Experiment: prepare revised HTTP knowledge A/B gate

.github/actions/install-agent-harnesses/action.yml

@@ -6,11 +6,18 @@ inputs:
     description: Full BC-ALAgents commit SHA (blank uses the default pin)
     required: false
     default: ""
+  copilot-version:
+    description: GitHub Copilot CLI version (defaults to 1.0.82)
+    required: false
+    default: "1.0.82"
 
 outputs:
   bc-alagents-path:
     description: Path to the pinned BC-ALAgents checkout
     value: ${{ steps.bc-alagents-path.outputs.path }}
+  copilot-version:
+    description: Installed GitHub Copilot CLI version
+    value: ${{ steps.copilot-version.outputs.version }}
 
 runs:
   using: composite
@@ -24,7 +31,7 @@ runs:
       id: engine-sha
       shell: pwsh
       env:
-        ENGINE_SHA: ${{ inputs.engine-sha || 'ecf8e31759d6ddd6d78e3a0b7836b40134368009' }}
+        ENGINE_SHA: ${{ inputs.engine-sha || 'da9236a6f43d47c37e0bd59ebced33b6fe73f277' }}
       run: |
         if ($env:ENGINE_SHA -notmatch '\A[0-9a-fA-F]{40}\z') {
           throw "engine-sha must be a full 40-character hexadecimal commit SHA."
@@ -35,8 +42,19 @@ runs:
       run: npm install -g @anthropic-ai/claude-code@2.1.221
       shell: pwsh
 
+    - name: Validate GitHub Copilot CLI version
+      id: copilot-version
+      shell: pwsh
+      env:
+        COPILOT_VERSION: ${{ inputs.copilot-version }}
+      run: |
+        if ($env:COPILOT_VERSION -notmatch '\A\d+\.\d+\.\d+(?:-[0-9A-Za-z.-]+)?\z') {
+          throw "copilot-version must be a pinned semantic version."
+        }
+        "version=$env:COPILOT_VERSION" >> $env:GITHUB_OUTPUT
+
     - name: Install GitHub Copilot CLI
-      run: npm install -g @github/copilot@1.0.82
+      run: npm install -g @github/copilot@${{ steps.copilot-version.outputs.version }}
       shell: pwsh
 
     - name: Checkout BC-ALAgents

.github/workflows/pr-review-evaluation.yml

@@ -9,17 +9,37 @@ on:
       model:
         description: "Copilot model used internally by BC PR Review"
         required: false
-        default: "gpt-5.6-luna"
+        default: "gpt-5.6-sol"
         type: choice
         options:
-          - "claude-sonnet-5"
-          - "claude-opus-5"
+          - "gpt-5.4"
           - "gpt-5.6-sol"
           - "gpt-5.6-terra"
           - "gpt-5.6-luna"
           - "gpt-5.3-codex"
           - "mai-code-1.1-flash"
-          - "gemini-3.7-flash"
+      leaf-model:
+        description: "Model requested for domain leaf agents"
+        required: false
+        default: "gpt-5.6-luna"
+        type: choice
+        options:
+          - "gpt-5.4"
+          - "gpt-5.6-luna"
+          - "mai-code-1.1-flash"
+      leaf-execution:
+        description: "Deterministic leaf scheduling mode"
+        required: false
+        default: "serial"
+        type: choice
+        options:
+          - "serial"
+          - "parallel"
+      max-leaf-concurrency:
+        description: "Maximum simultaneous leaves in parallel mode"
+        required: false
+        default: "4"
+        type: string
       engine-sha:
         description: "Full BC-ALAgents commit SHA (blank uses the default pin)"
         required: false
@@ -46,6 +66,11 @@ on:
           - "3"
           - "4"
           - "5"
+      entries:
+        description: "Optional JSON array of entry IDs for a focused experiment"
+        required: false
+        default: ""
+        type: string
       git-ref:
         description: "Branch to record for experiment tracking (auto-filled on requeue; leave blank for manual runs)"
         required: false
@@ -95,7 +120,7 @@ jobs:
       fail-fast: false
       max-parallel: 64
       matrix:
-        entry: ${{ fromJson(needs.get-entries.outputs.entries) }}
+        entry: ${{ fromJson(inputs.entries != '' && inputs.entries || needs.get-entries.outputs.entries) }}
     steps:
       - name: Checkout repository
         uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
@@ -121,13 +146,18 @@ jobs:
         uses: $/.github/actions/install-agent-harnesses
         with:
           engine-sha: ${{ inputs.engine-sha }}
+          copilot-version: "1.0.88"
 
