gate-benchmark-rebaseline

To update a benchmark baseline after a performance change, record the highest p95_current value from multiple runs into benchmark_baseline.json and commit the update with provenance linking the change to a specific commit and date.

Rebaseline p95 metrics in CI pipelines only after a legitimate performance change on the main branch, never to silence regressions. The workflow enforces SLO adherence and p95-based comparisons to prevent silent performance degradation.

Recording a new p95 baseline involves capturing p95_current from multiple test runs, updating the tests/crew/benchmark_baseline.json file with the highest value, and submitting a baseline update PR with commit provenance and date.

No, rebaselining gate-result benchmarks must never be used to silence regressions. The rebaseline workflow is strictly for documenting legitimate performance changes on the main branch while enforcing p95 SLO adherence.

The gate benchmark rebaseline workflow operates within existing CI pipelines and pull request processes without external dependencies, requiring only access to the tests/crew/benchmark_baseline.json file and multiple p95 run captures.

Benchmark baseline drift occurs after legitimate performance changes on main because the old p95 values no longer reflect current performance, requiring a rebaseline update to benchmark_baseline.json with new p95_current values and commit provenance.