# Final read-only proof audit — composer chip order

**The requested replacement's proof is sufficient. All eight mandatory rows pass; the four replacement C1/C2 videos close the sole remaining capture gap. No new code regression or mandatory proof gap found.** This is evidence acceptance for the exact revisions below, not current-PR-head merge/CI certification.

PR [#17892](https://git.cafe/pingdotgg/t3code/pulls/17892) / [#17892](https://github.com/pingdotgg/t3code/pull/17892).

- BEFORE: main snapshot [eea9895368](https://github.com/pingdotgg/t3code/commit/eea989536849e5b5049987ee24318b5a77326542), full SHA `eea989536849e5b5049987ee24318b5a77326542`.
- AFTER: local `takeover/pr-17892` [c772f425b1](https://github.com/pingdotgg/t3code/commit/c772f425b12ab22d7455d5de616370ece0d2a295), full SHA `c772f425b12ab22d7455d5de616370ece0d2a295`.
- Original provenance only: [NikitaMGrimm/t3code](https://github.com/NikitaMGrimm/t3code), `fix/composer-compact-chip-order`, [263670f223](https://github.com/NikitaMGrimm/t3code/commit/263670f2239d6118a60b4413c571fe8446d48627), full SHA `263670f2239d6118a60b4413c571fe8446d48627`.

Fresh GitCafe metadata still points to the original fork SHA, not the requested replacement. Local main has also advanced to [07d38b8d17](https://github.com/pingdotgg/t3code/commit/07d38b8d17167c8d17fba1184dbe13a307fada74). Neither substitutes for the explicit comparison. HEAD and replacement branch resolve exactly to AFTER; merge-base with the exact BEFORE resolves to BEFORE. Clean worktree. No claim that the published PR currently contains this replacement.

## Coverage derived from the source matrix

[Matrix](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/72ef6f316babd4df/matrix.md), [accepted index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5eedb81bae30cd35/capture-index.md), [prior independent audit](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b774d7e52e2f5032/audit.md), [desktop addendum](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/deb8dfa3e58fdbd4/addendum.md), [new manifest](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5378187cecdbad13/manifest.json).

| Row | Final status | Accepted proof and limits |
|---|---|---|
| C1 | PASS | 1440×1000 light, real meter. Accepted Compact/Full stills plus fresh BEFORE/AFTER replacement clips: Compact→Full→Compact→empty/chip absent/Resume→same draft/chip restored. |
| C2 | PASS | 1440×1000 dark, meter off. Accepted Compact and final clean Full stills plus same complete replacement sequence on both SHAs. |
| C3 | PASS, carried forward | 390×844 light, meter on: both chip widths, actual absence/Resume/restoration, tooltip, real Tab/Shift+Tab. Tab alters narrow layout; no stationary full traversal claim. |
| C4 | PASS, carried forward | 390×844 dark, meter off: both widths, absence/Resume/restoration, blur/collapse/reopen. Final clean Full accepted; independent replay exact. |
| S1 | PASS, carried forward | 1440×400 light, actual resting predicate and overflowing timeline; toggles, expanded return, meter popover, benign file select/clear. Deliberately different height. |
| S2 | PASS, carried forward | 1440×1000 light; actual held/released detail response, reserved blank→real meter. Attach (995,933), input/editor identity stable. Intentional loading UI accepted. |
| S3 | PASS, carried forward | 1440×1000 light; chipless Implement still identical, real menu/Escape, feedback→Refine, clear→Implement; no dispatch. |
| K1 | PASS, carried forward | Actual keyboard forward/reverse. BEFORE Attach→meter→chip→primary; AFTER chip→Attach→meter→primary. |

C1–C4 establish both requested widths and themes with real-meter/absent-meter paths. S1 proves actual resting separately; S2 proves reservation rather than loaded-empty state; S3 covers the distinct plan menu. Optional pending-question, Attach-hidden, queued-edit and running Stop/Queue/Steer branches remain source-reviewed, not newly captured. They require no extra Cartesian matrix. Responsive web is not native mobile; packaged Electron was not separately run. No provider/contract/backend/native changes; remote modes share this renderer. No broader platform or provider execution claim.

