[WaterrrForever]

What

hyperframes diff no longer counts a jump in yours as an extra cut when the original moves just as hard at that moment. Those go to not_cuts as fast.

Why

From the reverse-any-video feedback. A regenerated shot never lines up with a whip frame by frame, so diff listed 10 to 16 extra cuts on the museum film, all inside its fast moves. Those hid the real problems: the code-first render's visible purple block was one line among 10.

How

cutsBeyondMotion compares each unmatched cut in yours with the strongest change in the original within a quarter second. It stays an extra cut only above OVER_MOTION (1.5x) of that. The original's own cuts are left out of the comparison, otherwise a cut a few frames late would hide behind the real one. missing does not change.

Extra cuts from the real diff, on 8 renders. Before is after plus the cuts moved to fast; none of these originals has a fast of its own:

Render	Before	After	Still reported
museum, generated	16	2	53.38, 53.75 s: jumps with no match in the original
museum, code first	10	2	21.67 s (a join), 32.13 s (the purple block)
museum, current skill	13	3	49.75, 49.92, 50.58 s: around a late cut
ali	4	1	14.48 s
johnny	1	1	27.65 s: a late cut
ig03	6	3	2.77, 9.37, 17.67 s
astra	1	1	30.97 s
cindy	1	0

The other bench renders had no extra cuts, so this rule cannot change them.

Test plan

fails without the fix: diff.test.ts "reads a change no stronger than the original's motion as the same fast move"

• diff.test.ts, diff-segments.test.ts, diff-people.test.ts: 35 passed, including "reports a cut a few frames late as missing and extra".
• 15 real cuts spliced in at quiet and fast moments of the museum film, 7 in ali, 6 in ig03: all 28 the detector found stayed extra cuts, including those inside whips. Two splices were not found by the detector itself.
• By eye, 6 of the extras it dropped on live footage (ali, ig03, cindy) show the same change in the original.
• OVER_MOTION is tuned on these films. A weak real cut inside a fast move can still pass as motion.

[jrusso1020]

Reviewed at d29fb1537840d9593af85f0e5b4589a8d10f683a.

The motion comparison is applied only to unmatched cuts, leaves missing cuts unchanged, and excludes original cut spans from its local motion baseline.

Need: yes · Size: ok, two files · Reuse: classifiedCuts, frameChanges and existing cut matching · Simpler: already minimal · Structure: fine
Risk: low — local diff diagnostic/classification change, no generation or billing path changed.

Validation: node node_modules/vitest/vitest.mjs run src/measure/diff.test.ts src/measure/diff-segments.test.ts src/measure/diff-people.test.ts: 35 passed on this exact head, including the existing delayed-cut regression. Disabling the new unmatched-cut motion filter makes its suppression witness fail while its quiet-baseline/true-cut and delayed-original-cut unit controls still pass. Independent source-module controls cover 24/30/60 fps windows, first/last transitions and multi-frame original-cut exclusion.

The 1.5 threshold is a disclosed heuristic: weaker real cuts during fast motion can be suppressed. I did not rerun the eight-render museum/bench corpus or validate its reported counts.

Verdict: APPROVE
Reasoning: No blocking regression found in the changed path and its callers. The limited scope and disclosed validation limits support landing.

— hf-review

[jrusso1020]

Reviewed at ddbd50d9faa4cb1b95a5fca9603462e73e3277fc (merge-only re-review).

The resolution preserves both reviewed changes: cutsBeyondMotion and cutReport are byte-identical to approved d29fb1537; rateWarning, projectRate and their warning insertion are byte-identical to approved #6976 at 029fc5ee4. Against the merged main parent, only the two original diff files change.

Need: yes · Size: ok, two-file increment · Reuse: existing cut classifier and composition FPS parser · Simpler: already minimal · Structure: fine
Risk: low — combines the two previously reviewed diff changes without changing their logic.

Validation: node node_modules/vitest/vitest.mjs run src/measure/diff.test.ts src/measure/diff-segments.test.ts src/measure/diff-people.test.ts: 38 passed on this head, including frame-rate warnings, motion suppression and delayed cuts. Independent source review found no merge-resolution blocker. No film-corpus or full browser project-capture rerun; the disclosed 1.5x heuristic limitation remains.

Verdict: APPROVE
Reasoning: Both implementations and their tests survive the resolution; no new blocker found. Unrelated main changes are inherited, outside this focused merge review.

— hf-review