# Proof audit pass 2 — issue 24999

VERDICT: SUFFICIENT A, B, C, D — all 16 viewport/theme matrix entries.

Read-only audit of `/home/ubuntu/wt/issue-24999`, branch `fix/issue-24999`, head [72735278d0](https://git.cafe/pingdotgg/t3code/commit/72735278d0f638ac2220b0754bdf3748eedefee6), base [2ed92618e6](https://git.cafe/pingdotgg/t3code/commit/2ed92618e60f5ee6ecc694eb834c28105a502d09). Read actual full diff, component, heading measurement, accessible-name component, caller, capture/record drivers, fixture recipe, trace, observations, matrix and evidence index. No source edits/restores, forge reads, pushes, comments, merges, watches, server starts or stops.

## Source bug proof

The entire diff is one component: remove the grouped display-name alias and use `activeProjectTitle` for trigger text/accessibility, trigger tooltip, and hidden heading measurement. The caller passes `activeProject?.title ?? null`. Group identity, picker item labels/search/tooltips, target selection, and the same-group no-op remain unchanged. This fixes the selected-checkout/group-label mismatch at its three rendering uses without renaming the repository group. Web and desktop share this component; responsive 390px proof does not certify native mobile. No provider, protocol, backend, or connection-specific branch changed.

The baseline capture method is valid for this diff: substituting the base version of the only changed file into the head worktree is equivalent to the relevant base rendering. The trace attributes before/after to the requested SHAs. I did not repeat that source substitution because this audit was explicitly read-only. HEAD and clean status were independently verified, including after the rebuild.

Scratch, unresolved-title and no-picker guards were inspected in source; no extra runtime coverage of those unaffected branches is claimed. The null-with-group and empty-title defensive differences noted in pass 1 remain bounded and do not undermine the supported selected-title case.

## Matrix status

Each line covers before and after at the requested SHAs. A/C/D include both repeats and both delivered masks; B includes both recordings.

| Row | CSS width | Theme | Status | Observed proof |
| --- | ---: | --- | --- | --- |
| A | 1440 | light | PASS | Group label becomes selected title; inline, unclipped; breadcrumb unchanged; hover in B. Independently rebuilt exactly. |
| A | 1440 | dark | PASS | Same title correction and inline layout; hover in B. |
| A | 390 | light | PASS | Selected title replaces group label; stacked, readable; hover in B. |
| A | 390 | dark | PASS | Same title correction and stacked layout; hover in B. |
| B | 1440 | light | PASS | One group search result; group label/tooltip retained; same-group checkout retained; concrete first/second routes update head hero/tooltip. |
| B | 1440 | dark | PASS | Same grouping, search, selection and member-title observations. |
| B | 390 | light | PASS | Same observations in responsive web layout. |
| B | 390 | dark | PASS | Same observations in responsive web layout. |
| C | 1440 | light | PASS | Baseline inline becomes stacked for long selected title; full title and tooltip; no clipping. |
| C | 1440 | dark | PASS | Same measurement-sensitive stacked result and full tooltip. |
| C | 390 | light | PASS | Stacked heading; intentional trigger ellipsis; full two-line tooltip within viewport. |
| C | 390 | dark | PASS | Same intentional ellipsis and full tooltip. |
| D | 1440 | light | PASS | `server` inline, exact zero before/after pixel delta. |
| D | 1440 | dark | PASS | `server` inline, exact zero delta. |
| D | 390 | light | PASS | `server` stacked, exact zero delta. |
| D | 390 | dark | PASS | `server` stacked, exact zero delta. |

All three fixture checkouts independently have the same repository origin. The actual group label is `SpyC0der77/arc-ui-skeleton`, which is consistently preserved in picker rows and their tooltips. The real app-store draft seed recipe is suitable for this component proof; it creates concrete drafts rather than replacing rendered text. B navigates real draft URLs, not sidebar member controls, and the index correctly limits that claim. Clicking the current group preserves the second checkout; in some recordings the menu is visible immediately before navigation, so do not narrate a sustained closed-menu state there.

