+          fail-on-alert: false
+          auto-push: false
+          save-data-file: false
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index 177debb..a9e6b3a 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -275,7 +275,7 @@ jobs:
 
     steps:
       - name: Checkout code
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           ref: refs/pull/${{ github.event.pull_request.number }}/head
           persist-credentials: false
@@ -337,7 +337,7 @@ jobs:
           echo "✅ Configuration files secured - using versions from main branch"
 
       - name: Setup .NET
-        uses: actions/setup-dotnet@v5
+        uses: actions/setup-dotnet@a98b56852c35b8e3190ac28c8c2271da59106c68 # v6
         with:
           # Match the test stages' SDK list so the solution's older-TFM projects
           # (test projects target netcoreapp3.1 / net5.0-net7.0) restore and build
@@ -394,7 +394,7 @@ jobs:
             --no-build
 
       - name: Upload SARIF to Code Scanning
-        uses: github/codeql-action/upload-sarif@v4
+        uses: github/codeql-action/upload-sarif@e4fba868fa4b1b91e1fdab776edc8cfbe6e9fb81 # v4
         with:
           sarif_file: inspect.sarif
 
diff --git a/.github/workflows/scorecard.yaml b/.github/workflows/scorecard.yaml
index 38e7269..5246e26 100644
--- a/.github/workflows/scorecard.yaml
+++ b/.github/workflows/scorecard.yaml
@@ -45,6 +45,6 @@ jobs:
           publish_results: true
 
       - name: Upload SARIF to Code Scanning
-        uses: github/codeql-action/upload-sarif@e0647621c2984b5ed2f768cb892365bf2a616ad1 # v4
+        uses: github/codeql-action/upload-sarif@e4fba868fa4b1b91e1fdab776edc8cfbe6e9fb81 # v4
         with:
           sarif_file: results.sarif
diff --git a/.github/workflows/semgrep.yaml b/.github/workflows/semgrep.yaml
index 54c85f3..3845a86 100644
--- a/.github/workflows/semgrep.yaml
+++ b/.github/workflows/semgrep.yaml
@@ -57,6 +57,6 @@ jobs:
             --error || true
 
       - name: Upload Semgrep SARIF
-        uses: github/codeql-action/upload-sarif@e0647621c2984b5ed2f768cb892365bf2a616ad1 # v4
+        uses: github/codeql-action/upload-sarif@e4fba868fa4b1b91e1fdab776edc8cfbe6e9fb81 # v4
         with:
           sarif_file: semgrep.sarif
diff --git a/.zizmor.yml b/.zizmor.yml
index f9d2374..efcc5db 100644
--- a/.zizmor.yml
+++ b/.zizmor.yml
@@ -12,3 +12,11 @@ rules:
     config:
       policies:
         "*": hash-pin
+  dangerous-triggers:
+    # pr.yaml deliberately uses `pull_request_target` as a *gated* workflow: it
+    # runs from the trusted main branch, checks out PR code via refs/pull/*/head
+    # (never executing untrusted workflow YAML), and re-fetches analyzers/config
+    # from main so a malicious PR cannot disable checks. This is the canonical
+    # secure pattern, not the vulnerability the rule warns about.
+    ignore:
+      - pr.yaml