Compares cms e2e/render-bench/results/latest.json (written by the dojo
render-bench suite) against the committed baseline.json. FAILs when
home-page p50 exceeds baseline by >25%, with an explicit instruction to
investigate and never bump the baseline without Captain approval. p95
breach (>40%) WARNs only: measured ambient p95 variance on an idle dev
instance exceeds 100% between runs while p50 holds within ~7%, so a hard
tail gate would block unrelated commits on dev-stack noise.
SKIPs (never blocks) when: baseline.json absent (non-cms repo),
latest.json absent (bench not run), latest older than 7 days, or latest
predates the baseline — each with a nudge to run the suite from dojo.
Comparison logic is Reporter-free and unit-tested (renderperf_test.go).
Verified against the live tree in all three states: OK (p50 22.9 vs
27.3ms), FAIL (26% slower via delay proxy, mandated message), SKIP
(latest.json removed).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
