GC Ratchet fails on every main run with:
gc-ratchet error: 07_array_grow_evacuate: GC counters are not deterministic
across traced runs (copied_bytes, freed_bytes); they cannot be gated
The gate is behaving correctly — it refuses to gate a counter it cannot reproduce.
This is not baseline drift. The current run reports zero REGRESSION cells. (Earlier runs in the window did show 42 drifted cells, but those are gone; the nondeterminism is what remains.)
Two possibilities, and the order to rule them out
- The collector's copied/freed accounting became run-dependent on this workload. This is the more interesting answer and should be excluded first — it would be a real GC finding wearing a CI failure as a disguise.
- The fixture itself became nondeterministic.
Done when
07_array_grow_evacuate is deterministic across traced runs, or it is removed from the gated set with the reason recorded.
Context
GC Ratchet has been red on main continuously since 2026-08-31. It is not a pr-gate context, so nothing blocks on it and the failure has been invisible.
GC Ratchetfails on everymainrun with:The gate is behaving correctly — it refuses to gate a counter it cannot reproduce.
This is not baseline drift. The current run reports zero REGRESSION cells. (Earlier runs in the window did show 42 drifted cells, but those are gone; the nondeterminism is what remains.)
Two possibilities, and the order to rule them out
Done when
07_array_grow_evacuateis deterministic across traced runs, or it is removed from the gated set with the reason recorded.Context
GC Ratchethas been red onmaincontinuously since 2026-08-31. It is not apr-gatecontext, so nothing blocks on it and the failure has been invisible.