Repository navigation
Recorded A/V harness: media audio lands 13-18 ms after video, straddling the one-frame gate #750
Description
Activity
Owner ruling 2026-10-02: the steady ~15 ms audio-after-video offset for a media clip in the recording is acceptable. Remaining here: make the harness's A/V limit fit that ruling so it stops failing one run in four, and keep watching for the 34.5 ms (one extra frame) sessions, which are not covered by the ruling.
Fresh data point, main 88eae0d, dev D3D11 core, depth 3, 7.5 s flash/beep run (
program-buffer-recorded-av.mjs): source-corrected median audio−video = +26.2 ms (audio lags), max 36.2 ms, coverage incomplete (one interior pulse unmatched), soalignmentWithinOneVideoFrame = false. The 2026-09-30 range was 13 to 18 ms with occasional 34.5 ms; this run sits above that. Single run, not a trend; the ruling on whether a steady positive offset is acceptable is still the open question.
Found 2026-10-02 while measuring #724.
scripts/qa/program-buffer-recorded-av.mjs(flash-beep media clip taken to Program, Program recorded) reportssourceCorrectedAudioMinusVideoMsbetween 13 and 18 ms on almost every session. Its pass limit is one video frame (16.7 ms), so the gate fails about one run in four with nothing changed.Measured, same machine, two buffer depths per run:
mainc6bf15dTwo things to decide:
main, 2 of 14. Not enough runs to call it a regression.Until then the harness's A/V verdict cannot be used as a gate. Its frame-delivery verdict (added with the #724 remainder) is independent of this.