# Gallery-to-device: recorded variant results

September 15, 2026

## Recorded finding G1

**Previous adaptation** and **Previous default** have the recorded
intermediate-state separation problem: note thumbnails and text overlap, then
the following gallery heading crosses the notes. **Current adaptation** and
**Current default** keep separate note rows and a gallery below them at the
matched inspected states.

| Named variant | Preserved result |
| --- | --- |
| Current adaptation | Matched native-detail witnesses retain separate notes and gallery. The common source-fidelity differences remain. |
| Current default | Matched native-detail witnesses retain separate notes and gallery. The common source-fidelity differences remain. |
| Previous adaptation | G1 is visible in the inspected intermediate states. A later native-detail witness shows separate notes and gallery again. |
| Previous default | G1 is visible in the inspected intermediate states. A later native-detail witness shows separate notes and gallery again. |

G1 is directly supported at retained indices 64, 70, 74, 75, 78, and 84
(approximately 2.13–2.80 seconds). At index 70, the second and third note lines
intersect; at 74, 75, and 78, the gallery heading crosses the lower notes; at
84, the gallery remains above or against the final note. Index 90, at 3.0
seconds, is a clear later witness. These are witnessed states, **not a claim
that the exact first and last failing frames have been established**.

The gap remains the same note/gallery separation finding, assessed through
device-versus-body relationships, ordered section reveals, boundary behavior,
and legibility. It is not a new diagnosis of implementation or motion quality.

## Source differences remain

All four variants retain the previously recorded differences from the source
in opening and collage relationships, central image alternatives, relative
scales, section reveal order and positions, and clipping. The source's late
independent leftward image-strip displacement was not observed in the
inspected late variant states.

Artwork and language differences are source-fidelity observations, not
automatic product defects. The recorded text intersections are a separate
legibility issue that does not depend on the chosen theme.

**No overall winner or new ranking is established.** Separated current-variant
witnesses do not establish full source equivalence or all-frame correctness.

## Inspection scope and confidence

- Each variant previously had all 135 retained indices inspected in nine
  chronological sheets, plus 16 native-detail center crops. The crops are
  960 × 1080 from 1920 × 1080 movies, without spatial scaling; they do not
  cover the full movie width. The sheets supply reduced full-frame context.
- The source baseline previously covered all 135 indices in reduced sheets
  plus 26 retained-resolution witnesses at 960 × 540. Identical source
  material reuses that coverage without additional viewing credit.
- The verified retained clock spans 4.5 seconds at 30 frames per second.
  Full-master recovery is not claimed.
- Confidence and scope are unchanged: the listed states have native-detail
  support, while other indices were inspected only in reduced sheets. No new
  confidence score or exact failure interval is introduced.
- No moving playback, native-pixel-full coverage, or full-film viewing credit
  is claimed. Stills do not establish perceived smoothness, easing, pace,
  judder, or audio experience. No input, causality, or workflow completion is
  inferred.
- This is the same study. This report adds **zero new visual inspections,
  zero new viewing credit, and zero additional unique-study credit**.

Selection and publication remain outside this report.

## Selected public revision

The current variants above are R02; the previous variants are R01. This packet selects R02 default and adaptation. The source and earlier repeated comparison frames add no unique-example or whole-film coverage credit. The builder separately records faster vertical steps around frame 65 as an unresolved motion tradeoff, not a perceptually validated improvement. No new rank or primary-gap finding is inferred after reconciliation.