Remote-HLS subtitle anchor: separate muxer delay from keyframe-restart gap #763

Summary

The remote-HLS subtitle proxy anchors injected WebVTT renditions with X-TIMESTAMP-MAP from a probe of the segment the load opens on (Silo-Server#14). That probe observes one number, D - G, where the map needs D alone:

• D is the muxer delay, constant for the session (ffmpeg MPEG-TS: 1.4 s by default, 10 s with -max_delay 5000000; fMP4: 0).
• G is the keyframe-restart gap: how far before its playlist slot the first segment of a started or restarted transcode begins, when the origin restarts at the keyframe before the slot (Jellyfin-style -noaccurate_seek -copyts). Only that first segment carries it.

renditionAnchor(measured:) applies only a positive measurement, which is exact when D == 0 or G == 0. When both are nonzero:

• 0 < G < D: the map is D - G, every cue shows G early, AE#616 measures about 0, and sourceTime also leads by G.
• G > D: the plain body shows every cue D early, AE#616 measures G - D, and sourceTime is off by D.

AE#616 cannot see either error, because it measures item time against the same cues and its reference moves with the map. An MPEG-TS origin with a muxer delay that resumes by restarting at the previous keyframe hits this on every resumed load. Silo's own HLS output keeps G = 0, so this concerns third-party origins.

Proposed direction

Probe the segment after the start segment as well: same job, AVPlayer requests it next anyway, and it carries D without G, provided the origin forces keyframes on slot boundaries. Stream copy does not, so it needs a fallback there (for example: only trust the second segment when its measurement is consistent with its slot, otherwise keep the current rule).

References

• fix(hls): anchor injected subtitle renditions to the origin's timestamps Silo-Server/AetherEngine#14 (discussion and the doc comment on RemoteHLSTimestampAnchor.renditionAnchor(measured:))
• AE#616 (remote-HLS source-time correction)