## Artifact inspection and independent pixel checks

Opened and visually inspected every delivered A/C/D capture and Diff0/DiffT mask, including repeats, using full-frame contact sheets and readable hero crops. Inspected all four full-title tooltip pairs, including readable tooltip crops. Opened all eight MP4s through freshly extracted frames at 2 fps (317 sampled frames total), including the start, picker sequence, route transitions, concrete first-member state and return state. Audit sheets/frames are under `/tmp/issue-24999/audit2/`.

Independent recomputation from original PNGs exactly matches every published count and bounding box. All 24 delivered masks also match freshly computed masks pixel-for-pixel: Diff0 is max RGB delta >0 in magenta; DiffT is max RGB delta >16 in orange. All 24 before/after states have pixel-identical repeats; all four tooltip states have pixel-identical repeats. D has zero differences in all four cases. Every A/C changed pixel is confined to the hero. No unrelated composer, breadcrumb, panel or background difference was found.

| Pair | Diff0 | DiffT >16 | Bounding box |
| --- | ---: | ---: | --- |
| A-1440-dark | 8706 | 7925 | 314,295–966,322 |
| A-1440-light | 8705 | 7839 | 314,295–966,322 |
| A-390-dark | 6385 | 5633 | 173,632–606,668 |
| A-390-light | 6385 | 5570 | 173,632–606,668 |
| C-1440-dark | 11024 | 9773 | 314,267–966,322 |
| C-1440-light | 11032 | 9612 | 314,267–966,322 |
| C-390-dark | 9215 | 8573 | 24,632–748,668 |
| C-390-light | 9215 | 8491 | 24,632–748,668 |
| D, all four | 0 | 0 | none |

No settled still is ghosted, duplicated, skeleton-loaded, in the wrong theme, or contaminated by a stale overlay. C's narrow label is deliberately ellipsized and its full tooltip is intact. Tooltips above the stacked trigger can cover the question while open; that is an intentional hover overlay visible in baseline and head videos, not ghosted screenshot text. Transition frames during tooltip appearance and full reloads are not settled still proof. No resize stability, absence of oscillation over time, or loading-performance conclusion is justified by these stills, and the index explicitly avoids those claims. The measurement trace (734px base group phrase versus 1639.25px head long phrase in a 736px desktop lane) agrees with the visible inline-to-stacked result and source.

## Independent rebuilt head capture

Used the required `test-t3-app` skill, native T3 preview only, `preview_status` then `preview_open({reuseExistingTab:false})`. Own tab: `tab_3_b067f792-36e4-4f55-9ebd-aabb1db86f63`. No lead/capturer tab was operated. Reused existing port 6197. Paired against the isolated worktree `.t3`; seeded my own A draft through the driver's real `readProjects`/`setProjectDraftThreadId` recipe. Set 1440×900/light, waited for the real trigger and fonts, normalized both panels, and used the recipe's 900ms settle. No standalone browser driver was executed.

Measured Chrome 154.0.0.0 / Linux x86_64, DPR 2, root font 16px, CSS viewport 1440×900, light background `oklch(0.992 0 0)`, lane 736px, selected heading measurement 628.5px, inline true, sidebar absent. Breadcrumb and hero both identify `arc-skeleton-for-zen`. Preview's unaltered PNG is 1280×800, confirming the disclosed CSS/PNG distinction in the same rendering environment.

Rebuilt PNG `https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/409606b9e068af16/A-1440-light-rebuilt.png` versus delivered `/tmp/issue-24999/captures/A-1440-light-after-1.png`: **Diff0 0; DiffT 0; no bounding box**. Result saved in `audit2/rebuild-delta.json`; independently recomputed matrix saved in `audit2/recomputed.json`.

