[miguel-heygen]

What

Scrubbing a long film in Studio no longer slows down as the film gets longer. While the playhead moves, the editor's layer list skips the scenes the runtime has hidden, so each update costs what is on screen, not the whole document.

Why

Two consumers rebuild the layer list from collectDomEditLayerItems while a scrub mutates the preview:

• the off-canvas indicators (useOffCanvasIndicators), which are mounted whenever the DOM edit overlay is, and rebuild at most every 100 ms on any DOM mutation;
• the Layers panel, which re-collects on each playhead tick while its tab is open.

Each rebuild visited every element of every scene and called isInspectableLayerElement on it, which reads its computed style. On a film with 230 scenes of 600 elements that is about 276k style reads per rebuild, almost all of them in scenes that are not on screen.

The runtime already hides everything inside a clip its visibility pass hid: it marks the clip data-hf-clip-hidden, sets its inline visibility: hidden, and a stylesheet rule hides its contents (packages/core/src/runtime/timedClipHide.ts). Nothing inside such a clip can be a layer, so the walk can stop there.

How

• isClipHiddenByRuntime(el) in @hyperframes/core/studio-preview-mark: the clip has the mark and an inline visibility of hidden. Both are required: older shader builds reveal a clip through its inline style only and leave the mark, and such a clip is on screen.
• collectDomEditLayerItems returns early on such a clip.
• CLIP_HIDDEN_ATTR moves to studio-preview-mark so the runtime and Studio read one constant; clipReveal.ts re-exports it, so runtime imports are unchanged.

What changes in output

None. This is Studio editor code only; render and snapshot are untouched. The layer list and the off-canvas indicators list the same items as before: isInspectableLayerElement already dropped everything inside a hidden clip, because its computed visibility is hidden. The early return only skips reading those styles.

Test plan

• Unit tests added/updated

• Manual testing performed

• Documentation updated (if applicable)

• Comments follow CONTRIBUTING.md "Comments": they say why, not what, and a bug fix says what the code must do and how to reproduce the bug

• domEditingLayers.test.ts, "collectDomEditLayerItems and clips the runtime hid":

  • lists what is on screen and reads the computed style of nothing inside a hidden clip with 50 children. Without the early return it reads 51.
  • still lists a marked clip whose inline style shows it.
  • Passed 3 runs in a row.

• packages/studio src/components/editor: 155 files, 1,974 tests pass. packages/core timedClipHide and clipReveal tests pass.

No visible change

13 changed non-test lines, no .tsx, .css or .html. The Layers panel and the off-canvas indicators show the same items; only the time spent finding them drops.

Why under 100 lines: Miguel's asap P0 for the next Desktop build. One early return in one walk, plus the predicate next to the mark it reads; it shares nothing with any other open change.

[github-actions]

Edit accuracy: accurate 271 (base branch 271), smooth 149 of those

The gate passes.
Smoothness is reported in the artifact, not gated. A case fails only if it fails 2 of 3 runs.

Quarantined, measured but not gated (0)

Unstable (1)

• seqplainkeys-none-px-r0-root-z100: tracking 0.06, pressJump 0, drop 0, reload 0, render 0.03, renderKey -, undo false, teleport true / tracking 0.06, pressJump 0, drop 0, reload 0, render 0.03, renderKey -, undo true, teleport true / tracking 0.06, pressJump 0, drop 0, reload 0, render 0.03, renderKey -, undo true, teleport true

[heygengenesis]

Verdict: REQUEST CHANGES · 1 blocker · 0 important · 0 nits · head aa22718

blocker: the required Studio and player captures check fails on this head. The body has no ## Before/## After section and no ## No visible change section (scripts/check-pr-captures.mjs). This change qualifies for ## No visible change: 13 non-test lines and no .tsx/.css/.html. It is also accurate. Before this PR, isInspectableLayerElement already dropped a runtime-hidden clip and its contents, because their computed visibility is hidden, the clip through its inline style and its contents through the always-installed [data-hf-clip-hidden][data-hf-clip-hidden] * rule. The early return only skips the style reads. So the line "A hidden scene's elements no longer appear in them" describes no change. This is a body-only fix and the head stays the same.

I found no problem in the code. Every other workflow on this head is green, and domEditingLayers.test.ts passes locally (26/26).

[heygengenesis]

Verdict: APPROVE · 1 resolved, 0 open, 0 new · aa22718 → aa22718

✅ The body now has ## No visible change and no longer claims an output change. A re-run of the original run used the old event payload, so I re-ran the edited captures run. It passed, and so did every other workflow on this head.

[miguel-heygen]

Measured in the real Desktop app, before and after this change (the app is the only consumer that matters for the scrub target): a 23-minute film of 276,460 elements, a quiet Mac, three 6 s drags of the seek knob with a CPU profile of each.

	before (app on Studio 0.8.202)	after (this change, Studio 0.8.204)
seek knob drag, cold	9.3 fps	13.8 fps
seek knob drag, warm	13.0 / 14.1 fps	37.5 / 47.0 fps
film player seeking once per frame, warm	5.2 / 5.7 fps	33.5 / 35.5 fps
seek moves served in 6 s, warm	26 / 28	72 / 85

In the warm profile the layer walk went from about 65% of the sweep to about 45% (the selector index and findClosestByAttribute, addressed by #6427, plus the composition re-read, #6429), with the browser's own style and layout at about 20%.