# Proof audit pass 2

Findings first: evidence is insufficient for a complete affected-surface/runtime claim. No confirmed implementation defect was found in this narrowly scoped proof audit. The missing evidence is not proof of a defect.

Reviewed [#24914](https://git.cafe/pingdotgg/t3code/issues/24914), [b76a110aa4](https://git.cafe/pingdotgg/t3code/commit/b76a110aa4) to [8d2db6f7c2](https://git.cafe/pingdotgg/t3code/commit/8d2db6f7c2), all seven changed files and relevant render conditions. During review the implementation checkout advanced to [3a0a20061a](https://git.cafe/pingdotgg/t3code/commit/3a0a20061a), a merge of main with server/protocol changes. The independent rebuild remains pinned to the requested 8d2db6f7c2. A follow-up comparison verifies all five changed marker implementation files are byte-identical in 3a0a20061a and that its diff against b6c326d12f contains the same seven feature files. Static marker proof therefore transfers; the merged server/protocol changes themselves are outside this proof audit.

## Findings / proof limits

1. **Required mobile rendering proof is missing.** M1–M3 exercise new native symbol rendering, selected tint, caller layouts, and selection transitions. The EAS task produced no device session or images. Its reported GhosttyKit architecture mismatch is an environment/build blocker, not evidence that the indicator is wrong. M4 is also unverified, but is principally an unchanged visual control with changed shared accessibility-label calculation. [Mobile handoff](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bc1b32d341db4231/index.md).
2. **Runtime supplement materially closes T1 and core T2/T3 behavior, with remaining continuity limits.** T1 now passes. Both revisions record answering while Working; HEAD removes the marker while the timer and foreground command continue. Both record finishing with an unanswered async question switching to Input. T2 remains partial because uninterrupted A completion and full navigation-return coverage are absent. T3 remains partial because HEAD answer submission/clearing is only established by a later cleared-state segment. T4 gains a baseline blocking Input recording, but no matched HEAD control/answer sequence. See reviewed runtime addendum below.
3. **W3 and W9 are partial.** New content shares existing card status-slot gates and tooltip primitives. Recordings show some control/menu restoration and a full tooltip, but not a complete card gate cycle or compact/palette tooltip sequence. Existing gate mechanics and tooltip open/close infrastructure are unchanged; do not demand every unchanged interaction permutation as if it were new functionality. A short targeted HEAD sequence should prove the new content restores and the new tooltip text is reachable.
4. **Do not promote W7 to an independent sorting feature requirement.** The aggregate comparator and precedence are unchanged. W6 proves an affected hidden winning A displays the badge. S2 verifies resolver precedence. Reorder/equal-priority/higher-priority visual winner changes remain unrecorded, so the original matrix's W7 is partial, but exhaustive reorder recordings are supplementary regression controls, not a newly implemented path. The UI intentionally shows the winning status's question marker; it is not an “any hidden thread has a question” aggregate.
5. **Captions must remain honest.** D is visited ready, not a visible unseen Completed pill. G uses static goal metadata, not real goal creation. Static card A is selected and G unselected; there is no independent unselected A duplicate. The icon branch itself has no isActive condition, so this missing duplicate is supplementary, unlike mobile's explicit selected tint branch. Parent's real before/after are unmatched different threads: both show foreground sleep and question UI; after visibly adds the icon. Before was later cancelled by restart. They support the point-in-time reproduction, not lifecycle continuity or a matched visual delta.

## Row status against actual diff

| Row | Status | Required affected path versus control |
|---|---|---|
| W1 | Static pass, stated limits | 1440/390 light/dark, selected A Working+timer+question; N/B/C/D controls lack extra marker. No inactive A duplicate; shared marker branch inspected. |
| W2 | Static pass | Unselected G retains Goal+timer+question in all four card captures; real goal creation unproven. |
| W3 | Partial | New marker sits inside existing hide/drag/sweep slot. Some hover/menu steps seen; sweep/drag/snooze cycle and full restoration sequence missing. |
| W4 | Unverified control/accessibility | Slim has no newly added visual icon. Shared accessible-name change applies only when working+question is reachable; do not promise visual parity. |
| W5 | Static pass | Legacy full affected caller, all four appearance/width pairs. Narrow uses full component with text hidden by existing breakpoint. |
| W6 | Static pass | Actual hidden-row Show more compact winner, all four pairs; added badge preserves slot width. |
| W7 | Partial, supplementary winner controls | Show more/less seen; visual tie/reorder/higher-priority changes absent. Aggregation unchanged; W6 plus S2 cover new propagation and resolver branches. |
| W8 | Static pass | Palette full-pill caller, all four pairs, top-level palette. |
| W9 | Partial | Accessibility text verified for full/compact; full tooltip visible in head recording. Compact/palette hover/focus cycles incomplete; primitives unchanged. |
| M1 | Missing | New native card symbol and layout, both appearances, screen caller. |
| M2 | Missing | Native selected/unselected tint and navigation/primary-pane callers. |
| M3 | Missing | Actual selection transition and row reuse. |
| M4 | Unverified control/accessibility | Slim/Woke/fallback visual branches unchanged; shared label changed. |
| T1 | Pass, live base/head | N→question while Working; advancing timers; HEAD marker present, baseline absent. Same driver/environment, separate fresh threads. |
| T2 | Partial, core transition proven | Both answer while Working. HEAD marker clears while command and timer continue. A completion transition omitted; later HEAD settled/navigation state only; baseline navigation-return missing. |
| T3 | Partial, core transition proven | Both finish unanswered → Input. Baseline answer/clear/resume recorded; HEAD answer-cleared state only in later navigation clip, not uninterrupted submission. |
| T4 | Partial data plus baseline live control | Baseline Plan-mode blocking question → Input recorded. HEAD matched control, control answer and live legacy omission absent; legacy/data gates pass. |
| S1 | Pass | Independently reran public raw-shell driver at pinned head: working gates, request kind/blocking, legacy omission, null. |
| S2 | Pass for affected branch | Independently reran missing/false flag, approval/child input/guest/blocked precedence, waiting and connection statuses. Unchanged Plan Ready/Completed branches inspected; not claiming every unchanged state got a new runtime exercise. |

