# Source-bound treatment review

## Outcome

Role reconciliation does not change the independently established criteria,
findings, confidence, or evidence limits. Neither the default nor the adaptation
is preferred in any example. There is no overall ranking or winner. Moving
wins remain **0**. This review does not approve the quality of the source
footage, certify the finished kits, or replace technical review.

Three examples have a shared remaining gap in both outputs. For image isolation
and recombination, no additional material gap or comparative discriminator was
established. Different subjects, illustration choices, wording, and colors are
not failures by themselves.

## Evidence and limits

The source criteria were frozen before output inspection. Before that source
phase, no specific familiarity with these exact clips was recalled; general
model priors and possible training exposure cannot be excluded. Subsequent
source familiarity was reused openly, not represented as a fresh baseline.

All 1,413 retained packet indices are accounted for: 942 newly inspected output
indices and 471 previously inspected, byte-identical source indices. Reused
source coverage receives no additional viewing credit. The new outputs were
inspected in order on 40 full-context thumbnail sheets, with uncropped 384×216
tiles, followed by 22 selected native-resolution frame witnesses. The source
baseline comprised 20 thumbnail sheets and 13 native-boundary sheets, including
47 repeated native frame indices. Five source-sheet reopens added no coverage.
The output-review phase recorded 67 actual image-view events.

These are ordered still-image observations, not moving perception or
native-pixel-complete inspection of every frame. Native witness files were not
resampled during extraction, but display scaling does not guarantee that every
native pixel was individually resolved. No audio was auditioned; the inspected
files' probes reported no audio streams. No genuine moving surface was used.
Playback readability, smoothness, easing, realtime causality, and whole-film
experience remain unassessed. Generation, decoding, probes, and playback
callbacks are not counted as perception.

Some detached-preview and simulator frames contained variant-describing
captions. Those captions limited perfect blinding before role reconciliation;
they were not used as evidence of quality or to establish a preferred result.
Reconciliation added no viewing events and did not reopen the judgments.

All witness indices are zero-based. Times are local actual-movie presentation
timestamps, not original-master event times. Fractional times are exact; decimal
times below are rounded. The inspected bytes and integer frame clocks were
verified during the sealed output review. Later role reconciliation matched
those sealed file hashes; it did not inspect additional footage.

| Example, both outputs and reference | Retained indices | FPS | Last local PTS | Exclusive end |
| --- | --- | --- | --- | --- |
| Detached working preview | 0–95 | 24 | 95/24 s | 4 s |
| Gallery to device | 0–134 | 30 | 67/15 s | 9/2 s |
| Image isolation and recombination | 0–119 | 30 | 119/30 s | 4 s |
| Simulator selection | 0–119 | 24000/1001 | 119119/24000 s | 1001/200 s |

The full masters were absent. Retained master probes and rounded timestamp
joins do not establish fresh visual correspondence to original masters. No
unseen product behavior, successful edit, generation, or applied code change is
inferred from these excerpts.

## Detached working preview

**Single biggest remaining gap: the local pointer/control cue before separation
is missing in both the default and the adaptation.** No material discriminator
between the two outputs is established.

In both outputs, native frame 31, PTS 31/24 = 1.291667 s, shows the preview inside
its host without a visible corner control or pointer. At frame 38, PTS 19/12 =
1.583333 s, the preview straddles the retreating host edge without that cue.
The ordered thumbnails around this interval do not show a corresponding
pointer/control stage. The reference has changing corner controls at frames
31–33 before the frame 37→38 separation boundary. This is a gap in visible
agency and context, not proof that a particular click or drag is required.
Tiny details outside the native witnesses remain bounded by thumbnail resolution.

Both outputs preserve the host-to-embedded-to-detached relationship, retain the
diagram's identity across separation, and visibly change the internal page at
frame 72 = 3 s while keeping the preview window. Neither is merely a frozen
detached screenshot. The route and material diagrams, page contents, and
wording are meaningful original-content differences, not reasons to prefer one.

The judgment preserves the frozen criteria for contextual detachment, preview
identity, and the local control cue. It does not require the reference wallpaper
or application. **Confidence:** high for the witnessed omission; moderate for
its material effect on explanatory clarity, with playback impact unassessed.

## Gallery to device

**Single biggest remaining gap: staged sections collide before settling in both
the default and the adaptation.** No material discriminator between the two
outputs is established.

