Candidate construction guide · native Resolve not tested

Keep the wider view: SAME-ID R04 public kit

The existing R04 default and adaptation inputs are unchanged. The completed actual pair is bound in PUBLIC-KIT-PROVENANCE.json; movies and source footage are deliberately absent. PUBLIC-ASSESSMENTS.md distinguishes the exact public static assessments from historical repair material.

Reconstruct and test offline

Use Python 3.11+ and an existing Node 22.18+ with type stripping. Unzip into a new directory. No install, browser, encoder, network or global toolkit is needed for these checks. The included core archive is local to this kit; reconstruction does not restore any deleted global project or agent guide.

python3 -B reconstruct.py --destination work/reconstructed
python3 -B test_kit.py --project work/reconstructed/project
python3 -B test_pose.py -v
python3 -B test_repair.py --project work/reconstructed/project -v
python3 -B test_guide.py -v
python3 -B test_public_kit.py -v
python3 -B handoff.py verify --guide guide.json --project work/reconstructed/project
python3 -B resolve_controls.py verify --guide guide.json --project work/reconstructed/project
node --experimental-strip-types --test work/reconstructed/project/tests/timing.test.mjs work/reconstructed/project/tests/assembly.test.mjs work/reconstructed/project/tests/text-fit.test.mjs

The original tests cover 27 unit cases, archive safety, core/overlay bindings, all-frame controls, original artwork and authored geometry. The public tests check provenance, exact filenames, scope, manifests and preservation. Read the delivery's VERIFICATION.md for the executed command results. No dependencies are included; hashes of package metadata do not attest installed global bytes.

What stays identical

The recipe, both content files, both CSVs, source clock, core archive and seven-member active overlay retain their R04 bytes. Four original SVG assets include the two scenes and matching detail wrappers. Their historical -detail-r03.svg names remain intentional. Use their hashes, not suffixes, to identify the R04 bytes. Both CSVs retain CRLF records, including the quoted embedded LF in resolve-controls.csv.

The canvas is white, silent 1920×1080. Local frames [0,59) run at exact 24000/1001 fps. Additive onsets remain 0/11/19; states last 11/8/40 frames. The end is 59059/24000 seconds. Nothing here retimes, redraws or animates it.

Historical status is not current output status

repair-contract.json, input-preservation.json, inspection-limits.json, historical/R04-BUILD-README.md and criticism/PUBLIC_*.md are preserved generation-time evidence. Their unrendered/pending wording described the original builder before the parent rendered R04. Existing helpers report renderVerified: false because those helpers check input controls, not movies. Do not rewrite their tests to turn that into a render or Resolve approval.

Current actual-file evidence is in PUBLIC-KIT-PROVENANCE.json and actual-proofs/. The public manifests remove only the recorded browser's local executable path; both raw-manifest hashes and derived-manifest hashes are named. Source hashes in inherited recipes describe full-film provenance, not the separately supplied source derivative. No source master was opened here.

Parent assembly

Keep the exact ZIP after independent review when possible. A generic assembler may recreate START-HERE.md, bindings.json or render-verification.json, but must preserve the exact bytes at the exact paths in ASSEMBLY-PRESERVATION.json. An alias cannot satisfy a hash-bound filename. If anything changes, regenerate the complete kit-files.json inventory and rerun the extracted ZIP's tests. The current verification does not certify that future artifact.