name: Weekly Eval # Report-only benchmark run. This workflow surfaces benchmark drift in the # job summary and the uploaded CSV artifact, but it must NOT fail the default # branch on regressions (yet) — eval failures are informational until the # co-change baseline has enough history to set thresholds against. on: schedule: - cron: "23 6 * * 1" # Mondays 06:23 UTC (off-minute to dodge load spikes) workflow_dispatch: permissions: contents: read jobs: eval: runs-on: ubuntu-latest timeout-minutes: 45 steps: - uses: actions/checkout@v4 - name: Set up Python uses: actions/setup-python@v5 with: python-version: "3.12" - name: Install with eval extras run: pip install -e ".[eval]" - name: Run benchmarks (2 smallest pinned configs) # httpx (~60 files) and flask (~83 files) are the two smallest # pinned repos. Report-only: `|| true` keeps regressions and # transient clone failures from failing the default branch. run: | code-review-graph eval \ --repo httpx,flask \ --benchmark token_efficiency,impact_accuracy,agent_baseline \ --output-dir evaluate/results || true - name: Upload result CSVs if: always() uses: actions/upload-artifact@v4 with: name: eval-results-${{ github.run_id }} path: evaluate/results/*.csv if-no-files-found: warn retention-days: 90 - name: Write job summary if: always() run: | python - <<'PY' >> "$GITHUB_STEP_SUMMARY" from code_review_graph.eval.reporter import generate_full_report print("# Weekly eval (report-only)") print() print( "Configs: `httpx`, `flask` (the two smallest pinned repos). " "Regressions are reported here and in the CSV artifact but do " "not fail CI." ) print() print(generate_full_report("evaluate/results")) PY