Replace the single 83% global threshold with per-domain thresholds (async, classes, modules, regex, JSON, Symbol, Proxy, fs, crypto, stream, …). CI fails when any category regresses, not just the global average.
Seed data: @proggeramlug has per-module diff counts across 5 sweeps (v0.5.713 → v0.5.910) ready to drop into threshold.json as the initial baseline. This sub-issue composes with #812 (per-module behavioral matrix) — that matrix is the natural data source for these thresholds; this issue is just the regression-gate layer on top.
Part of #793.
Replace the single 83% global threshold with per-domain thresholds (async, classes, modules, regex, JSON, Symbol, Proxy, fs, crypto, stream, …). CI fails when any category regresses, not just the global average.
Seed data: @proggeramlug has per-module diff counts across 5 sweeps (v0.5.713 → v0.5.910) ready to drop into
threshold.jsonas the initial baseline. This sub-issue composes with #812 (per-module behavioral matrix) — that matrix is the natural data source for these thresholds; this issue is just the regression-gate layer on top.Part of #793.