Candidate construction guide · native Resolve not tested
Historical builder snapshot: 2026-09-15. Earlier README and handoff statements about being unrendered or unpublished describe that authoring stage. The current selected report verifies both content exports at 2026-09-15T04:23:05.909779+00:00; those earlier statements do not describe their current availability. Native Resolve execution and source-quality approval remain unverified.
sora-recut-03 R02: preview-only builder correction
Two original contents, Coast Survey and Letterpress, retain all seven original SVGs and all 67 layer IDs, kinds, order, text and palette. Only preview-wide and preview-detail image x/y/width/height and clip insets change. Preview world bounds stay [748,240,1452,636), and initial arrangements at 0 and 150 are exact. The 180-frame/60-fps/1920×1080 clock, cut 150, cursor, all readouts, endpoint pulse 59/restore60, clip bounds, thumbnails and every unrelated control stay unchanged.
This adds zero examples or coverage: same ID, revision r02. The implementation uses authored affine overscan, not copied source frames, source-camera recovery, nonrigid subject simulation or product-result choreography. See SOURCE-FIRST-R02.md. This candidate has no R02 render, moving-quality pass or native Resolve credit.
Actual portable ZIP handoff
Extract download-kit.zip into an empty folder. It contains the exact retained core-publication-v08.zip (SHA-256 307fb6b927e35e6d73779652873d994f5c3e36cb7deba65f22fa64748057a0de). In this new R02 guide, active-assets.zip is the R02 archive selected by overlay-manifest.json. active-assets-r02.zip is a byte-identical named copy. The standalone kit also retains original-active-assets.zip with R01's unchanged bytes/hash for audit, not assembly. That baseline archive is not required in parent-repackaged kits. The original frozen guide and its active-assets.zip remain untouched. Apply only the R02 archive after the unchanged core.
From the extracted kit directory, with Python 3 and Node >=22.18 available:
python3 -B -m zipfile -e core-publication-v08.zip .
python3 -B -m zipfile -e active-assets.zip .
python3 -B handoff.py verify --guide guide.json --project project
python3 -B test_kit.py --project project
python3 -B test_repair.py --project project
python3 -B handoff.py frames --guide guide.json --project project > controls-regenerated.csv
python3 -B resolve_controls.py --project project > fusion-controls-regenerated.csv
cmp controls.csv controls-regenerated.csv
cmp fusion-controls.csv fusion-controls-regenerated.csv
No install or dependency tree is required for these checks. The baseline 18-test suite retains its mechanism checks; test 17 now checks this builder's honest non-blind review order and 12 own source-sheet witnesses instead of claiming R01's source-first order and 15 witnesses. The added eight-test regression uses the actual extracted core validator/sampler and projects visible SVG anchors, including both shots and readout hold intervals. This is code/geometry evidence, not rendered proof. baseline-recipe.json and baseline-svg-hashes.json bind R01. The public packager may flatten inputs/defaults.json and inputs/adaptation.json to defaults.json and adaptation.json without changing their contents.
Parent rendering and review
Parent must assemble from the pinned R02 archive and retained core, using the
existing authorized runtime, and render default/adaptation separately. The core
entry remains npm run render -- --recipe sora-recut-03 --variant default --out out/default.mp4
and the corresponding --variant adaptation --out out/adaptation.mp4 command.
These are parent instructions, not commands run by this builder. Do not install,
retime, add holds/audio or reuse the older movies as R02 evidence. Exactly 180
frames at 60 fps, 1920×1080; parent confirms actual encoded format and no audio.
The separate unchanged source reference remains the comparison authority.
Fresh critic must inspect actual R02 outputs at 104/130/149 and 150/159/169/179, plus full playback, cut149/150, pulse59/restore60 and unchanged outside-preview behavior. No local test closes the prior C5 gap or changes sealed findings. The remaining limitation is affine original artwork versus nonrigid source motion/material detail; the secondary cursor/handle regression is not repaired.
Resolve and raw controls
RESOLVE-HANDOFF.md section 7 gives exact preview construction and clipping; earlier sections retain all other controls. resolve-recipe.json contains the actual graph and unchanged SVG primitive lists. controls.csv has local numeric tracks; fusion-controls.csv has evaluated world geometry for both contents. Both are raw CRLF CSV; Fusion text values also contain quoted LF. Do not normalize newlines or split records on bare LF. No .drp/.setting/importer exists.
overlay-manifest.json binds active members. kit-files.json binds download members except itself and kit-files.sha256. This local staged ZIP is not published and no download URL or served-delivery verification is claimed.