![Independently rebuilt head A at 1440 CSS pixels, light theme](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/409606b9e068af16/A-1440-light-rebuilt.png)

## Recording narration timestamps

Approximate seconds from each recording's start, derived from visual inspection at 0.5s intervals; allow roughly ±0.5s. These are annotation ranges, not measured latency. Use separate BEFORE/AFTER ranges because reload duration differs. “Navigate first” and “return second” include the reload; settled ranges identify the actual title proof. Group selection is brief and sometimes immediately followed by navigation; retained concrete draft identity is additionally recorded in the raw observations.

Common narration: hover the selected checkout trigger; open picker and search `arc-ui-skeleton` (one repository group); clear search and hover its unchanged group tooltip; select that same group (second checkout remains selected); navigate to the first concrete checkout draft; return to the second. BEFORE keeps the repository label in the hero/tooltip for both members; AFTER follows the concrete member title. Do not describe reload splash, temporary no-project state or reopened panels as part of the fixed behavior.

| Recording | Hover | Open/search | Clear/group tooltip | Same-group selection | Navigate first (settled first) | Return second (settled second) |
| --- | --- | --- | --- | --- | --- | --- |
| 1440 light BEFORE | 0–3.2 | 3.3–4.4 | 4.5–6.2 | ~6.2 | 6.4–14.7 (12–14.7) | 14.8–23.9 (20.5–23.9; normalized ~22.5) |
| 1440 light AFTER | 0–2.7 | 2.8–3.8 | 3.9–5.6 | ~5.7 | 5.9–14.6 (11.5–14.6; normalized ~13.5) | 14.7–21.4 (18.5–21.4; normalized ~19.5) |
| 1440 dark BEFORE | 0–2.7 | 2.8–3.7 | 3.8–5.1 | ~5.2–5.7 | 5.8–12.7 (9.5–12.7; normalized ~10.5) | 12.8–19.7 (17–19.7; normalized ~18) |
| 1440 dark AFTER | 0–2.8 | 2.9–3.7 | 3.8–5.7 | ~5.8 | 6–13.7 (11–13.7; normalized ~12) | 13.8–20.8 (18–20.8; normalized ~19) |
| 390 light BEFORE | 0–2.3 | 2.4–3.3 | 3.4–4.7 | ~4.8 | 5–12.2 (9.5–12.2) | 12.3–18.9 (16–18.9) |
| 390 light AFTER | 0–2.3 | 2.4–3.3 | 3.4–4.8 | ~4.9 | 5–12.4 (10–12.4) | 12.5–19.3 (16.5–19.3) |
| 390 dark BEFORE | 0–2.2 | 2.3–3.6 | 3.7–4.6 | ~4.7 | 4.9–10.2 (8–10.2) | 10.3–15.9 (13.5–15.9) |
| 390 dark AFTER | 0–2.3 | 2.4–3.6 | 3.7–4.8 | ~4.9–5.2 | 5.3–11.6 (9–11.6) | 11.7–18.7 (16–18.7) |

The explicit narrated timestamps are absent from the current index but can be added directly from this table. This formatting omission does not weaken the source bug proof or require recapture.

## FRICTION

- My fresh preview context was unpaired. The authorized pair command generated a fresh token; navigating the pairing link did not submit it. Manual input and Continue succeeded. No token or pairing URL is included in this report or saved audit artifacts.
- Browser-local draft IDs were not shared. The published recipe successfully seeded an independent concrete A draft; the result still reproduced the delivered image exactly.
- Desktop snapshot output is 1280×800 despite independently measured 1440×900 CSS layout; narrow output is 780×1688 at DPR 2. This is disclosed and reproducible, not evidence of a wrong CSS viewport.
- Recordings contain reload transients and panel normalization. They prove the bounded title/grouping interactions, not timing/performance or sidebar-navigation coverage. Still artifacts are settled and uncontaminated.
- Fixture/driver scripts are intentionally uploaded outside the repository under the user's instruction; do not apply stale matrix language as a commit requirement.
- No blocking proof gap or source defect found. Existing server remains running; source remains clean at the requested head. Own tab remains on rebuilt A/light/1440 with closed panels.

Audited by gpt-6.1-sol through the Codex harness in T3 Code.