[vanceingalls]

What

The filter tier appended its hidden <svg><defs> (the per-mover SVG filters) to the composition root and left it there until teardown, so #root > :last-child or :last-child selectors on the root's children matched the injected svg and a stray square could draw at (0,0) (sweep defect I: 1980 px differ from the oracle). The svg now goes under document.documentElement instead (not document.head: a display:none subtree can stop a filter reference resolving in Chrome). One line in filterTier.ts layerOf.

Removal still works from the new parent (teardown calls defs.remove()), and at most one filter tier exists at a time, so filter ids (hf-mb-<n>) cannot collide across two svgs.

Tests

• index.test.ts: keeps the filter tier's svg out of the root and removes it on teardown: none inside the root, exactly one in the document after a paint, still one after re-init, zero after teardown.
• The filter-tier last-child-selector sweep case loses its bug field and S.I and passes as an ordinary outside-the-trail check. The plain filter case in the same file is what proves the filter reference still resolves from documentElement.
• With only the one-line move reverted, the new unit test and the un-pinned sweep case both fail. Core captureMotionBlur 213 pass; engine sweep + differential 90 pass on Chrome 153.

Limits

• The svg now sits under <html>, outside <body>, while the filter tier is active. Nothing in core, engine, cli or studio clones or serialises the live documentElement.
• html > :last-child and body:last-child author selectors now see the svg as a sibling of body; far less likely than the #root case.

🤖 Generated with Claude Code

[github-actions]

Edit accuracy: accurate 267 (base branch 267), smooth 123 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): fail / pass / 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): 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): 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): fail / pass / 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: APPROVE · 0 blocker · 0 important · 0 nits · head fe037536

Scope

• Read filterTier.ts end to end. teardown removes the svg from any parent, and the tier was already building it with the global document (same as createElementNS and landThenUndoDom(…, document)), so moving it under documentElement adds no new document assumption. Nothing else in captureMotionBlur/ appends to root. The canvas tier wraps the root and doesn't add children to it.
• Checked locally: core index.test.ts passes 125/125. With only the one line reverted, the new test fails.
• CI: the first Studio: edit accuracy (14/15) failure was a puppeteer WS-endpoint launch timeout, unrelated to this change. After a rerun the whole CI run is green. captureMotionBlur.sweep.browser.test.ts ran in Test (17 passed). Windows render verification and Safari perf were still running when this was posted.
• The html > :last-child / body:last-child exposure is documented in the PR body. I accept it as is.

— Genesis