# Batch06 public candidate checkpoint

September 15, 2026. No quality approval, moving blind win or native Resolve
execution is claimed. These five kits are candidates with open visual gaps.
The source-first and anonymous label records remain sealed. Roles were reconciled
only afterward; familiarity with the source is disclosed, not treated as hidden.

## Selected exports and review scope

The selected report binds ten actual content movies: two newly rendered type
variants and eight byte-identical movies retained from the earlier verified
report. Retained movies are not ten new encodes. Each is checked against its
actual file, manifest, original report pin and native frame count. The editable
inputs remain original code/SVG assets, not flattened movie replacements.

The selected five excerpts contain 845 source frame instances and the selected
ten outputs contain 1,690 candidate frame instances. The four unchanged pieces
retain their earlier source-first static judgments. The type comparison adds
old-versus-new actual outputs without increasing distinct examples or unique
source coverage. Every retained chronology was inspected in reduced full-frame
sheets with selected native-detail follow-ups. This does not establish perceived
motion, all-pixel inspection, whole-film coverage or clean corpus passes.

## Type comparison: split result

The revised adaptation clears the old word/parenthesis overlap at local 104.
The old adaptation's partial `stonew` crosses both parentheses there; the new
one does not. This is a bounded static legibility improvement, not proof of
universal collision freedom or a moving win.

The revised default regresses on intermediate second-word positioning. At
local 104, leading inset measures 30 pixels in the source, 41 in the old default
and 87 in the new default. Across local 90–119, the ranges are 30–30, 36–41 and
47–87 pixels. These are distances from the opening parent's outer left bound
to the word's dark-pixel left bound, not optical whitespace. Different source
glyphs prevent treating 30 as a universal target. The same-content old/new
comparison remains valid. The new adaptation also retains excess spacing.

The largest next correction is to reduce excess and changing reveal inset,
especially in the default, without reintroducing adaptation overlap. The repair
has been returned for correction. A blanket successful or behaviorally isolated
adaptation repair is rejected. The original axis-specific ranks remain unchanged.

The six principal onsets tie: opening square 27, empty square pair 29, first
interior character 38, opening parenthesis 74, empty parenthesis pair 78, first
interior character 90. Empty-pair dwells are 0.150000 and 0.200000 seconds. The
outputs' connector begins at 65 versus source 66. Source-only transcription errors
are appended, not rewritten: `Go from` completes at 20 rather than 19; `de` begins
at 94 rather than 93; `depl` begins at 99 rather than 100. The source full phrase starts at 116.

All type comparison files are 1920×1080, 186 frames, 60 fps, with retained time base
1/15360 and PTS 256 times the local frame index. Local 104 occupies
[1.733333…, 1.750000) seconds. Local 90–119 spans [1.500000, 2.000000) seconds.
These are actual retained-file times, not recovered absolute source-master PTS.

## Four retained judgments

| Treatment | Single biggest remaining gap |
| --- | --- |
| Rowwise code emphasis | Token hierarchy is flattened into proportional, whole-line coloring. |
| Three-file handoff | The thumbnail slot is reserved early, losing attachment-pill reflow at 101–102. |
| Canvas range/camera return | Partial selection endpoints detach from the I-beam's text position. |
| Shopping table partial return | Early rightward travel under-advances before catching up; the larger renewed left step is authored 113 versus source 112. |

The source remains preferred in each earlier static judgment. The canvas action
chip still disappears at 125 versus source 122. No later default/adaptation replay
or fresh independent moving review is inferred for the eight retained movies.

## Documentation and clock corrections

All public entry instructions require Python 3.11+ for the unchanged helper's
`hashlib.file_digest` contract. The plugin README no longer contains another
treatment's delimiter-reveal explanation. Shopping's authored checkpoint is 113;
112 remains the source target. Raw CRLF control tables are reproduced without
normalizing their bytes. These are packaging/factual corrections, not motion fixes.

The plugin source retains its 1/16000 time base, 528/544-tick PTS steps, 533-tick
reported durations and 4.0033125-second end. Its candidates preserve exact
30000/1001 nominal CFR and 4.004-second end. Same-ordinal PTS differ by up to
13/15000 seconds. Both clocks remain explicit; neither footage nor timing was
silently normalized. Brief silent source excerpts are research references only.

Actual downloadable-kit execution and deployment checks are separate gates.
This review does not grant them in advance or convert guidance into a tested
native Resolve project.