## Artifact inspection and provenance

Inspected all 32 final whole-viewport PNGs, all 32 threshold0/16 diff PNGs, all 12 superseded diagnostic captures, remaining width/rebuild/review PNGs, and both parent real reproduction images. Recordings were decoded and inspected in one-second frame sequences across all five complete durations (6.29s, 13.29s, 44.27s, 11.88s, 16.07s); this is sampled visual inspection, not a claim to inspect every encoded frame. Shorter missed events cannot be certified by this sampling. Invalid-live-attempt shows a submitted prompt into synthetic A and does not prove a lifecycle.

All 16 supplied final pair pixel counts were independently recomputed from the local full PNGs and matched diff-stats.json at thresholds 0 and 16, without masks/crops. Visible final differences match the changed code: cards gain the indigo icon and shift/truncate neighboring status/project text; legacy/palette add an overlaid badge. No final clipping/title collision observed. Palette threshold0 backdrop differences are real and must not be described as exact identity outside the dots. Earlier diagnostic narrow layout width displacement is visible; final narrow captures restore the former width.

Capture SHAs are declared by the index rather than visibly watermarked in screenshots. Verified c7481a95e4→8d2db6f7c2 changes only ThreadStatusIndicators.tsx, so card sources are identical in final. Final legacy/palette revision is declared in index; independent rebuild below anchors one affected final path. Environment metadata agrees with screenshot sizes: CSS 1440×900 is saved downscaled 1280×800; CSS390×844 saved780×1688 DPR2. Frozen clock/reduced-motion make these static evidence. Public/local media byte validation was attempted but stopped on HTTP403; no blanket public/local byte-identity claim is made.

## Independent public-driver rebuild

Downloaded public prepare.py and browser-driver.md. Created a sibling detached checkout pinned to 8d2db6f7c2, installed dependencies, ran its own dev server with --no-seed to create an empty isolated schema, stopped it, executed the public seed, and restarted. No live database was opened. Only public fixture patches modified disposable ProjectionStore.ts and index.html; no PR source was edited.

Rebuilt legacy-full / CSS390×844 / light, selected proof-main-0-A, main expanded and overflow collapsed, sidebar width localStorage208, frozen2026-10-25T12:01:30Z, reduced motion true. Linux Chrome154, DPR2. Measured sheet378.5 CSSpx and correct Working · Question waiting accessibility names. Against supplied head-legacy-full-390-light.png: **385 pixels differ at threshold0, bounds[757,1396,758,1635], maximum channel delta3; zero pixels differ above16/255**. Differences are confined to the sheet's right border, not status content. This is close reproduction within stated tolerance, not pixel identity.

Local deliverables for embedding/review: audit-rebuilt-head-legacy-full-390-light.png, audit-rebuild-diff0.png, audit-rebuild-stats.json, audit-data-results.json in this evidence directory. Existing base/head assets and public URLs remain available through index.md. Own dev sessions stopped via retained handles; preview tab closed; sibling worktree removed. No forge edits, merge, watch, or implementation checkout edits.

## Minimum additional proof

- Finish the remaining runtime tails: A completion and settled navigation-return (baseline navigation-return remains absent); uninterrupted HEAD Input→answer→cleared state for B; matched HEAD blocking control and answer. T1 and the already recorded answer-while-working / finish-unanswered transitions need no repeat. Legacy omission can remain data-only. Sequential same-recipe recordings must not be described as pixel-matched.
- One short HEAD interaction recording can pack card controls enter/exit and marker restoration plus full/compact/palette tooltip text. Sweep/drag should be exercised only where legitimately reachable; report unreachable combinations rather than fabricating state. W4 can be a slim control/name inspection in the same pass. Reordering exhaustive W7 ties is optional unchanged-control coverage.
- Native EAS proof after resolving build environment: light/dark screen and navigation callers, selected/unselected active async rows, selection transition, one slim/Woke control. Label store injection HARNESS ONLY if the prepared fallback is used; it does not establish backend/runtime behavior.
- Source equivalence of marker paths on the final merge head was verified. Review runtime supplements against their actual revision/environment.