       - name: Run BC PR Review for entry ${{ matrix.entry }}
         timeout-minutes: 120
         shell: pwsh
         env:
           COPILOT_GITHUB_TOKEN: ${{ github.token }}
           GH_TOKEN: ${{ github.token }}
+          COPILOT_REVIEW_CLI_VERSION: ${{ steps.install-harnesses.outputs.copilot-version }}
+          COPILOT_REVIEW_LEAF_MODEL: ${{ inputs.leaf-model }}
+          COPILOT_REVIEW_LEAF_EXECUTION: ${{ inputs.leaf-execution }}
+          COPILOT_REVIEW_MAX_LEAF_CONCURRENCY: ${{ inputs.max-leaf-concurrency }}
         run: |
           Write-Output "::add-mask::$env:COPILOT_GITHUB_TOKEN"
 
@@ -142,9 +172,32 @@ jobs:
         if: always()
         with:
           name: evaluation-results-${{ github.run_id }}-${{ matrix.entry }}
-          path: ${{ env.EVALUATION_RESULTS_DIR }}/**/*.jsonl
+          path: |
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/*.jsonl
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/_run-manifest.json
           retention-days: ${{ (inputs.test-run || inputs.modified-only) && 1 || 30 }}
 
+      - name: Upload allowlisted review diagnostics
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        if: always()
+        with:
+          name: review-diagnostics-${{ github.run_id }}-${{ matrix.entry }}
+          path: |
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/_run-manifest.json
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/_run-metrics.json
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/_review-report.json
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/al-code-review-findings.json
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/agent-output.txt
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/agent-transcript.log
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/leaf-results/*/stdout.txt
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/leaf-results/*/stderr.txt
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/_filter-report.json
+            ${{ env.EVALUATION_RESULTS_DIR }}/**/bcquality/schemas/findings-report.schema.json
+            ${{ steps.setup-env.outputs.repo_path }}/review.json
+            ${{ steps.setup-env.outputs.repo_path }}/judge_results.json
+          if-no-files-found: warn
+          retention-days: 7
+
   summarize-results:
     needs: evaluate-with-pr-review
     uses: $/.github/workflows/summarize-results.yml
@@ -157,7 +210,7 @@ jobs:
       agent: "BC PR Review"
       # Scores only the modified entries, so publishing it to Braintrust/Kusto or the
       # leaderboard would record a partial corpus as a benchmark result.
-      mock: ${{ inputs.test-run || inputs.modified-only || inputs.engine-sha != '' }}
+      mock: ${{ inputs.test-run || inputs.modified-only || inputs.engine-sha != '' || inputs.entries != '' }}
       category: code-review
       git-ref: ${{ inputs.git-ref || github.ref_name }}
     secrets: inherit
@@ -174,4 +227,4 @@ jobs:
       repeat: ${{ inputs.repeat }}
       existing-tag: ${{ needs.pin-commit.outputs.tag-name }}
       workflow-inputs: |
-        {"model": "${{ inputs.model }}", "engine-sha": ${{ toJSON(inputs.engine-sha) }}, "test-run": "${{ inputs.test-run }}", "git-ref": "${{ inputs.git-ref || github.ref_name }}"}
+        {"model": "${{ inputs.model }}", "leaf-model": "${{ inputs.leaf-model }}", "leaf-execution": "${{ inputs.leaf-execution }}", "max-leaf-concurrency": "${{ inputs.max-leaf-concurrency }}", "engine-sha": ${{ toJSON(inputs.engine-sha) }}, "test-run": "${{ inputs.test-run }}", "entries": ${{ toJSON(inputs.entries) }}, "git-ref": "${{ inputs.git-ref || github.ref_name }}"}