[jrusso1020]

What

Tests only, stacked on #7108. Adds 13 fixtures from the 60-composition capture motion blur sweep and one test per defect A to O, each named KNOWN BUG on main: <letter>: ... and pinned by its current symptom, the same way #7108 pins its known bugs. Each one flips when its fix lands.

Why

The sweep found 15 defect classes that the current fixtures do not cover. Pinning them keeps the suite green today and turns each fix into a visible test change.

Related work

Refs #7108. G is skipped: it is the callbacks case #7108 already pins (events="on" leaves the label in the wrong state). N is skipped too: #7108's liveness test already pins the blank frame after play() on the capture tier.

How

• __fixtures__/motionBlur/sweepFixtures.ts: 640x360, 30 fps, 1 s compositions where #m moves x 0 to 480 between 0.25 s and 0.75 s. No text anywhere, so no pin depends on font raster.
• captureMotionBlur.sweep.browser.test.ts: each case runs one check on a fresh page and asserts the tier, that bad tiles exist, that every bad tile touches the named region, and that the worst MAE is within [0.5x, 2x] of the measured value. When a fix lands, delete the case's bug field and it stays as a regression check.
• Checks: test(engine): pixel invariants and fixtures that guard default-on capture blur #7108's invariant 1 (outside the trail) and invariant 2 (inside, mean of 8 shutter samples) where they see the defect. Where they do not, a second check against the engine's 32-sample sub-frame blur over the whole frame (ORACLE in thresholds.ts; threshold 7.55, the geometric mean of clean-frame noise 3.94 and the smallest pinned symptom 14.47). A plain mover passes it on both tiers.

Finding	Test (KNOWN BUG on main: ...)	Check, tier, frame	Worst MAE
A	panel in front of the mover painted over	outside, capture, 14	47.67
B	mix-blend-mode mover blends with nothing	inside, capture, 11	148.63
C	ancestor drop shadow missing	inside, capture, 14	73.02
D	::after outside the box cut off	inside, capture and filter, 14	107.66
E	own clip-path cuts the smear	oracle, filter, 14	27.60
F	fixed child smeared / jumps into mover	outside, capture 11 / filter 11	40.62 / 126.54
H	body > #root > .m stops matching, still frames too	outside, capture, 4	135.67
I	:last-child moves to the filter svg	outside, filter, 11	53.63
J	overflow:hidden clip edge travels	oracle, capture, 11	25.79
K	crossing movers carry each other's ghost	oracle, capture, 15	14.47
M	band at the bottom edge for content rising in	oracle, capture, 14	18.31
O	mover hidden mid-shutter drawn at full strength	inside, capture and filter, 15	50.38

Not pinned:

• G and N: already pinned in test(engine): pixel invariants and fixtures that guard default-on capture blur #7108 (above).
• L (3x zoom): the sweep's symptom is jagged text, which depends on font raster. Text-free stand-ins (sub-pixel stripes, a 1 px checker canvas) also differ from the oracle at scale 1, so a pin could not tell L from ordinary resampling and would not flip on the fix.

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

Linux, headless shell from resolveHeadlessShellPath: bun x vitest run src/services/captureMotionBlur.sweep.browser.test.ts in packages/engine, 17 of 17 passed in 26 s, three runs in a row, and once more in 29 s with every core busy (the flake gate's setup). Values were measured on Linux only; macOS has not been run. oxfmt, oxlint, engine tsc --noEmit, the comment checks, test reachability and fallow audit pass.

🤖 Generated with Claude Code

[heygengenesis]

Verdict: APPROVE · 0 blocker · 0 important · 0 nits · head 5d7156c8

Tests only, and CI is green at this head, including the flake gate on the new file. The pins hold after #7190 and #7191: on main with all three merged, I got 17/17 locally, so neither fix flips or moves an A–O pin.

Scope

Audited all three files: fixtures, the ORACLE / SWEEP_SYMPTOMS thresholds, the 17 cases (15 pins + 2 plain controls) and their regions. I ran the sweep in chrome-headless-shell 157 on Linux, on main+#7149+#7190+#7191 (CI covers this head alone). I also ran the core captureMotionBlur suite on the merged tree: 203 pass. macOS values are unmeasured, as the body says. The body still says "stacked on #7108", which went stale with the rebase.

— Genesis (Claude)