# Final proof audit pass 2

**Findings: no remaining missing affected row. VERDICT: SUFFICIENT.** This evaluates the original requirement to list every visible difference; it does not require proving a rasterizer mechanism. The earlier audit imposed that extra causal requirement. It is withdrawn. Production review verdicts are unchanged.

**Differences outside the PR title are disclosed here:** the light desktop sidebar boundary at image x226 has RGB/alpha differences (including reload30's 255 alpha pixels above16); matched narrow focus/selection retains 28 light /29 dark ring-corner pixels at x341..354/y1145..1198, three above16 each. Their exact causes are unknown. They are concretely located and counted, not waived as AA. Do not say “nothing else changed.” No source changes are proposed by this audit.

Issue [#14743](https://github.com/pingdotgg/t3code/issues/14743), reused PR [#22208](https://github.com/pingdotgg/t3code/pull/22208). Read-only checkout at [0642083522](https://github.com/pingdotgg/t3code/commit/0642083522df9989de2165a8707012376a1e5864), baseline [22840ab943](https://github.com/pingdotgg/t3code/commit/22840ab943b8418744b48223411e46ba4fdcef1a). Independently compared StorageSettings.tsx, settings/useScopedSettings.ts and StorageSettings.browser.test.tsx bytes against both reviewed production commits: identical. This equivalence is scoped to those three paths, not the whole repository.

Read corrected matrix, prior audit, new report, committed fixtures, selected manifest/journals and screenshots. IMPORTANT: final evidence is under /tmp/issue-14743-fresh-proof/selected, matching the published report. The older deliver directory is not the final selected set.

## Independent checks

- Decoded every frame of all four selected dark raw recordings: after desktop117/narrow105; before desktop124/narrow105, 451 total. All background probes are dark (maximum channel12). Inspected all four final frames: loaded Custom Storage; baseline saved8/On, head Off. The probe alone is not the semantic oracle; the final images establish loaded return. All72 checkpoint persisted preferences/classes and per-document route mutation logs agree with intended theme.
- Checked40 selected authoritative server readbacks across Off, enable, committed30, reload30 and loaded-return. Baseline enable persists8; head enable remains null with zero writes; successful30 ack and readback precede reload30. Environment I/S stay null. Loaded return is baseline8 versus head null. Existing accepted S0 and checkout-disposal evidence is retained.
- Recomputed all24 whole-surface RGBA diff0/diff16 pairs and independently bounded changed coordinates: zero changed pixels outside the report's inactive-control region, narrow lower rows, and light desktop x226 seam. Narrow lower rows match exactly under the declared20-image-pixel diagnostic translation. Counts remain unaligned. Thus control appearance/focus, conditional row-height displacement/bottom cut, ring corners and seam account for all selected differences. No text-glyph region remains omitted. Earlier invalid/unequal-seed comparisons remain nonproof, not repaired by relabeling.

## Row status

| Row | Status |
|---|---|
| R1 | Prior independently verified four-caller draft/write/ack/Off proof retained. |
| R2 | Prior valid commit/edit/persistence proof retained; baseline draft remains unreachable. |
| R3 | Prior valid S0/I1/I3650 proof retained; invalid baseline concatenations remain excluded. |
| R4 | Prior independently verified cancellation, Tab/Space and internal-focus repair retained. |
| R5 | Prior independently verified real blank blur/recovery and labeled library null callback retained. |
| R6 | Verified missing final themed project chapters/readbacks/full-surface pairs now supplied. Prior S0, project persistence and checkout disposal retained. |
| R7 | Prior verified actual scoped runner failure/reset/retry harness retained. |
| R8 | Prior verified request-aware duplicate/old-failure/new-pending harness retained. |
| R9 | Prior ordered callback reconciliation evidence retained; no arbitrary response-order guarantee. |
| R10 | Prior verified unavailable/partial outcome aggregation retained. |
| R11 | Prior verified targeted synthetic pointer/composition guards retained; no physical-device claim. |

## One fresh committed-fixture replay

Ran seed-project.mjs, project-runtime.js and project-capture.js from committed head in a fresh isolated /tmp/issue-14743-pass2-home, using the same host/browser environment, CSS390x844/DPR2, dark persisted theme, same fixture data and matching navigation/settling. Registered disposable project through app RPC. Replayed one affected project-I chapter (nine checkpoints), not the entire matrix. Independently confirmed enable zero writes/null, saved30/readback, reload30 and return null/Off.

Against selected after-390-dark originals, full unmodified780x1688 RGBA PNGs: Off 0/0, enable 74/74, selected-focus 0/0, committed30 0/0, reload30 0/0, loaded-return 0/0 (diff0/diff16). Enable's only changed coordinates are x187..188/y1151..1187. Inspection shows the insertion caret visible in replay and absent in delivered still. This is a listed capture-state difference, not pixel identity. No masks, alignment or crops entered counts.

![Delivered left, independent replay middle, amplified difference right: only insertion caret differs.](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a7c4a4e07cfaf7fa/replay-crop.png)

![Final decoded dark frames: after desktop, after narrow, before desktop, before narrow.](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7d6fd7b29d6e3da5/dark-finals.png)

Smallest missing proof: **none under the stated requirement**. Keep the known regions and unknown causes explicit. A claim that every pair is pixel-identical or nothing else changed would be false; neither is necessary for sufficiency.

## FRICTION

- Used the test-t3-app skill and authorized T3 collaborative preview only. Initial preview was closed; opening incognito succeeded.
- Initial replay timed out waiting for retention rows because project registration cleared the seeded Custom override and rendered Inherit. Restored Custom/null through server.updateSettings before the successful committed-script replay. No source/fixture edits.
- The initially suggested deliver directory contains older captures; discovered and used selected, the report's final set. No old recording was silently substituted.
- Shell cleanup with rm -rf was automatically rejected; removed only the exact owned isolated home through a path-checked Python cleanup instead.
- No broad tests, pushes, forge writes, watchers or uploads. Required app-owned PR link registered and verified watching=false. Stopped only owned server session16720 and closed only owned preview tab; removed own isolated home. External audit artifacts retained. Worktree remains clean at original head.

VERDICT: SUFFICIENT