Ledger

SFX-SETUP-20261004

Install local Stable Audio 3 MLX and reusable sfxgen

← Ledger

donestatus · tooling

13✓ 6✗tests passed / failed

3.09Mtokens · 16 min

0evidence files

Evidence

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

No evidence files in logs/evidence/SFX-SETUP-20261004/.

Tests

fail Wrapper tests before implementation (expected missing-module failure) · 0.1 s · raw log$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python -m unittest discover -s tools/audio -p test_*.py

fail Official MLX offline inference (1.5 second shield impact) · 0.2 s · raw log

$ env HF_HUB_OFFLINE=1 /Users/king/AI/stable-audio-3/optimized/mlx/sa3 --prompt Short fantasy sword hitting a metal shield, sharp metallic clang, compact impact, isolated sound effect, no music, no intelligible speech, clean start. --seconds 1.5 --dit sm-sfx --decoder same-s --seed 20261004 --out /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/raw-test/shield.wav

pass Official MLX offline inference and audio validation · 6.1 s · raw log

$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/raw-test/run.py

pass Audio processing and CLI safety unit tests · 0.2 s · raw log$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python -m unittest discover -s tools/audio -p test_*.py

pass Real wrapper inference A-F and audio verification · 13.9 s · raw log

$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/e2e.py

fail Repository data tests · 6.3 s · raw log$ pnpm data:test

fail Regression tests for asymmetric attack and tiny-duration preroll (expected red) · 0.1 s · raw log$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python -m unittest discover -s tools/audio -p test_*.py

pass Unit tests including asymmetric-transient and tiny-duration regressions · 0.2 s · raw log$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python -m unittest discover -s tools/audio -p test_*.py

pass Python lint and format checks · 0.1 s · raw log$ uvx ruff check tools/audio tools/sfxgen && uvx ruff format --check tools/audio tools/sfxgen

pass Repository docs lint and index regeneration · 0.3 s · raw log$ pnpm docs:lint

fail MP3 codec-overshoot regressions (expected red) · 0.1 s · raw log$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python -m unittest discover -s tools/audio -p test_*.py

pass Unit tests with MP3 peak compensation · 0.2 s · raw log$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python -m unittest discover -s tools/audio -p test_*.py

pass Final A-F inference after review fixes · 13.9 s · raw log

$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/e2e.py

pass User-local command and remaining CLI options · 4.1 s · raw log

$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/extra-checks.py

fail Sub-block onset regression at 44.1 kHz (expected red) · 0.2 s · raw log$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python -m unittest discover -s tools/audio -p test_*.py

pass Final 15 audio and safety regression tests · 0.1 s · raw log$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python -m unittest discover -s tools/audio -p test_*.py

pass Final Python lint and format checks · 0.1 s · raw log$ uvx ruff check tools/audio tools/sfxgen && uvx ruff format --check tools/audio tools/sfxgen

pass Final inference A-F with sample-accurate onset detection · 13.2 s · raw log

$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/e2e.py

pass Final documentation lint and index · 0.2 s · raw log$ pnpm docs:lint

Notes

2026-10-04 08:28 UTC · Identified Zoen by exact audio-request indicators and active-project instructions. Use its -3 dBFS SFX ceiling, no gain boost by default. Bootstrap audio stays in the Codex work directory; production catalog is unchanged.

2026-10-04 08:40 UTC · Offline official inference and all wrapper A-F tests succeeded. Metal GPU operational; Python3.11.16 MLX0.32.3. Review fixes: preserve float decoder peaks before limiting, robust sample-level onset, bounded preroll, decoded MP3 ceiling compensation. No model credentials configured or written. Data tests75/77; two preexisting missing ART-UIK-window-plate reference failures, left untouched. No production audio generated.