docs: close runtime quality review with consolidated lesson
Intrepid/Prometeu/Runtime/pipeline/pr-master There was a failure building this commit

This commit is contained in:
2026-10-14 14:09:12 +01:00
parent 3f9d235546
commit 22769aa438
9 changed files with 1129 additions and 201 deletions
@@ -95,3 +95,27 @@ availability guarantee. No human listening assessment is claimed.
Strict host Clippy passes. Repository-wide strict-gate status remains separate
from these behavioral and device observations.
## Production decode and save pressure
A complementary release observation on 2026-10-14 used the same 1,000-callback,
16-voice synthetic harness while a separate process repeatedly ran
`qualify_file_commits_with_production_asset_decoder` from `prometeu-system`.
Each pressure round performs twenty production glyph loads and durable 64-KiB
FileMemcard commits. Fourteen rounds completed successfully during the audio
process interval. The two test binaries were built before measurement; a worker
started the first pressure process before audio and repeated it until audio
exited, then joined the final invocation. Monotonic process intervals confirm
concurrency in the [measurement record](audio-production-pressure-qualification.json).
Callback p50/p95/p99/max were 14.542/97.667/399.041/1,070.708 microseconds for the
nominal 5,333.333-microsecond period. Allocation and deallocation counts remained
zero; sixteen live sample owners became zero after teardown. All test processes
exited successfully. The record identifies the committed source, toolchain,
per-round save observations and the additional policy tests running on the host.
This measures host contention between processes, not an integrated console
session or a physical callback under this exact workload. The previously named
physical-device observation remains separate. The pressure fixture alternates
loads and saves; deterministic production-decoder tests establish save preemption
during decoding. None of these host-specific maxima is a real-time guarantee.