[jpsts]

Summary

Only timestamp jumps of 10 s or more were folded out of the software path's live timeline. A shorter hole, which a feed that pauses and skips what it missed leaves behind, played out as a standing picture with the clock running and the phase reading playing.

What changed

• LiveHoleFold: on the rewind route, a forward hole of 3 s or more in both streams folds. Audio decides what is a hole, so a sparse picture over continuous sound stays as stamped.
• Audio reaching the hole before the video waits for the fold; audio still short of it keeps the fold it was stamped under.
• Each fold is logged, up to 20 per session.

Test plan

• LiveHoleFoldTests.
• aetherctl play --live --sw --dvr-window 600 on a local real-time stream that pauses 8 s and skips what it missed, a Mac, macOS 27: stock 7.56.0 stops twice; with this change once, the second stop gone.
• Full suite on a Mac: one failure, LAN peers cannot take the slots kept for loopback, which times out the same way on stock 7.56.0.
• Known cost: after a fold the picture resumes about 2 s behind the newest media rather than 4 s, because playing out the hole no longer rebuilds the lead. One brief extra rebuffer showed up right after the resume in the test.
• The no-rewind loop keeps its 10 s threshold.