## Fresh replacement artifact inspection

Downloaded all 42 addendum manifest artifacts and independently checked SHA256 and byte size: zero mismatches. Public addendum and manifest match their local originals; public prior audit matches its local original byte-for-byte. Fully decoded all four downloaded MP4s with `ffmpeg -v error -xerror -i video -f null -`: all exit 0, no decoder errors. Each is 1440×1000, 30fps. Re-extracted encoded frames at **1/3/5/7/10.5 seconds from each downloaded video** and opened all 20 at original composer crop scale in four fresh sheets; additionally opened all four full-frame 7-second absence frames. This is actual decoded-video inspection, not structural/trace-only acceptance or a claim every frame was manually viewed. No browser/server/capture launched.

All four show the same original draft, actual Full toggle, Compact restored, empty placeholder/no chip/blue Resume, then exactly “Review the synthetic fixture.” and Compact returned. Light/dark and meter on/off match their cases. Controls are visible and unclipped. No selected-text-only deletion, ghost duplicate, obstructing dialog or unrelated loading state in these checkpoints. Lossy H264/caret/cursor differences are not presented as new still pixel comparisons. All accepted prior stills remain unchanged; rejected dark Full captures stay superseded by final clean controls-only index entries.

| Video | SHA256 | Encoded duration |
| [after-C1.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7c9b646150d20466/after-C1.mp4) | `aabc8e7d06e32bf6ac44bc651875160375489756a1106992c55053f33284b866` | 11.733333s |
| [after-C2.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7ded2390b2e73b4e/after-C2.mp4) | `39bafd0b237638140b7d0ed318614e91bbd27a45aa2e8f2c64a0b15d1325f4e8` | 11.666667s |
| [before-C1.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f0733d523f87d5f8/before-C1.mp4) | `04208f20cde92e742d9eac14ed5d4265f8cda7d593bd6c165291764fd5a47753` | 11.633333s |
| [before-C2.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/007d2c04f376cc69/before-C2.mp4) | `62324a2a91b5d2a4cdff2ed7bb57189386260aa493b94bd74e635fa1ccf677f2` | 11.766667s |

## Geometry, identity and focus checked against the trace

Independently asserted all 20 labeled checkpoints in the downloaded trace, not just the summarized geometry. Viewport1440×1000, DPR1/zoom1 throughout; fixed synthetic clock; empty file input/value with zero files. Original-reference equality and original connectedness hold for Attach, input, editor and enabled meter on BOTH builds. Equality is document-local, not across reloads. Meter remains absent in C2. Focus is editor initially, chip after each desktop pointer toggle, editor during absence and restoration, matching normal expanded-desktop behavior; no claim focus stays in editor during desktop clicks.

| Case/build | Attach x: Compact / Full / Compact / absent / restored | Meter x |
|---|---|---|
| C1 BEFORE | 855.453125 / 888.609375 / 855.453125 / 995 / 855.453125 | 891.453125 / 924.609375 / 891.453125 / 1031 / 891.453125 |
| C1 AFTER | 995 / 995 / 995 / 995 / 995 | 1031 throughout |
| C2 BEFORE | 891.453125 / 924.609375 / 891.453125 / 1031 / 891.453125 | absent |
| C2 AFTER | 1031 / 1031 / 1031 / 1031 / 1031 | absent |

Attach/meter stay y933, size28×28. Primary stays (1067,931,32,32) on BOTH builds in all states, including Resume. Editor stays (393,849,710,78). BEFORE toggle moves Attach +33.15625px; AFTER toggle and absence/restoration move it 0px in x/y. BEFORE primary node IDs change entering/leaving Resume (C1 5/5/5/7/9; C2 4/4/4/6/8); AFTER retains its primary (C1 all5; C2 all4). Chip itself legitimately disappears/remounts. The driver uses real click, Control+A, Backspace, awaited empty text/chip-null/visible Resume, then real typing; no DOM/style mutation to manufacture absence.

