fix(lint): warn when a video's slot outlasts its source - #7696

Context: I've been using HyperFrames every day for the past two weeks to make Korean Shorts and long-form videos, and it's been great to work with. This PR is one of a batch of fixes for things I ran into along the way, so you'll see several PRs from me at once — sorry for the pile-up, and thanks for your patience. Each one is reproduced on current main, independent of the others, and can be reviewed, merged or closed on its own.

What

hyperframes lint / check now report video_slot_outlasts_source (warning) when a <video>'s data-duration is longer than what its file has left to play, so the render holds the last frame for the rest of the slot.

⚠ video_slot_outlasts_source: Video "clip.mp4" runs out after 1.9s of its 9.02s slot (source 1.9s, starting 0s in). It holds its final frame for the last 7.12s.
  fix: Shorten data-duration to 1.9s, add loop, or use a longer source if this final-frame hold is unintended.

Why

A Short had <video data-start="0" data-duration="9.019"> pointing at an mp4 that is only 1.9 s long. The render showed 1.9 s of footage and then a still frame for 7.1 s; check passed with no media finding. Holding the final frame is the documented behavior (#2516, docs/reference/lint-rules.mdx video_media_start_at_or_past_eof), so this is not a render bug. What was missing is the warning: the existing probe-based rule only fires when data-media-start is at or past the end of the file, and it skips every video that has no data-media-start at all, which is the usual way to write a clip that starts at the top of its file.

Related work

• Extends video_media_start_at_or_past_eof (same ffprobe duration probe, same skip list).
• fix(video): hold final frame through composition #2516 (hold the final frame through the composition) is the behavior this warns about; Video coverage gate aborts renders using the new hold-last-frame feature (#2516) #2565 / Regression: video coverage gate aborts renders of a looping video shorter than its slot (#2516 / #2606) #2665 were about the render-side coverage gate, not lint.
• No open issue or PR covers this (searched "video shorter than slot", "holds final frame", "freeze last frame", "media-start").

How

packages/lint/src/hevcPreviewLint.ts:

• readFiniteVideoSlot now also keeps slots with no data-media-start (it defaults to 0, as the runtime's readMediaOffsetSeconds does) and records data-duration and data-playback-rate (readPlaybackRate from @hyperframes/parsers/media-duration, the shared reader).
• For each probed slot, a start at/past EOF keeps its existing finding; otherwise resolveNaturalDurationSeconds(source, mediaStart, rate) (the shared "one rule" for media length) gives the timeline seconds the source covers, and a slot longer than that by more than 0.1 s gets video_slot_outlasts_source. 0.1 s keeps a rounded-up data-duration (e.g. 5 s on a 4.97 s file) quiet.
• Still skipped: loop, data-var-src, data-playback-start, and a rate lane in data-automation (the static probe cannot follow a ramp). The skip checks moved into hasUnknownSourcePosition to keep readFiniteVideoSlot under fallow's complexity limit.
• Side effect: a finite video with no data-media-start is now duration-probed too (one ffprobe per unique file, same concurrency cap). The HEVC "probed once per file" test now counts codec probes only.

docs/reference/lint-rules.mdx: new video_slot_outlasts_source entry.

Test plan

Unit (packages/lint/src/project.test.ts, ffprobe mocked like the existing rule's tests):

• warns for data-start="0" data-duration="9.019" on a 1.9 s source (message, fix hint, element id, file)
• counts data-media-start (4 s slot from 3 s into a 6 s file) and data-playback-rate (4 s at 2x on a 6 s file); 3 s at 2x on a 6 s file is quiet
• quiet for a 0.05 s hold, loop, a rate lane, and a start past EOF (which still gets only the old finding)

On main (affa03b6) with the fix stashed: 2 of the 3 new tests fail (× warns when a slot outlasts the source…, × counts data-media-start and data-playback-rate…). With the fix: packages/lint 27 files / 1312 tests pass; tsc --noEmit -p packages/lint clean.

Real CLI on a repro project (index.html with one <video id="clip" src="clip.mp4" data-start="0" data-duration="9.019" muted>, clip.mp4 = 1.9 s testsrc2):

$ hyperframes check --no-cache        # main
  0 error(s), 1 warning(s), 0 info(s)   (the one warning is craft's dead_stretch, which doesn't name the video)
◇  Check passed

$ hyperframes check --no-cache        # this branch
  ⚠ video_slot_outlasts_source: Video "clip.mp4" runs out after 1.9s of its 9.02s slot (source 1.9s, starting 0s in). It holds its final frame for the last 7.12s.
◇  Check passed

The freeze itself, rendered on main (-q draft): the frames at 2.5 s and 8.5 s match (PSNR 41 dB, encoder noise only), while 1.0 s vs 2.5 s is 22.6 dB (moving test pattern).

Gates: oxlint / oxfmt clean, fallow audit --base origin/main no issues, comment ratchet at or under share, lefthook pre-commit (format, fallow, typecheck) passed.