test(whale): stabilize realtime headroom probe

[alexdermohr]

Summary

Repair the post-merge main validation failure from Audio PR #43 without changing the Organic engine, frozen candidate, study inputs, reports, audio output, or the existing 0.65-second performance threshold.

The prior test took one perf_counter() sample on a shared CI runner. PR validation passed, while the subsequent main-push run failed at 0.686 s because one wall-clock sample included scheduler contention. The revised test measures process CPU time and uses the best of three fresh voices. This keeps the same computational budget and still rejects a sustained regression, while excluding unrelated runner stalls.

Evidence

• Failed main run: 30748935704
• Failure: 0.6863655660000063 not less than 0.65
• Focused Organic tests: 12/12 PASS
• Full just check: 566 tests PASS
• Compileall: PASS
• Local finalization receipt: 69f47634df4ab42260e9b3f3288f70ee9094b01a9408273f5b230a06dd12a190

Boundaries

• No engine or runtime source changes
• No study/evidence artifact changes
• No active/default profile changes
• No physical playback or audio-routing effects
• Threshold remains exactly 0.65 seconds

Bureau context: AUDIO-CONTROL-PLANE-V1-T029; follow-up evaluator candidate candidate-269040f93c99c2e672fd688a.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[alexdermohr]

Revision-bound self-review: PASS

• Base: 704e2aff6ea9ed28fa45a43c0e183a99b4aedb49
• Head: 1d6f718902a56bade93e919ef079ca2b48b74303
• Diff SHA-256: b5677f40b6765edb1dea488d99b8f6b0c09bf956db99cc9cc316d751113bc84b
• Scope: exactly one test file; 11 insertions, 6 deletions
• Findings: none

Separated review:

1. Root cause: main-push run 30748935704 failed only one wall-clock assertion at 0.686 s after the exact PR head had passed safety.
2. Measurement: process CPU time excludes unrelated shared-runner scheduler stalls; three fresh voices prevent warm-state reuse.
3. Gate strength: the threshold remains exactly 0.65 s and best-of-three still rejects a sustained computational regression.
4. Evidence preservation: no engine, runtime, study, source, candidate, report, WAV, profile, or default file changes.
5. Validation: focused Organic suite 12/12 PASS; full just check 566/566 PASS; compileall PASS; local receipt 69f47634df4ab42260e9b3f3288f70ee9094b01a9408273f5b230a06dd12a190.
6. Boundaries: no audio playback, routing, service, profile, deployment, or physical-device effect.