## Corrected narration suitable for the PR

**Encoded-video observed sample times, not transition onset times or ffmpeg-spawn wall time.** B/A means BEFORE/AFTER seconds. New C1/C2 replace the old desktop recordings completely; all other times below carry forward unchanged from the independently inspected prior audit. Every listed time is within its corresponding encoded recording.

| Row | Visible sequence at encoded sample times B / A (seconds) |
|---|---|
| C1, replacement | Compact 1/1 → Full 3/3 → Compact 5/5 → empty+chip absent+Resume 7/7 → exact same draft+chip 10.5/10.5. |
| C2, replacement | Compact 1/1 → Full 3/3 → Compact 5/5 → empty+chip absent+Resume 7/7 → exact same draft+chip 10.5/10.5. |
| C3 | Compact 0/0 → Full 3.5/3.0 → Compact 5.5/5.5 → empty+Resume 6.5–7.0/6.0–6.5 → draft+chip 7.5/7.0 → tooltip 9.0/8.5 → Tab/provider focus 9.5/9.5 → reverse/editor 10.0/10.0. |
| C4 | Compact 0/0 → Full 3.0/3.0 → Compact 5.5/5.5 → empty+Resume 6.0–6.5/6.0–6.5 → draft+chip 7.0/7.0 → collapsed 8.0–8.5/8.0–8.5 → expanded 9.0/9.0. |
| S1 | Expanded 0/0 → resting 1.5/1.5 → Full 3.5/3.5 → Compact 4.0/4.5 → end/expanded 5.0/5.5 → popover 6.5/6.5 → attachment 7.0/7.0 → cleared final frame 9.13/9.23. |
| S2 | Loading+reserved blank 0.5–1.5/0.5–1.5 → actual detail release and meter visible by 2.5/2.0. No rejection for intentional loading UI. |
| S3 | Implement 0/0 → menu 4.5–5.0/4.0–4.5 → Escape/menu closed 5.5/5.0 → feedback+Refine 6.0/6.0 → clear+Implement 7.0/7.0. |
| K1 | Editor 0/0; forward Attach/meter/chip/primary at 2.5/3.0/3.5/4.0 BEFORE; chip/Attach/meter/primary 2.5/3.0/3.5/4.0 AFTER. Reverse visible 4.5–6.5 on both. |

Do not reuse the old desktop “selected draft equals absence” narration, or older spawn-relative seconds (notably S2 release and S3 restoration). The new encoded-narration.json agrees with the fresh samples. Older corrected rows retain their 0.5s sample resolution; final clear frames are explicitly identified by their actual sample times.

## Carried-forward pixel audit and independent replay provenance

As explicitly authorized, carry forward the prior exhaustive artifact checks unchanged: 18 raw stills, 18 diff images, all 16 videos and independent recomputation of all nine still pairs. Prior public verification downloaded all86 visual artifacts with zero hash mismatches. This final pass does **not** claim to have repeated those full inspections. Their exact results and proof files remain in the linked prior audit. Amplification ×8/clamped255 with disclosed nonzero floor [90,20,90], no masks, thresholds max-RGB >0 and >8/255, inclusive bounds:

| Still pair | Dimensions | Changed >0 | Changed >8 | Changed bounds |
|---|---|---:|---:|---|
| C1-compact | 1440×1000 | 2023 | 1886 | [855, 931, 1058, 962] |
| C1-full | 1440×1000 | 1721 | 1528 | [889, 931, 1058, 962] |
| C2-compact | 1440×1000 | 1457 | 1346 | [891, 931, 1058, 962] |
| C2-full | 1440×1000 | 1161 | 885 | [925, 931, 1058, 962] |
| C3-compact | 390×844 | 2110 | 1980 | [113, 779, 320, 814] |
| C3-full | 390×844 | 1764 | 1551 | [147, 779, 320, 814] |
| C4-compact | 390×844 | 1567 | 1466 | [149, 779, 320, 814] |
| C4-full | 390×844 | 1302 | 1022 | [183, 779, 320, 814] |
| S3-implement | 1440×1000 | 0 | 0 | None |

All changes remain inside chip/Attach/meter reorder regions; S3 is pixel-identical. No unrelated sidebar, composer outline, timeline or primary-button changed pixels accepted. Final clean C2/C4 Full had zero change outside actual chip/Attach union.