At frame 65, PTS 13/6 = 2.166667 s, suggestion labels and icons occupy overlapping
row space. The decisive native witness is **frame 78, PTS 13/5 = 2.6 s**, in both
outputs: the gallery heading overlaps the second suggestion label, while the
final suggestion and icon encroach on the gallery's top row. This is geometric
interference, not a complaint about wording. Both outputs have separated
sections again at frame 100, PTS 10/3 = 3.333333 s. The issue is transitional
hierarchy, not a claim that the final layout remains broken. These witnesses do
not establish the exact first and last collision frames.

The reference's native frames 73–74 and 99–100 retain separated suggestion rows
above the gallery while upper content appears. Its corresponding reinspection
did not show the same heading collision. The finding preserves the frozen
criterion of staged interface hierarchy, rather than merely matching the final
arrangement.

Both outputs preserve the dispersed-image field, tightening three-column
gallery, recognizable image relationships through the device change at frame
45 = 1.5 s, and staged top content. Conservatory and coastal imagery are valid
different subjects; illustration itself is not a failure. The first two frames
were inspected as low-weight boundary context, not silently excluded.
**Confidence:** high for the native-frame collision; moderate for its material
effect during playback, which was not perceived.

## Image isolation and recombination

**No additional material gap and no default-versus-adaptation discriminator are
established.** This is a bounded absence of an established gap, not proof of
equivalence, flawlessness, or source-quality approval.

Both outputs distinguish a rectangular image, isolated silhouettes with pose
changes, an intermediate overlap, and a later contained two-person scene.
The ordered sheets show the silhouette stages at frames 15, 29, 44, and 58
(1/2, 29/30, 22/15, and 29/15 s), followed by the overlap and later related paired
results at frames 101 and 116 (101/30 and 58/15 s).

The decisive native check is **frames 86→87, PTS 43/15→29/10 =
2.866667→2.9 s**, in both outputs. At frame 86, an oversized foreground figure
extends outside a smaller room image. At frame 87, both figures fit within a
wider room composition at comparable scale, with added sofa/table context.
This satisfies the observed distinction between an intermediate overlap and a
new integrated composition; it does not merely leave the oversized sticker in
place.

The reference's closer, seated pair involves stronger relational reposing.
That difference does not establish a failure of the frozen criterion for these
original illustrated subjects. Literal sitting, touching, photographic
rendering, or the same people was not required. Clothing, character identity,
props, palette, and standing poses do not justify a forced failure or winner.
**Confidence:** moderate in the no-gap-established conclusion; the strength of
the recombination payoff during actual playback remains unassessed.

## Simulator selection

**Single biggest remaining gap: the broad-region-to-small-target inspection
stage is omitted in both the default and the adaptation.** No material
discriminator between the two outputs is established.

At frame 45, PTS 3003/1600 = 1.876875 s, both outputs still show distributed green
element outlines. At **frame 48, PTS 1001/500 = 2.002 s**, the next distinct target
state is already the small first-row icon and its metadata tooltip. The
intervening ordered thumbnails do not show a broad selected browser/phone
region or a visible large-to-small tooltip/outline sequence.

The reference instead shows a broad browser highlight at frames 45–46, a
toolbar target at frame 47, and a broad phone/feed-to-small-button boundary at
frames 58→59, PTS 29029/12000→59059/24000 = 2.419083→2.460792 s. The exact blue
hue is not mandatory; the missing relationship is spatial target hierarchy.
The judgment preserves the frozen criteria for broad-to-specific inspection
and distinct highlighting states.

Both outputs retain a sidebar, dark preview field, nested phone, and three-row
content. Both hand off from the small target and tooltip to an anchored composer
at frame 72, PTS 3003/1000 = 3.003 s, and end during unsent text entry at frame
119, PTS 119119/24000 = 4.963292 s. Neither is a freestanding phone poster or
invented completed fix. Different list content and icon designs are valid
adaptations, not wording-based failures. **Confidence:** high for the omission
across the witnesses and ordered chronology; moderate for its explanatory
effect, with moving clarity unassessed.

## Disposition

These conclusions remain unchanged after role reconciliation. They authorize
no quality approval, adoption decision, deployment, or release. Packaging and
technical review are separate from this bounded visual critique.

## Packaging boundary

Four original editable kits accompany eight verified content renders. Three retain the specific structural gaps above. Cutout recombination remains moving-review outstanding without an established additional material gap. Neither status earns quality approval. All selected retained indices were inspected at thumbnail scale with selected native-excerpt details, not all pixels at native resolution or actual moving playback. Captions limit perfect role blinding. Actual public-kit execution, delivery and native Resolve remain separate checks.