# Batch07 source/output review and public corrections

All five source references are preferred over their default and adaptation
candidates. The two candidate variants tie within each treatment. This is
static native-frame inspection with disclosed source familiarity, not a moving
blind win, a source-quality approval or native Resolve execution.

The frozen review covers 973 selected source frames and 1,946 candidate frames.
Another 155 source-context frames were inspected around the carousel selection.
Those context frames do not increase the selected-frame or distinct-example
count. Nominal source cadence, actual retained-file timestamps and candidate
timestamps are kept separate.

## Single largest remaining gap per treatment

| Treatment | Gap to correct |
| --- | --- |
| Layer insert, resize and place | Both candidates resize too slowly early in the gesture, local frames [102,110). |
| Topic stack and release | Both candidates omit independent presenter/body/face/background changes below the text, especially [29,63) and [69,76). |
| Resolving carousel | Both candidates introduce a hard in-card replacement front instead of intact-image resolving, [84,105). |
| Chart representation change | Both candidates reduce pointer scale and outline/shadow separation, [0,51) and [146,240). |
| Chart-to-chart pan | Both candidates travel too slowly early and too crisply, lacking front-loaded travel and directional smear, [11,34) and [137,178). |

## Documentation corrections, not motion repairs

The carousel guide previously described the authored vertical replacement mask
as observed source behavior. That claim is corrected. The adjacent fourth-card
replacement analogy is withdrawn because the source review did not establish it.
The editable candidate keeps its mask instructions, explicitly labeled as an
authored departure. Its third/fourth entry gates also remain later than the
source observations. No recipe, SVG or rendered motion changes for this correction.

All five handoffs require Python 3.11+. Public control tables are regenerated
from raw helper output and checked byte-for-byte. The two editorial preassembly
tables used LF while the helper emitted CRLF; their parsed controls agreed.
This serialization correction does not change motion or erase the earlier failed
byte checks. Portable archive checks are a separate delivery gate.

## Timing and execution limits

The layer source uses a retained 1/16000 clock with 656/672-tick PTS increments;
its reported frame durations and nominal 24000/1001 cadence are distinct fields.
The carousel's 145-frame public excerpt is a brief silent re-encode, selected
from 300 context frames and rebased by 1.55 seconds. Rational selected timing is
retained; pixel identity to the source is not claimed. The chart menu fade is
authored; its opacity timing is not recovered source timing. The pan's source
opening caret is not evidence of typing or newly generated data.

Resolve instructions describe manual construction, not a tested project or
render. Automated playback and static frames cannot establish perceived moving
quality. All five gaps remain open until corrected and independently retested.