The prior audit independently rebuilt actual AFTER C4 Full from the public synthetic SQL in a fresh isolated home, using actual AFTER source, Chromium148, DPR1/zoom1, viewport390×844, dark/meter-off, fixed clock and matched rendering flags. **changed0=0, changed8=0, max RGB delta=0; no bounds** against the accepted AFTER C4 Full. [Replay delta](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bdf4becc512e5e61/replay-delta.json). This was an independent executed replay in the prior pass, not a replay newly launched here. Production source and accepted stills are unchanged, so no redundant browser/server replay was needed or performed.

## Source claims and audit limits

Fresh exact-SHA diff is still exactly three web files, 106 additions/83 deletions. [ChatComposer](https://github.com/pingdotgg/t3code/blob/c772f425b12ab22d7455d5de616370ece0d2a295/apps/web/src/components/chat/ChatComposer.tsx#L3904) supplies stable attachment/focus callbacks; [FooterControls](https://github.com/pingdotgg/t3code/blob/c772f425b12ab22d7455d5de616370ece0d2a295/apps/web/src/components/chat/ComposerFooterControls.tsx#L195) passes Attach/meter/reservation as leadingActions; [PrimaryActions](https://github.com/pingdotgg/t3code/blob/c772f425b12ab22d7455d5de616370ece0d2a295/apps/web/src/components/chat/ComposerPrimaryActions.tsx#L133) returns fixed chip/leading/action fragment slots in every branch. Removed nested wrapper and reordered siblings explain the measured movement fix and keyboard change. Chip suppression for Resume/queued edit remains explicit. Narrow pending-answer caller supplies neither chip nor leading controls. Existing branch contents, action selection, callbacks, eligibility, labels and contracts remain unchanged.

Visible change is placement/order and corresponding Tab order: chip before Attach, optional meter and existing primary action. Without chip, controls retain their former order. This audit supports the bounded evidence claims above, not every statement in the live PR body's historical evidence on other SHAs. The supplied two fresh reviewer MERGE conclusions remain supplied provenance, not newly commissioned reviews. No tests or CI rerun or certification in this final proof pass. No source changes, commit, push, comments, watch, browser, dev server or live-userdata access. All shell commands explicitly cd into `/home/ubuntu/wt/pr-17892`; new scratch only `/tmp/pr17892-audit-final`. Existing capture-process cleanup is retained provenance; this pass started no capture/server process.

| FRICTION category | Happened | Avoidance |
|---|---|---|
| Revision scope | Local main advanced; live PR metadata names original fork head | Pin exact BEFORE/AFTER SHAs and separate replacement proof from current-PR-head claims |
| Prior input gap, resolved | Old desktop deletion selected text and retained chip | Real keyboard deletion plus empty editor/chip-null/Resume predicates, visible encoded checkpoint |
| Prior timing gap, resolved | Spawn-relative narration outran encoded footage | Use final-MP4 sample times; separate wall-clock telemetry; retain corrected prior table |
| Trace identity | BEFORE primary replaces nodes on Resume; AFTER preserves them | Assert actual IDs and original references per document; distinguish identity from fixed rectangles |
| Final inspection | No download, decode or assertion failures | Download public bytes, verify hashes, full-decode videos, inspect fresh encoded samples |
| Upload | First ghdrop POST failed with a transport error | One bounded retry; verify returned public report bytes |

Audit by GPT-6.1-Sol through Codex in T3 Code. Final proof acceptance for the exact requested replacement; no mandatory missing item remains.

VERDICT: SUFFICIENT