## Runtime addendum — bounded read-only inspection

Reviewed the [runtime index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8492124ff1eb2c51/index.md) and [timestamped public driver/steps](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5f28bf55250bdaef/steps.md). Downloaded and decoded all seven recordings; valid encoded sizes are 2880×1800 and durations are 30.16, 8.71, 44.63, 26.12, 25.82, 18.62 and 65.59 seconds. Inspected complete durations as one-second sampled frame sequences, including start/end and transition frames; this does not certify every encoded frame or subsecond continuity. No new browser, server or checkout was created for this addendum.

The declared environment is dark CSS1440×900, same isolated origin/state/checkout, real Codex GPT-6.1-Sol Low, identical prompts, baseline b76a110aa4 and HEAD 8d2db6f7c2. Visible dimensions, appearance and provider agree. SHAs and no-injection provenance come from the supplied driver/index; they are not visibly watermarked. Sequential fresh threads and accumulating sidebar controls mean these are comparable same-recipe runtime runs, not pixel-matched pairs. The earlier verified marker-source equivalence to merged HEAD remains applicable; no claim that this footage executed the merged server revision.

| Recording | Independently visible credit and limitation |
|---|---|
| [Baseline A](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3702c1cbc63af078/baseAPath.mp4) and [tail](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a749d31684596675/baseATail.mp4) | Question appears around 7–8s with Working and no marker; timers advance. Around 24–25s Blue enters history, question clears, Working and Running sleep persist. Separate tail still shows Running sleep; approximately six wall-clock seconds between files are omitted. Final completion is not recorded. |
| [HEAD A](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5c479261cdc6d603/headAPath.mp4) | Around 9s question appears while Working; 12–16s marker and increasing timer coexist. Around 17–19s Blue is answered, marker/question disappear while Working and Running sleep continue. At 44s still Working; completion transition absent. T1 passes and T2's changed behavior is proven. |
| [Baseline B](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/506b2e6aa37a3f9f/baseBPath.mp4) | Async question persists while Working. Around 21–22s Input replaces Working with Live B complete and unanswered question. Around 23–25s answer clears question and Working resumes. |
| [HEAD B](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a77eb75acc2854bd/headBPath.mp4) | Working+marker advances through approximately 20s; 21–23s switches to Input after Live B complete with question still present. Blue selection near the end does not itself prove submission. T3's finish-unanswered transition is proven. |
| [Later HEAD navigation](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7282981ba954d441/head-completion-navigation.mp4) | Opens already settled B, then A and B through app-loading splashes. At 14–18s answered B has Blue history, completion, normal composer and no stale marker. This proves a later persisted cleared state, not uninterrupted answer submission or A completion. |
| [Baseline blocking control](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/20fb5a0d451668cc/baseCPath.mp4) | Plan-mode real blocking question and Input pill appear around 8s and persist to the end. No matched HEAD blocking control or answer recorded. T4 remains partial. |

No runtime contradiction to the changed marker behavior was found. M1–M4 remain entirely unverified on a native device. W3/W9 remain partial for the specific control-restoration/tooltip gaps above. W4 remains an unverified slim/accessibility control and W7 partial supplementary winner controls; neither is a newly missing visual marker feature. These distinctions should be preserved in the PR disclosure. The verdict lists all incomplete matrix rows, not a claim that each is an independent merge blocker.

Scratch downloads, decoded frames and contact sheets were removed after inspection. Combined audit and the earlier requested independent-rebuild deliverables remain in the evidence directory; public media URLs are preserved here.

## FRICTION

Fresh isolated dependency install and dev startup cost time; initial browser route needed token entry; sidebar button is named Toggle main sidebar; narrow sheet defaults depend on persisted width. Public media byte check returned HTTP403. Independent rebuild retains small ≤3-channel sheet-border variation. Mobile native build has no simulator output. Runtime supplement subsequently reviewed below; encoder tails omit A completion and HEAD B answer submission, and navigation includes loading splash. These gaps remain explicit. Parent reports current-head check:changed passed 96 tests and seven typechecks; this was not rerun by this proof auditor. Native unverified-surface disclosure permits the parent’s instructed PR handoff, but does not make M1–M4 proven. Temporary review sheets/extracted frames are removed; requested deliverables retained separately.

VERDICT: INSUFFICIENT W3,W4,W7,W9,M1,M2,M3,M4,T2,T3,T4