# Final proof audit supplement — issue 11873

VERDICT: SUFFICIENT

The sole outstanding R1 AFTER dismissal → selector-reopen gap is closed. Independently downloaded the public supplementary MP4, updated evidence report and upload index; all three match local bytes. Video SHA256: d5f445d9d3d7cd8446e27816535980459d61f8ecfda2b94b2c95c51594551625. H.264, 2880×2000, duration 19.104233s.

Extracted and visually inspected actual recording frames at 0.000, 6.400, 6.600, 7.400, 10.000, 14.000 and 18.000 seconds. Report visible at 0 and 6.4, showing the two demo source accounts; dismissed by 6.6; model/provider selector visibly open at 7.4 and still open at 10, 14 and 18, listing Codex (via cliproxy) models. These are frame observations, not inferred exact click times. The supplement is separate from the original recording; it does not pretend to repair the original clip in place.

Capture SHA: [330326c7db](https://git.cafe/pingdotgg/t3code/commit/330326c7db0bbc8165d550c9c5082d13b7dee9d7). Current clean worktree SHA: [92e1f121c9](https://git.cafe/pingdotgg/t3code/commit/92e1f121c9bc12fead91a51e5a8406d83960130e). No capture at the latter SHA is claimed. Independently inspected the entire delta: only apps/server/src/provider/testFixtures/usageEvidence/peer.mjs changes. Default Node built-in imports become namespace imports with matching reference renames; the HTTP callback and all remaining changes are formatting. After explicit import/reference normalization and whitespace removal, full source equality holds. Runtime checks confirm readFileSync, appendFileSync, createServer and createInterface are identical functions through namespace versus default exports. No builtin export mutation exists in this fixture. Production, fixture data and seeder are byte-unchanged. Thus the captures remain sufficient at current head without recapture.

Inherit /tmp/issue-11873-evidence/proof-audit.md acceptance of S1, S2, S3, R2 and R3, all 69 original assets/diffs, threshold statistics, and the independent S1 AFTER rebuild with zero changed pixels at >0 and >8 (maximum channel delta 0). Those checks were not rerun. Original audit limitations remain: protocol fixtures do not prove real subscription login, and responsive web does not prove native mobile, Electron shell, remote/relay/tunnel or streamed-turn behavior. No new gap in the requested proof matrix remains.

[Verified recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/692bac30d3df4ec4/R1-after-dismiss-reopen-1440-light.mp4) · [Updated report](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f3e043f1d48ae547/evidence-report.md) · [Updated index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4824c77beaf8019c/upload-markdown.md)

FRICTION: Original R1 AFTER omitted the reopened selector; the separate supplement resolves it. Captured SHA precedes current SHA; independently verified fixture-only equivalence resolves recapture concern. No browser, server, checkout edits, pushes, comments, merges or watchers used. Only /tmp audit artifacts and authorized public audit upload.

Audit model/harness: GPT-6.1 Sol / Codex.

VERDICT: SUFFICIENT