# Tooltip proof audit — pass 4

**SUFFICIENT for the scoped file-path specification and representative non-file behavior.** The two remaining pass-3 proof gaps are resolved: six uncontaminated static parity controls and an explicit M1 cancellation/reopen pair. No missing required row remains within that scope. This is an evidence verdict, not a CI/merge authorization or a claim that every shared-tooltip caller retains on entry.

Read-only audit of [#11558](https://git.cafe/pingdotgg/t3code/pulls/11558), [GitHub mirror #11558](https://github.com/pingdotgg/t3code/pull/11558). Reviewed original proof-audit brief and its final dispatcher notes through the lead thread, pass-1 matrix, pass-2/pass-3 audits, replacement body draft, initial/supplemental indexes, and the normalized round-3 index/archive. Latest scope clarification takes precedence over the original matrix's overly broad glass/P1 wording.

## Findings and publication corrections

No new production blocker established. P1 error-entry closure is the accepted pre-existing controlled-PreviewCard exception outside the file-path spec, not a retention PASS and not a requested fix. Glass selection remains unverified. Harnesses are explicitly component evidence, including P1's query stub; no real PR-detail route or network integration is certified.

The replacement body draft is stale rather than contradicted: it still says M1 cancellation, normalization and final proof audit are pending. Before publication the lead should replace those placeholders with this verdict and the normalized index, update M1 to right-click + focused Shift+F10 → tooltip suppression → outside-click menu absence → reopen, and append final-SHA provenance. Keep the existing P1/glass/harness/clipboard limits. Original table SHA labels remain historical capture pins; do not relabel those images as final-HEAD captures. Link final proof separately.

## Final revision and code-derived coverage

Final merge [a429aa3](https://github.com/pingdotgg/t3code/commit/a429aa3561ed9a0d7afc588285b5aa747299f645) has parents capture [3490923](https://github.com/pingdotgg/t3code/commit/3490923fb6c723fb50139d7184173ba18f4ee444) and updated main [76da264](https://github.com/pingdotgg/t3code/commit/76da264a0ccaeff5fa88bcb5cfada1be6ae320c4). `git diff --quiet 3490923 a429aa3 -- apps/web/src` exits 0: all web source is byte-identical across capture and final. Against updated main the complete diff remains exactly two files, +57/−2: 52 test lines and the two production hunks in shared tooltip.tsx. Final worktree is clean. The main merge introduces no additional PR-owned production behavior to prove.

The popup class affects default, code and glass variants through their shared primitive; positioner remains pointer-transparent. The containment guard changes only scroll dismissal inside the sole production TooltipScrollDismissArea caller, MessagesTimeline. Followed current Markdown file primary-anchor/fallback-button convergence, scroll child, ordinary URL, controlled MediaActions, controlled PR PreviewCard, sidebar mini-list and disabled/overlay caller shapes. Sidebar footer still supplies no activation callback. Noninteractive label-only callers collapse into the same event shapes; equivalence is a source decision, not an individual runtime PASS. Provider, transport, contracts, scheduler and native mobile code are outside this PR diff. Web is the defined surface; no separate Electron/mobile capture claim.

