Studio bench (perf/bench)

A built studio measured in Chromium against an in-process mock of the Sanity API: no tokens, no network, one clock. The interesting output is an A/B verdict per metric from interleaved reference/experiment sessions with a cluster bootstrap — 🔴 regression, 🟢 improvement, ✅ neutral, ⚪ inconclusive (CI too wide within budget; never a coin flip). Absolute numbers are host-relative and only comparable with the run's calibration score alongside. perf/bench/README.md is the full reference; this skill is the operating card. Stored results are read through sanity-radar.

Local