Record measured warm startup and remaining HD replay blocker

This commit is contained in:
firestar5683
2026-09-19 17:39:14 -07:00
parent 86b6077c8c
commit 4dcde78413
+5 -1
View File
@@ -1,4 +1,8 @@
2026-09-19 standalone recovery: six current-planner roles are prepared and validated (16.1 MB, no PCM). Main includes standalone local conditioning, corrected Preparing placement/keep-awake, exact archived Mac replay, and conservative-v2 presentation with per-passage beat assessment. The comma connection recovered; the hardware owner restored native replay, staged both local favorites and the plan bank, verified 100 W, and launched the first standalone fresh run. Native acceptance is still in progress; no completed fresh replay or warm-start improvement is claimed. Protected gold and fallback remain unchanged.
2026-09-19 current standalone status: the integrated HD demo is not yet passing. Native ACE resident reuse passed same-seed PCM/latent equivalence and fresh-seed variation without model reload or graph compilation. First accepted audio took 16.8–17.3 seconds; 110 seconds of accepted buffer took 91.3–92.1 seconds at 100 W. The normal native command now selects resident preparation automatically.
JLab Bluetooth output started during user-operated replay. The latest run produced 36.4 seconds of audio with no underflows, but HD replay stalled at the first segment boundary and the synchronization guard stopped it. Initial cache/frame priming fixed the earlier startup stall. A separate native decode fixture reproduces only 17.2–17.4 fps against the required 20 fps; sustained HD decoding is the current blocker. Decoder changes require measured frame equivalence and throughput before another user test. The warm worker, generated music, original gold and fallback assets remain preserved. User owns the next replay launch.
Galaxy RoadScore settings and tap calibration exist only in an isolated local branch and are not installed on the device. They require adaptation from PulseAudio to the device's BlueALSA output. They are deferred behind the replay fix.
Mac archived replay passed actual UI capture and exact final-sample playback against a complete historical archive. Presentation passed 33 offline tests; an actual 27-second opening gives a coherent estimated beat after removing the former 30-second minimum. Native alert accents are sparse, beat-aligned and suppressed during competing cues. Subjective cue salience still requires listening.