What

• Defects are decided once per timeline. The mover scan already seeks the whole timeline, so the defect scan samples along it: a defect that only appears mid-timeline (a root that rotates, a card that flips) turns blur off from frame 0, not partway through a render. Defects now apply on every Chrome version, because the preview, the pinned render browser and Docker's browser can differ and a render must match what the preview showed. GSAP's identity matrix on a rewound root transform no longer counts as a transform.
• The per-frame check is a backstop that costs one frame. For what the scan missed (a root shake shorter than the scan step, a backface flip between samples, a canvas a script adds later) the frame renders sharp from the plain DOM and the runtime logs once; blur resumes when the defect is gone, instead of ending for the rest of the render. Measured with the review fix on the CLI render path (Chrome 152, quality draft, frames compared with blur off), before the change: a 1-frame shake sharpened 179 frames of a 6 s 60 fps clip, a 60 s clip with a 10.05–10.2 s shake sharpened 598 frames; after, 1 and 4 frames.
• Four detector gaps closed. A root moved by the individual scale, rotate or translate properties (identity values ignored) is a root transform; a root mix-blend-mode other than normal is a defect (root-blend-mode); the backface check treats an individual rotate with an axis or 3D angle (rotate: y 180deg, rotate: 0 1 0 180deg) and a translate with z as 3D (shared individualIs3d); and the iframe-in-root entry covers everything that is torn down and re-initialised when the root moves: <iframe>, <object>, <embed> and defined custom elements (hyphenated name with customElements.get). Measured on Chrome 152 against the previous build: a scale: .5 root, a rotate, a translate, a multiply root, a rotate: y 180deg face, <object>, <embed> and a defined custom element all reported on and now report off with their code; plain compositions and flat uses (scale: 1, rotate: 90deg on a face outside 3D) still report on.
• A block raised by the scan is lifted when its cause is removed or the timeline changes. Capture-failure blocks stay for the life of the instance.
• The WebGL / WebGPU check covers the whole document: a render's shader-transition compositor appends its canvas to <body>, while preview puts it inside the root.
• Blur is off on routes that cannot host it. HDR renders and the layered compositor screenshot each DOM layer after hiding the others, which does not reach blur's canvases; a distributed (Lambda / cloud) chunk render has no Studio preview to match and no guarantee of html-in-canvas in the worker's Chrome. The producer sets pageMotionBlur to hdr-render, layered-composite or distributed-render there, and the runtime reports the renderer turned it off (…). The distributed plan's compile-time hint no longer forces a data-motion-blur composition onto screenshot capture: it freezes the same capture mode as the same composition without the attribute.
• The mover scan reads its baseline at t=0, so every worker of a parallel render finds the same movers whatever time its scan starts. An ancestor's pan is not counted as its child's move (a first change is judged against the element's own baseline style at that time).

Why

Part 7 of 9 in the capture motion blur stack: #5875 → #5967 → #5968 → #5969 → #5970 → #5971 → #5990 → #5991 → #5992. It builds on the defect list in #5971.

Stack goal: opt-in capture motion blur for compositions, in Studio preview and in renders, drawn with Chrome's html-in-canvas. A composition must blur in every frame the scan trusts, and be sharp in the rest.

Notes

• Defects apply on every Chrome version, so a composition that Chrome 157 would capture correctly still has blur off if an older preview browser would not. Revisit when the render pin and the support floor both pass the last affected major.
• Enabling blur runs a timeline scan that can leave about one level of noise inside blend-mode / filter effects even when blur ends up off.
• Documented, not fixed: an ancestor's pan can still be read as a child's move in some edge cases; the per-frame backstop re-reads backface on its cached candidates, which costs more when a stylesheet resets *{backface-visibility:hidden}. The candidates include the animation-derived faces (feat(core): block capture motion blur on html-in-canvas defects #5971); each rescan adds to them and they are kept for the rest of the scan, because the last sample can fall after an animation has ended (the per-frame backstop then still reads them).
• Output is not bit-identical across worker counts, at about the size of the chunk-split noise a render without blur already has. Motion shorter than one scan step (SCAN_STEPS) is under-measured.
• The HDR / layered / distributed routes render without blur, as before this stack existed.

Testing

• htmlInCanvasDefects.test.ts: the scan finds a defect that appears at any sampled time and says where it first appeared; identity transforms are ignored; a root scale / rotate / translate (identity values ignored), a root mix-blend-mode, <object> / <embed> / defined custom elements (an undefined hyphenated tag is ignored), and a backface-hidden face turned by rotate: y 180deg, rotate: 0 1 0 180deg or a z translate (a flat rotate: 90deg is not); the document-wide WebGL check; facesNow reads the scan's backface candidates at the current time (a hidden face that flips into 3D after the scan is seen without rescanning; a face an animation keys to hidden is a candidate, stays one after the animation is gone, and is found by the scan and by facesNow).
• Core captureMotionBlur/index.test.ts: blur off from frame 0 for a mid-timeline defect; a defect the scan missed renders only that frame sharp and blur resumes; an engine frame that is sharp resolves at once and leaves the composite step idle; a block lifts when its cause is gone or the timeline changes, and a capture-failure block does not; the distributed route is named as the reason.
• discover.test.ts: the same movers whatever time a scan starts, an ancestor's pan is no child's move when the child's style first changes, and the scan reports each step after its seek.
• Engine pageMotionBlur.test.ts and producer motionBlurRoute.test.ts: the off reason for the HDR, layered and distributed routes; producer distributed plan.test.ts (the chunkFrames.test.ts for every chunk session is in feat(core,engine): report why blur is off, and what each render session did #5991): the plan probe session and every chunk session stand blur down, and the frozen capture mode matches the plain composition's; orchestrator request validation.
• bun run build, tsc --noEmit in core, engine and producer pass; comment ratchet against feat(core): block capture motion blur on html-in-canvas defects #5971's branch passes; every package is within its size budget.

🤖 Generated with Claude Code