| Row | Final scoped status / evidence boundary |
|---|---|
| F1 | PASS retained: four viewport/theme modes, native full-path selection/copy/paste and leave. Pass-2 independent 0-pixel reproduction retained. |
| F2 | PASS component branch: actual fallback button, native exact selection/copy/paste, full/relative options, cancel; desktop/narrow. Real fallback route not exercised. |
| F3 | PASS real timeline desktop/narrow: entry and horizontal popup scroll retain; unfocused timeline movement dismisses. Final-HEAD independent repeat below. Keyboard-only focus diagnostic from pass 3 retained; no universal mixed-hover/focus guarantee. |
| F4 | Source-equivalent unchanged dispatch; F2 menu smoke representative. Individual editor/browser/media/directory destinations not run. |
| N1 | PASS retained ordinary URL four-mode hover/focus/Escape/leave. Actual destination navigation not claimed. |
| N2 | PASS real attachment preview dynamic label/action, native paste containing fixture contents, sibling transfer/Escape. No exact plain-text clipboard equality claim. New static parity control accepted. |
| G1 | PASS scoped real sidebar hover/sibling transfer; normalized desktop/narrow controls. Footer activation only in labeled callback-combination harness; real footer informational. Text selection UNVERIFIED. |
| G2 | Source-equivalent existing explicit pointer-events-auto interactive badge; no new independent badge list/scroll pass. |
| G3 | Source-equivalent default/glass switch; dynamic/overlay shapes represented by N2/M1/E1/P1. No updater-install runtime claim. |
| C1 | PASS labeled production-component copy/wrapping harness desktop/narrow, full native paste and Copied state. Exact close latency not measured. |
| C2 | Source-equivalent code wrapping; no direct diagnostics/changed-files runtime pass. |
| M1 | PASS: new paired real MediaActions recording explicitly proves right-click and Shift+F10 suppression, menu absence after outside click, dialog retained, tooltip reopening on both pins. Escape cancellation not certified. |
| P1 | ACCEPTED UNCHANGED EXCEPTION: correct error tooltip opens then closes on entry on both pins; success PreviewCard excludes tooltip. Query-stub component harness only. |
| H1 | PASS real snooze suppression/cancel/reopen, sortable drag with tooltip disabled, neighbor control; desktop/narrow. Other hidden/stale-label roots source-equivalent. |
| E1 | PASS labeled production disabled-wrapper/nested-menu harness with zero disabled callback count and dismiss/reopen. No PR-route integration claim. |
| L1 | Collision code unchanged, representative narrow wrapping/cleanup covered. Existing F3 width 480 at x=5 exceeds 390px viewport on both pins. Native-titlebar placement not run. |
| A1, Z1 | Source equivalence to covered label/wrapping/menu/disabled shapes; no blanket runtime claim. |

## Normalization and M1 checks

[Normalized evidence index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/eadf711206a24c91/index.md) and its [public archive](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/dd5ee0fb48dddcfe/ev11558-r3-public.tar.gz) downloaded with curl. Index byte-matches local final index; all 92 archived files byte-match corresponding local public files. Source snapshots match git-show pins: before SHA-256 `21a0bb262e60917f81372d6ea7aab3d8889a7517837a04af833e4c877d3319fc`; capture/final SHA-256 `80c5237b1cb9a74933f0684748cb75953501d326df4362a2ba43900ea19ab4b3`.

Independently recomputed every new pair with Pillow/NumPy. Six pairs F3 desktop/narrow, G1H1 desktop/narrow, M1 desktop and N2 desktop have **0 total changed pixels at >0 and >16**; all twelve delivered diff images exactly match recomputation. No masking, cropping, resampling or alignment. State JSON independently agrees on viewport/DPR, timeline top/height/client height, trigger rectangle, draft, attachments, focus and fonts; correct visible popup text and geometry agree, pointer-events changes none→auto. These are **hover-trigger static parity controls**, never entry/retention proof. Existing valid recordings remain the behavioral evidence.

Opened all new controls, all new diffs and all sixteen M1 state stills plus paired review frame. Independently decoded both full M1 raw videos and paired captioned video into one-second temporal samples spanning their duration and inspected all samples. Re-inspected retained initial F1/N1/G1 video contact sheets and supplemental capture/diff/video galleries; prior pass-2/pass-3 full inspections and independent diff computations remain part of the audit record. This is state-sampled video review, not a claim of continuous playback of every historical clip. No blank/loading/ghost frame is credited. Deliberate attachment dialogs and the narrow sidebar drawer are valid fixture state.

M1 JSON and visuals now distinguish menu absence from tooltip absence. On BOTH pins: initial demo.png tooltip; right-click Save image/Copy image menu with tooltip null; movement over exposed image still suppressed; outside click produces empty visible menu-item array while dialog remains; leave/rehover restores demo.png; focused Shift+F10 repeats menu/suppression; outside click empties items; final rehover restores tooltip. First click also zooms image on both pins; second returns the smaller image. These are existing image actions and explicitly disclosed, not accidental changed background. Paired captions correctly name each transition. Exact millisecond timing is not inferred from one-second samples.

## Independent final-HEAD reproduction

Reused retained synthetic route, headed Chromium CDP 9158 and Xvfb :158, 390×844 light, DPR1, fine pointer, reduced motion, fonts loaded and frozen browser Date. No source edit, checkout swap, response override, provider turn or live T3 access. Reloaded actual app and read its **executed** Vite tooltip module using Debugger.getScriptSource. Both final class and containment guard are in that executed module; inline sourcemap sourcesContent exactly matches git-show final tooltip source. This is the production component served through Vite, not a production-build bundle claim.

