Ledger

SFX-AUDIT-20261004-01

Audit generated SFX prompt and usage alignment

← Ledger

donestatus · audio-review

2✓ 1✗tests passed / failed

0tokens · 5 min

2evidence files · 16 kB

Evidence

From logs/evidence/SFX-AUDIT-20261004-01/. Click a thumbnail for the full image.

Logs and data (1)

features.json · 13 kB · open

{
  "method": "signal-feature summary only; does not assess perceived semantic match",
  "takes": [
    {
      "id": "sfx-impact-light",
      "path": "audio/source/combat-core/sfx-impact-light.mp3",
      "duration_s": 0.7,
      "rms_dbfs": -24.786496385105565,
      "peak_dbfs_mono": -13.274496043260292,
      "spectral_centroid_hz_first_0_186ms": 589.583847528993,
      "low_20_150_hz": 0.1465270848711487,
      "mid_150_1000_hz": 0.6589267026540846,
      "high_1000_plus_hz": 0.18468985133422833,
      "first_50ms_rms_dbfs": -24.283579016820823,
      "last_100ms_rms_dbfs": -25.11646933431304,
      "active_span_ms_at_minus40db": 700
    },
    {
      "id": "sfx-impact-light",
      "path": "audio/source/combat-core/sfx-impact-light_2.mp3",
      "duration_s": 0.7,
      "rms_dbfs": -23.64632174604168,
      "peak_dbfs_mono": -10.409703352091118,
      "spectral_centroid_hz_first_0_186ms": 329.28006777022375,
      "low_20_150_hz": 0.23786080398631024,
      "mid_150_1000_hz": 0.6602363833424019,
      "high_1000_plus_hz": 0.045384006869078605,
      "first_50ms_rms_dbfs": -23.372505343151637,
      "last_100ms_rms_dbfs": -22.641621852735124,
      "active_span_ms_at_minus40db": 700
    },
    {
      "id": "sfx-impact-light",
      "path": "audio/source/combat-core/sfx-impact-light_3.mp3",
      "duration_s": 0.7,
      "rms_dbfs": -24.911045266988673,
      "peak_dbfs_mono": -12.029831163334755,
      "spectral_centroid_hz_first_0_186ms": 551.714158389907,
      "low_20_150_hz": 0.10638376013810735,
      "mid_150_1000_hz": 0.7405510885612528,
…

Other files (1)

features.py · 3 kB

Tests

fail measure per-class spectral and envelope differences · 0.0 s · raw log$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python logs/evidence/SFX-AUDIT-20261004-01/features.py

pass measure per-class spectral and envelope differences · 0.4 s · raw log$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python logs/evidence/SFX-AUDIT-20261004-01/features.py

pass documentation lint after audit note · 0.3 s · raw log$ pnpm docs:lint

Notes

2026-10-04 09:23 UTC · Read-only audit of the 20 generated P0 takes and catalog use contract. No local acoustic captioner or audio-input path is available here, so signal features are diagnostic only and do not substitute for listening. Source prompts describe complete strike scenes while AUDIO.md defines short separable impact/material layers; requested durations are 0.6-1.5s versus the documented 0.1-0.3s hit/material layers. Light/medium/heavy median RMS and first-186ms spectral centroid overlap closely (RMS -24.85/-24.93/-24.18 dBFS; centroid 458/443/517 Hz); fur prompt also requests a whole blade strike. The wrapper used the intended sm-sfx/same-s runtime and outputs pass file validation; no evidence of broken local inference, but semantic quality is unconfirmed without audition. No assets/prompts replaced. Evidence in logs/evidence/SFX-AUDIT-20261004-01. docs:lint passed.