fix: preserve audio samples across Miruro timestamp jitter

2026-10-13 16:13:33 +02:00
parent 51382bc63a
commit 4ef8ceee24
6 changed files with 11 additions and 5 deletions
@@ -117,6 +117,8 @@ Handle `null` duration as ineligible. This measures position, not viewing covera
 | Scenario | Observed result |
 | --- | --- |
 | Explicit fresh direct Sun native smoke | Video/audio decode and advancing playback; default English VTT; expected cue text after forward/backward acknowledged seeks |
+| Live Sun audio continuity | Dragon Ball episode 1: actual Core Audio PCM diverged from FFmpeg after about 0.6 seconds before the fix; after tolerating millisecond-scale timestamp jitter, all 960,008 stereo frames compared across twenty seconds and multiple HLS segments matched exactly |
+| Audio regression | Deterministic AAC/MPEG-TS timestamp-jitter fixture failed before the shared continuity fix and matched FFmpeg exactly afterward; existing authored-gap, resampling, gain and stepped-resume checks passed |
 | Local resolved playback model | Independent stream/subtitle headers, nested playlist tokens, video/audio playback, cue timing and stable v2 progress |
 | Local native network checks | Encrypted MPEG-TS and fMP4; image-suffixed resources; headers on master/variant/segment/key/init; seeks/frame steps; redirect/foreign-origin isolation; diagnostic redaction; existing YouTube redirects |
 | Actual CLI-to-GUI handoff | Two independent Miruro episode windows; authenticated encrypted HLS; resource-specific default VTT; v2 progress and quit draining |