# Builder inspection: limits retained

Actual review objects: both new 1920×1080 Remotion movies and the retained
960×540 silent source excerpt. Every local frame L0–L59 was independently
decoded and placed in six labeled full-frame sheets. The title crops L10–L29
were additionally opened to inspect the entire underline construction. The
builder viewed all six full-frame sheets and those two title sheets. No
continuous moving-source or blind aesthetic judgment is claimed.

## Observed states

- L0–1 have no title. The lead arrives whole-word at L2 and L4.
- The first keyword character remains through L7. Letters advance at L8–L14.
- L13 shows the first underline with an incomplete seven-character keyword;
  the word completes at L14 while that stroke is still extending.
- L18–19 adds a close opposing return without removing the first path.
  L20–24 adds the irregular retrace over both. All three persist to L59.
- The later frame sheets show changed corner/material positions while the title
  remains. The late title enlargement does not enlarge the background with it.
- No fade to blank or premature title departure occurs before the final frame.

These observations agree with the exported per-frame controls. All 177 adjacent
background-crop comparisons across the three movies have pixel differences.
That last count alone can include codec noise; it is not proof of perceived
motion or an aesthetic score. Independent changes in the prescribed x/y/angle
tracks are also checked for every background branch on all 59 frame intervals.

## Differences, not waived source scores

The source excerpt says “Helps with homework”; the existing replacements say
“Build new pathways” and “Work with drawings”. The font, glyph footprints,
original corner art, field colors and drafting marks are substitutions. Source
equations are absent. The replacement right curl and folded surfaces do not
have the same silhouettes, shading or source texture. Their filled gesture
curves also differ from the source's visible returns. The source excerpt is
half the nominal source dimensions, limiting fine-edge comparisons.

No source-vs-candidate pixel threshold was invented or relaxed to accept those
differences. The zero-difference result is strictly **new replay versus retained
review02 candidate**. Each of the two movies is byte-identical and every RGB24
and YUV420P decoded frame matches. The frozen critic inputs must not be replaced
by a later design iteration under these hashes.

## Preparation and controls

Actual preparation produced 36 RGBA PNGs. The initial preparation and r02
preparation have identical PNG bytes. R02 adds explicit bound font provenance
and adapter identity, then verifies it along with browser identity and unique,
complete, correctly sized/hashed PNG inventory. The initial insufficient
provenance record remains in the private evidence; it is not used as a verified
public preparation record.

Geometry controls and the guide preserve ten separate leaf layers plus their
title parent; no combined screenshot replaces the background or emphasis
branches. Exact zero source error, native Resolve execution and moving-quality
approval remain unclaimed. The separate critic's result is not anticipated here.