Add noise-aware whale evaluator v2

[alexdermohr]

Summary

Implements AUDIO-CONTROL-PLANE-V1-T030 as a separate, preregistered evaluator layer without changing the whale engines, profiles, routing, locked source audio, or live audio state.

• explicit unvoiced decisions for silence, low periodicity, low prominence, and search-boundary ambiguity
• octave-family diagnostics without silently forcing a lower F0
• bandwidth-aware feature availability
• controlled synthetic and prior-annotation corpus
• locked legacy-vs-v2 sensitivity report on Alaska, Stellwagen, and American Samoa
• byte-reproducible reports, including source-archive reproduction without Git history

Bound methodology

The methodology was committed before external v2 evaluation:

• a8d15d16e17ee8196e76aca39d17d530001ec435
• bfe237b4fa21a89a712ad49b4bde709ab46d6106

No thresholds, search bounds, success gates, or external inputs were relaxed after observing external results. Two post-external implementation corrections are recorded in the final sensitivity report: voiced-count semantics and exact conformance to the frozen octave multipliers.

Results

• legacy Stellwagen voiced lag-3 counts reproduced exactly: 36 / 40 / 34
• evaluator v2 voiced search-boundary hits: 0
• controlled corpus: all preregistered cases pass
• Alaska remains fully unvoiced because all frames are search-boundary ambiguous; this is reported as a limitation, not absence of whale signal
• American Samoa high-band feature is explicitly unavailable at the source bandwidth

Artifact SHA-256:

• definition: c4dc3c5d579847d78799a95202616b6ce5a4cbf05941b681b7e4fc8378484045
• controlled report: e2a3e01c9ec04d10588f43b7f8b9b976f04183525be920b889952f54af1c9a30
• sensitivity report: 8fc32f5cd60fd435e5b16d68771d1c2d86812fefba4c9baa9a64931f5e274f71

Verification

• 579/579 repository tests pass on head 734ed786ddd5568c82962008730f408e33e91333
• Ruff, py_compile, compileall, and git diff --check pass
• both reports reproduce byte-for-byte from a git archive with no .git directory
• final local test receipt: 7c91dd20b38816aa73a6668ac194155ad32406b5bbc392f5395fb3e4727120da

Normal merge is required to preserve the preregistration commits.

[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.