[vanceingalls]

What

Adds the three pure modules capture motion blur is built from, under packages/core/src/runtime/captureMotionBlur/:

• pose.ts: an element's box as a 2D affine in composition space (layout offsets plus every CSS transform up the ancestor chain), and the in-between poses a shutter sweeps through. It reads the individual translate, rotate and scale properties in CSS order (translate percentages against the border box), adds the offsetParent's border back and subtracts ancestor scroll, so the pose matches getBoundingClientRect for a box centered with translate: -50% -50%, a turned box or a box in a scrolled container.
• discover.ts: finds the elements that move fast enough for a smear to show (inline-style changes and CSS / WAAPI animations reaching 2 px per frame, outside any 3D context, outermost of nested candidates only). An element on an offset-path, or animated through offset-distance, is no mover: the pose has no offset-path term, so blurring it would only paint the element twice with no smear.
• composite.ts: the per-frame draw. It crops each mover from a capture of the scene and redraws it at every sample pose into a float buffer, additive at weight 1/n.

Nothing calls them yet, so this PR changes no behaviour.

Why

Part 1 of 9 in the capture motion blur stack: #5875 → #5967 → #5968 → #5969 → #5970 → #5971 → #5990 → #5991 → #5992.

The whole stack delivers opt-in motion blur for compositions, drawn with Chrome's html-in-canvas (drawElementImage), in Studio preview and in renders. A composition turns it on with data-motion-blur="on" on its root; nothing blurs without that attribute. This first PR is the math and drawing with no wiring, so the later PRs can be read against modules that are already reviewed.

Notes

• The modules are not reachable from any entry point until feat(core): capture motion blur runtime lifecycle #5967.
• The pose composes individual transform properties, but the scan's skip of 3D rotate axes and z translates (individualIs3d, exported here) is wired into discovery in feat(producer,studio): wire capture motion blur into renders and Studio #5969.
• Not modelled by the pose, documented rather than fixed: an element under CSS zoom, a position: fixed element, and content inside <foreignObject>.
• Limits of these modules, all documented rather than fixed: image-space approximation (a mover's box contents move together; ink outside the box is not redrawn); trails are drawn over foreground elements above a mover; a transparent mover carries a halo of its background inside its box; a 2D flip through scaleX = 0 blends as a rotation; sub-pixel left / top steps land in whole pixels (offsetLeft is rounded); a fast child inside a parent that moved earlier is not blurred (only the outermost candidate is kept); SVG elements are not scanned.

Testing

Unit tests next to each module: affine multiply / invert / interpolate, the sweep taps and the individual translate / rotate / scale properties (each compared with its equivalent matrix transform, the translate: 120px and 180px sweep included), a bordered offset parent and a scrolled container (pose.test.ts); mover discovery thresholds and nesting, and that an offset-distance animation or an offset-path element is no mover (discover.test.ts); the compositor's draw calls against a recording canvas fake (composite.test.ts). bun run build, tsc --noEmit in core and the core runtime tests pass. Comment ratchet against origin/main passes; no package size budget changes.

🤖 Generated with Claude Code

[github-actions]

Edit accuracy: accurate 271 (base branch 271), smooth 164 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 (2)

• seqfastkeys-none-px-r0-root-z100: tracking 7.76, pressJump 0.05, drop 0, reload 0, render 0.02, renderKey -, undo true, teleport true / tracking 0.06, pressJump 0.05, drop 0, reload 0, render 0.02, renderKey -, undo true, teleport true / tracking 0.07, pressJump 0.05, drop 0, reload 0, render 0.02, renderKey -, undo true, teleport true
• seqplainkeys-none-pct-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

[miguel-heygen]

Reviewed at full head f78f8a6, bottom of the six-PR stack. No prior human reviews or inline findings in the saved and fresh snapshots.

P2 blocker: pose.ts:196–205 omits CSS individual translate/rotate/scale although discover.ts explicitly selects that motion. An individual translate mover is discovered, but x=120→180 returns pose.e=0 at both times and zero sweep; the equivalent matrix control returns e=150. Those computed properties are separate from computed transform. Compose them in CSS order with transform origin/ancestors, or exclude unsupported motion consistently. This head has no caller yet; #5967 activates the same unchanged pose code, so repair the foundation before shipping the stack.

Reviewed source contracts and the independent exact-module control; synthetic layout/Canvas, not a Chrome pixel proof. The private six-suite run passed124 authored tests and two additional controls reproduced the gaps. Current123 hosted checks pass/skip, and the inspected original core log executes the new19 pure tests. No render/performance/provider call, merge or release.

— Magi (Home)
Verdict: REQUEST CHANGES
Reasoning: Ordinary motion selected by discovery is omitted from the pose calculation once the next PR enables it.

[miguel-heygen]

P2: Read the individual transform properties as well. Discovery explicitly includes translate/rotate/scale, but these remain separate computed properties and are not serialized into cs.transform. With transform:none and translate:120px 0→180px 0, the exact reader returns e=0 for both poses, so the discovered mover has zero sweep and stays sharp. The equivalent matrix control returns e=150. Compose the individual transforms in CSS order (including origin/ancestor handling), and add individual-property versus matrix controls; alternatively consistently exclude unsupported movers. Dormant here, but this unchanged code becomes live in #5967.

[miguel-heygen]

Reviewed exact head af29b40ef15eb11e1c347b1feed2a62e6b398ae0 relative to 852f6f9f6386fa88b434504233449a9babf995dc.

The individual-transform blocker is repaired: pose.ts composes translate/rotate/scale with the transform/origin chain. The independent exact-module individual translate120→180 control now reads pose.e120→180 with nonzero sweep; the equivalent matrix control remains150. This pose production blob is unchanged through the tested stack tip. Original exact-head hosted runtime log executes23 pose tests (core4834pass/1skip overall).

Independent revision-bound review of the nine-PR composition. Earlier approvals/requests covered older heads. Current saved head/review/inline/check snapshot has no fresh duplicate finding and no failing/pending latest check. Stacked broad-suite skips are not executed tests.

Audited: Pose/discovery/compositor: individual transform order, origins/ancestors, scroll/offset geometry, crop/sweep and alpha composition.
Limits: Composed local focused run:196 authored core tests pass; separate WAAPI/CLI44 pass. Synthetic geometry/canvas controls, not fresh browser pixel/performance proof. No app/model/provider call, source fix, merge or release.

— Magi / desktop-review

Verdict: APPROVE
Reasoning: No additional blocker in this own delta. The later #5971 defect must be repaired before landing the complete stack.