- Executed module SHA-256: `a4c7faaaf06b525df78b9a7870d1eb964eff426b136c4f932a69fc7ba119583d`.
- Saved sourcemap JSON SHA-256: `904b032b82f11a994fe573ddf0e364c4bdeb120525048647a899a43962f52012`.
- Source hash: `80c5237b1cb9a74933f0684748cb75953501d326df4362a2ba43900ea19ab4b3`.

Recorded steps: approach 0.000s; cross into correct long popup 2.994s; wheel popup 6.903s; actual timeline scroll 9.609s. Correct identity retained on entry; scrollLeft 0→240 with popup retained; popup :hover true immediately before actual timeline scrollTop 1250→1170; popup then null. Recording duration 12.316s. Captioned export adds a 70px black header above the unchanged 390×844 recording; raw footage remains in the bundle. Independent screenshots use ffmpeg x11grab with mouse excluded and exact repeated-frame equality 500ms apart.

First repeat had branch notice/Scroll-to-end/footer state contamination: **26,861 pixels >0, 3,335 >16, mean absolute RGB 1.10564** against delivered normalized F3 narrow after. Preserved as rejected normalization output, not anti-aliasing or production regression. Used the actual branch-notice dismiss control and restored the prescribed normalized state, then repeated: **0 / 329,160 pixels >0; 0 >16; mean absolute RGB 0.0** against delivered F3-390x844-after.png. Popup geometry identical (x=5, y=397.703125, width=480, height=37.875). The final normalized still is supplementary final-SHA evidence; the recording separately proves affected behavior. No network-idle observer was installed for the independent repeat, so network-idle is unverified, not a claimed pass or detected load.

[Final-HEAD reproduction recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0298e4bea571e86a/final-F3-repro-captioned.mp4) · [Executed-module/source proof](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4cf651ad8cc643e9/final-source-proof.json) · [Normalized repeat state and pixel result](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/91c6f0c869a03413/normalized-repro.json) · [Complete reproduction scripts, module, sourcemap, raw/captioned video and audit verification JSON](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4982858ad1e3ebe1/audit11558-pass4-repro.tar.gz).

![Final HEAD narrow F3 normalized repeat: correct long-path tooltip visible; zero-pixel delta](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8aace497d64c5206/final-F3-normalized.png)

Earlier evidence remains [initial index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d9363ff4f654d120/index.md) and [supplemental index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ce407a750ecd0788/index.md). Before source is precisely the two production hunks reversed through CDP in the final surrounding app: component A/B, not whole-checkout baseline. Reproduction dependencies come from those public archives and round-3 archive; scripts retain their documented /tmp prerequisite paths.

## Limits, handoff and friction

P1 error-entry retention and glass selection are not certified. No real PR-detail network/route integration for harness fixtures, exact N2 plain-text clipboard equality, every equivalent tooltip caller, coarse-pointer/native-mobile behavior or OS-titlebar geometry is certified. Pre-existing narrow path clipping remains. No requirement to fix these was added. No tests rerun: final scoped test/CI gating belongs to lead; this pass only checks proof and source identity.

No source/data writes, commits, pushes, forge comments/reviews, watch or process teardown. Only synthetic browser UI, temporary audit outputs and requested public ghdrop uploads. PR linked; watching=false. CDP released, no driver remains. Existing server/browser/Xvfb processes retained for lead cleanup; no process killed or replaced. Old server PID 2063373 no longer had a readable cwd at inspection; retained route was healthy, so no restart was attempted. Vite 2063360 and Chromium 2014664 still own 5953/9158; Vite cwd is this worktree's apps/web. Worktree remains clean at exact final HEAD. Model/harness: GPT-6.1-Sol, Codex through T3 Code.

FRICTION:
other | First independent frame retained branch notice and scroll control, yielding 26,861 changed pixels | Run the normalized driver’s real dismiss-control step before measuring; final repeat is exactly zero.
tool | Initial subtitle export overlapped popup text | Re-export captions in a separate black header and inspect the resulting frame.
brief | Body draft still carries superseded pending proof statements and historical SHA columns | Update pending statements and append separate final-SHA evidence; preserve historical capture labels and explicit limits.

VERDICT: SUFFICIENT