Adapted from versioned Process native prepare/verify tools. Explicit completed runs only; raw bytes and historical helpers are archived without extraction or execution. Collector refuses existing output and checks raw run bytes again after collection. Run-1 remains aborted preflight, run-2 remains failed baseline verification, all other runs remain unreviewed. Archive consistency does not decide native acceptance.

python -B prepare.py --root CHECKOUT --runs run-1 run-2 run-3 run-4 run-5 run-6 --check-only

After review, replace --check-only with --output FRESH_DIRECTORY. Verify with python -B PACKAGE/verify.py. Do not include active runs. Exact raw members, source capsules, fixture/snapshot maps, build receipt/artifact relationships, logs, captures, helper versions and Allocation history/expected-path/sentinel checks are independently read back. Historical source capsules are included once for self-contained provenance; the seven source-commit proofs remain in the sibling gate package and should be reviewed with it. No new Git comparison or binary execution occurs.

Missing capture-executed.js is recorded as null, not reconstructed from current capture.js. Prepared capture.js bytes remain pinned; this cannot establish exact executed UI code or PID-window binding. Images/accessibility/lifetime metadata cannot prove physical gesture sequence, one-step history, held input, focus/camera, or usability. Parent observations and independent review establish those bounded claims. Prefix sentinels establish exact leading fixture comments only. Existing 18 model warnings remain advisory fixture incompleteness; inspect the acceptance report. The archive records executable hashes and Cargo artifact receipts, not executable binaries themselves.