Review Surface — CLIPFORGE

RENDERS — 3

Artifacts for this claim come from docs/state, preserved from the handoff session. Re-render to produce fresh ones.

lernquadrat · demo-standortfinder·felix·9x16DRAFT WATERMARK BURNED IN

CONTACT SHEET — THE REVIEW SURFACE

ZOOM PLAN — 5 SEGMENTS · COMPUTED FROM THE MANIFEST

KIND	FRAMES	SCALE	EVENT	EASING
push	39-57	1.04 → 2.75	e3	spring
hold	57-131	2.75 → 2.75	—	easeInOutCubic
chain	131-149	2.75 → 1.77	e5	easeInOutCubic
hold	149-204	1.77 → 1.77	—	easeInOutCubic
release	204-222	1.77 → 1.00	—	easeInOutCubic

DER SCROLL-STOPP-TEST

92/ 100SHIP

THRESHOLD 85/100 · SET BY REVIEW.MD

Hook & thumbnail truth23/25

Craft & polish17/20

Motion quality13/15

Capture cleanliness15/15

Brand compliance15/15

Narrative clarity, muted9/10

AUTO-FAIL CHECKS — 12

PASSconsent banner: none in any frame
PASSbroken image regions: none
PASSsafe areas (judged at camera 1.0): clean
PASSmax 2 text lines: ok
PASSem dash: none
PASSend card matches audience (felix): ok
PASSmetrics carry evidence (80 → live-site title + hero line, screenshots in out/film/evidence/): ok
PASSDRAFT watermark present while antonio_approved=false: ok
PASSduration 10.2s ≤ 15s: ok
PASSmotion linter passed before review: ok (0 violations)
PASSAI imagery: none (no b-roll in spec)
PASSzoom budget 3/3, every zoom lands on a manifest interaction: ok

DEFECTS BY LOOP — DISPLAY ONLY

LOOP 178

[F83-F166] zoom levels were raw camera multipliers over a 0.56x-scaled device: '2.2x' gave only 1.24x effective content magnification — close-up didn't read close, result text illegible muted → redefined MOTION.md §10 levels as effective magnification M {1.0 medium, 1.55 close}, camera computed per layout (skills/MOTION.md v4.2 + src/compose/zoom/compiler.ts)
[F185-F240] naming line (light) over white page content → expected resolved by fix 1 band geometry
[F83-F166] close band top-anchored left ~45% of canvas dead ~3s → resolved by fix 1 magnification

LOOP 284

[F139-F167] result heading left-cropped: frame showed '9 Standorte in Wien' where the site says 19 — a framing choice may never misquote the site's own proof → anchor visibility clause: anchors must keep interaction+result union in the visible window (skills/MOTION.md §10 v4.2 + compiler pickAnchor)
[F194-F240] naming line over white again: result anchor float-tie-broke to the LOWER thirds row pushing the band over the naming zone → result targets now pin to the upper thirds row (results read downward)

LOOP 392

No defects.

Would I stop scrolling: yes — the poster reads at 300px, and the close-up beat (pill flips lime, site renders '19 Standorte in Wien' in-frame) is a genuine money shot. Weakest moment is the ~0.9s wide idle between hook exit and push (F48-F75): rule-compliant, slightly plain; tightening establish to ~1.0s is a taste option for the full build, not a blocker.

NOTES

Loop 1 and 2 contact sheets were reviewed live and then overwritten by the next render (single sheet path); loop 3 sheet is preserved. Full build will persist per-loop sheets.
Latency represented honestly: 1372ms measured; capture-during-wait recorded 2 real page states; result pops at its true elapsed frame (never early).
Ripple verified on final MP4 frames 101/103/105 (compose-side, zooms with the page).

client_permissionTRUE

antonio_approvedPENDING

OPEN APPROVAL