[vanceingalls]

What

Motion blur now reports WHAT a composition holds, as counts, in the render log and in render telemetry, so the next failure class shows up as a rate ("films with a mover under a panel") before anyone files a bug. Five of the six blur bugs so far came from a composition feature no fixture held.

• MotionBlurFingerprint (one definition in core/src/motionBlurFingerprint.ts, imported by engine and cli): canvas2d, webgl, video, iframe, customElement, svgClip, mask, scrollContainer, threeD, overflowingMover, frontOfMover, blendMode, fixedInMover, pseudoOutside, all numbers. Computed once per scan in discover on both tiers, never per frame; the element walk is capped at 4096 nodes and the per-mover descendant walk at 256, so the counts are lower bounds. holdsFixedChild (now in frames.ts) and pseudoBox are reused. Report.fingerprint (null before a scan) and Report.readEvents.
• frontOfMover counts movers that something else inside the composition root sits above at any sampled scan step (elementsFromPoint at the mover's centre every ceil(steps/24) steps, at most 25 samples per candidate, a counted mover is not tested again; the tier's own canvases and the filter svg sit outside the root so never count). The first version sampled once at the restore pose and read 0 for anything entering after t=0: reviewed and changed before this PR. Real Chrome 153 on a mover passing behind a panel (behind-panel): the old code printed frontOfMover 0 (and, with two capture workers, a worker-dependent 0 or 1); this code prints frontOfMover:1 in every case.
• Log line: formatPageMotionBlur appends readEvents=true (only when true) and fp=webgl:1,video:2 (non-zero counts, field order); the existing [motion-blur] on reasonCode=... movers=... failedFrames=... tier=... text is byte-identical when there is no fingerprint.
• Render result: RenderPerfSummary.motionBlur?: { tier, reasonCode, movers, failedFrames, readEvents, fingerprint }: first reporting worker wins, failedFrames summed. failedFrames is fresh at session end because the session holds one pageMotionBlur object that getCapturePerfSummary hands out and closeCaptureSession updates in place.
• Telemetry: trackRenderComplete (CLI renders and Studio renders) sends motion_blur_tier, motion_blur_movers, motion_blur_failed_frames and one motion_blur_fp_<snake_case_field> per non-zero count, nothing when blur never ran. readEvents is not sent.

Privacy (please check this part)

Counts only: the runtime only increments numbers; the engine rebuilds the object from the 14 known keys with Number(x) || 0 (a test feeds id: "secret" and video: "many" and neither gets through); the CLI iterates the same field list. moverNames is never forwarded and reasonCode is not sent to telemetry. The existing opt-out (trackEvent returns when !shouldTrack(): HYPERFRAMES_NO_TELEMETRY, DO_NOT_TRACK) and the Studio telemetryOptOut path cover the new properties without extra code. Cardinality: 3 scalars plus at most 14 small integers under fixed names.

Tests

core captureMotionBlur 242 pass (fingerprint: one composition per field, carries no names, the cap, the sampler cases, onFast called once per scan step on the first pass only), engine 80, producer 64 (vitest) + 86 (bun test), cli 219; tsc clean in all four packages. A real producer render of webgl-canvas logs [motion-blur] on reasonCode=tier-auto movers=1 failedFrames=0 tier=filter fp=webgl:1; a forced drawElementImage failure across two workers gives failedFrames 30 and 31 at session end, summed to 61 in the perf summary and in the telemetry payload (mapping run offline; the durable proof is the two-worker aggregation test and the in-place close test). Per-file flake timings 1 to 6 s.

Limits

• frontOfMover is a lower bound: a covering overlay with pointer-events: none (vignette, light leak) is not in elementsFromPoint; a mover passing behind a thin cover between two samples is missed; a transparent full-size container with pointer events counts as covering.
• Past the 4096-node cap the counts undercount silently.
• Cost on a 10,000-node page is not measured; the sampler is bounded at candidates x 25 calls (2650 for a 106-mover film).
• Deferred minors from review: canvas2d includes the capture tier's own canvases on a rescan while wrapped; the recorder keeps the latest failedFrames rather than the largest (a capture-to-filter hand-off mid-session resets it to 0).

Size budget. @hyperframes/inspector unpacked goes 3,645,432 to 3,684,092 B in scripts/package-size-budgets.json: the runtime fingerprint code is bundled into inspector, which put the build 2.2 KB past the 1% tolerance. docs/packages/cli.mdx now lists the composition counts under what telemetry collects.

🤖 Generated with Claude Code

[github-actions]

Edit accuracy: accurate 267 (base branch 267), smooth 107 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 (8)

• seqplainkeys-none-center-r0-nested-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): pass, passes
• seqplainkeys-none-center-r0-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): fail / pass / pass, passes
• seqplainkeys-none-center-r30-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run
• seqplainkeys-none-pct-r0-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): fail / pass / pass, passes
• seqplainkeys-none-pct-r30-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run
• seqplainkeys-none-px-r0-nested-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): pass, passes
• seqplainkeys-none-px-r30-nested-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run
• seqplainkeys-none-px-r30-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run

[heygengenesis]

Verdict: REQUEST CHANGES · 1 blocker · 1 important · 1 nit

Privacy holds: only counts reach render_complete, the engine rebuilds the 14 keys, tier is narrowed to capture/filter, and both CLI and Studio go through shouldTrack/telemetryOptOut. No names or reasonCode leave.

• blocker: CI Build is red at verify:packed-manifests. @hyperframes/inspector unpacked is 3,684,092 B against a 3,645,432 B budget, about 2.2 KB past the 1% tolerance. Main's Build is green, so this PR causes it. Raise scripts/package-size-budgets.json and give the reason in the body (the runtime fingerprint code is bundled into inspector), or trim.
• important: docs/packages/cli.mdx:2267 lists each telemetry category. Counts of what a composition holds (videos, iframes, canvases, custom elements…) is a new category. Add it there so the disclosure matches what the CLI sends.

Scope

Read the full non-test diff and the CLI/engine/producer tests. Locally, core captureMotionBlur (181), engine pageMotionBlur (20) and cli renderMotionBlur + studioRenderTelemetry (20) pass; I did not run producer. The Docker CLI path sends no blur properties because it has no perf summary. The producer server's HTTP response and log.info carry the full summary, reasonCode included; that is not telemetry. I am deferring to a CLI-telemetry owner on the second-eyes ask. CodeQL, regression and Windows were still running.

— Genesis

[heygengenesis]

nit: Number(x) || 0 still lets page-supplied Infinity, negatives and fractions through. movers (line 82) gets no sanitizing at all, so NaN can reach motion_blur_movers. Clamp both to finite non-negative integers so "small integers" holds at this layer too:

Suggested change

	const count = (field: string): number => Number(Reflect.get(raw, field)) || 0;
	const count = (field: string): number => {
	const n = Math.trunc(Number(Reflect.get(raw, field)));
	return Number.isFinite(n) && n > 0 ? n : 0;
	};

Do the same for movers.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
hyperframes	🟢 Ready	View Preview	Nov 6, 2026, 9:53 AM

💡 Tip: Enable Automations to automatically generate PRs for you.

[heygengenesis]

Verdict: APPROVE · 0 blocker · 0 important · 1 nit · 2 resolved, 1 open (nit) · 6ae6059 → 473b458

✅ Inspector budget raised to the measured 3,684,092 B, with the reason in the body. CI / Build is green on 473b458.
✅ docs/packages/cli.mdx now lists the composition counts under telemetry.

The Number(x) || 0 clamp nit on pageMotionBlur.ts:63 is still open and optional.

— Genesis