# Supplemental proof audit — PASS2 followup

VERDICT: SUFFICIENT for the agreed web proof scope.

PR [#24051](https://github.com/pingdotgg/t3code/pull/24051), issue [#24041](https://github.com/pingdotgg/t3code/issues/24041), final head [9c7f68c](https://github.com/pingdotgg/t3code/commit/9c7f68c2a440c58c8e1c35b33aa004c8a61069b2). Reviewed repaired local artifacts against the prior `/tmp/issue-24041-audit-pass2/report.md`; [public coverage index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/60794bc67024199a/coverage-index.md). This supersedes the prior insufficient verdict only for the repaired findings and the explicitly authorized platform limitation. It is not a code approval or universal platform certification.

No tests, independent rebuild, browser, server, new capture, provider execution, source edit, push, or PR comment. Prior independent committed-fixture reproduction and 42/42 byte-integrity and pixel calculations are retained, not repeated or reattributed to a newer base. Read-only git confirms exact full HEAD `9c7f68c2a440c58c8e1c35b33aa004c8a61069b2` and clean status. Independently checked all 15 revision/path blob entries in repair source-provenance.json: all match git, including absence entries. The three affected product blobs match earlier consolidation source; fixture-script provenance remains separate. Runtime SHA attribution still depends on the disclosed capture procedure, not a signed bundle stamp.

## Findings resolved

| Row | Followup result |
|---|---|
| 1 web | Closed. Independently decoded both continuous core MP4s. Primary shows14→focused1 held through the long interval→30; final endpoint focus is recorder-qualified. RPC DOM observations bracket successful periodic sweep `023bafda777de97f` at17:51:25.473810–17:51:25.478918 with focused1 before17:50:33.774 and after17:51:29.919. Persisted17:51:29.861 remains14, old/fresh both unsettled; Enter request17:51:33.342 changes to30, DOM BODY17:51:33.479 and persisted30 at17:51:33.676. Clock-overlay take independently decodes to unfocused30, overlay UTC17:55:11.408/focus BODY. Do not use that take's nearly coincident initial focus/sweep as the sole across-sweep proof. |
| 2 | Closed. Decoded cancellation clip visibly contains45→focused30→blurred30, empty focused input→blurred30, and final45. PTS guide from decoded samples: draft45 at1.037s/2.961s; restored focused30 at4.597s/6.587s; blur30 at7.794s; empty at8.918s/10.938s; restored30 at12.780s; final45 at18.272s. Timestamped Escape/Tab observations identify the actions rather than inferring keys from pixels. Independently counted all server updateSettings spans in the four supplied UTC windows:0 Escape→Tab,0 empty→Tab,0 unchanged Enter,1 valid Enter45 (`216cb795ba1cc7f6`, Success). Outgoing request log contains one45 patch; persistence observes45. These are accepted settings mutation counts, not filesystem syscall counts or notification counts. |
| 6 | Closed. Independently decoded checkout and held-clear supplement. Original clip shows A30, B inherited45→20, A30, short clear45, then B20 (decoded samples around1.041s,15.465s/16.022s,19.791s,28.533s,32.908s). RPC patch scopes and persisted maps distinguish the two registered checkouts: environment45/A30/B20, then A:null leaves environment45/B20. Supplement holds A40 then clear→inherited30 (around21.954s and37.399s), then B20 at53.075s; its environment baseline is30, not45. Route/caption and endpoint stills supplement the fast navigation/loading frames. This proves checkout destination isolation; no claim a checkout picker was exercised. |
| 1 Electron / 9 native IME | Still unverified. Linux Chromium does not establish macOS Electron, native composition, or Enter during composition. Explicit user hard stop permits this disclosed limitation, so it is not missing required web proof. Retain it visibly in the PR body and terminal handoff. Earlier ArrowUp/Down result remains retained. |

Rows3–5 retain prior sufficient results with disclosed fault injection, ordering provenance, and same-host destination limitations. Row7 retains partial surrounding-gate coverage; it is not a new acceptance blocker. Row8 retains exact prior threshold/pixel arithmetic and responsive findings, including sidebar hover/caret/raster deltas. Row10 retains neighboring-consumer smoke results. Rows11–12 retain agreed mobile/provider/transport exclusions. No new gates or platform requirements introduced.

## Remaining actual limitations

No outstanding required web evidence gap against the prior findings. No matched before/after cancellation pair was added; the historical base core remains destructive-bug before proof at its original SHA, not fresh proof at requested current base [66ee628](https://github.com/pingdotgg/t3code/commit/66ee628e0875b2b735b199f42e6151e347833637). The repaired index acknowledges this. The primary recording's stale endpoint and encoded/wall-clock mismatch remain disclosed; separate DOM/RPC/persistence/still evidence and the fresh clock endpoint close the behavioral claim together. Neither video alone proves persistence or settlement. Unavailable macOS/native IME capture is a documented platform limitation, not an undisclosed web gap.

## FRICTION and own cleanup

No new tool blocker. Existing recorder cadence/stale-frame behavior required decoding rather than trusting report endpoint wording; no recapture requested. Local OpenCV decoded the five supplied MP4s (87/72/40/136/215 frames respectively); generated inspection contact sheets only. Sampling sheets are aids, not replacement source recordings or exact wall-clock maps. No claim of tamper-proof capture or cryptographic runtime attestation.

Exact own cleanup root: `/tmp/issue-24041-audit-followup/`. Own files: `report.md`, `final-cancellation.jpg`, `final-checkout.jpg`, `checkout-clear-supplement.jpg`, `final-core-continuous.jpg`, `final-core-clock.jpg`; upload output may also be retained there. No processes, ports, browser tabs, isolated DB, secrets, or worktree files created by this followup. Nothing deleted. Prior audit and repair roots/state are not mine to clean; their private logs, pairing data, and DB were not uploaded. Only this sanitized report is uploaded through the [ghdrop skill](/home/ubuntu/.codex/skills/ghdrop/SKILL.md).

Model/harness: gpt-6.1-sol / Codex in T3 Code.