1. **P01: comparison masks include a change outside this PR.** Every P01 pair, including the tool-region crop, includes `Worked for 30s` versus `Worked for 14s`. Both thresholds light up those unrelated duration glyphs. The 1280/light masks have 1,111 changed pixels at Diff0 and 970 at DiffT16. The real inline panel appears correctly, but the brief explicitly rejects a diff lit by something the PR does not change. The latency difference is truthfully disclosed under “What else changed”; disclosure does not make this a controlled comparison. Files: all `P01-Diff0-*` / `P01-DiffT16-*`, particularly [P01-DiffT16-1280x800-light.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/90ac39cfed7117c2/P01-DiffT16-1280x800-light.png). Keep the original real-provider captures and supply an independently controlled comparison that isolates the changed region.

2. **P02: an inspector capture is visibly overlapped, and the settled comparison has mismatched hover state.** In [P02-after-terminal-inspectors.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2ecb4641666ecff6/P02-after-terminal-inspectors.mp4) at 1.000s, `Synthetic ACP fixture complete.` lies across the expanded JSON, and the interruption divider remains across its lower portion. The corresponding before recording also shows overlap during inspector expansion. This is an unacceptable frame under the brief; it is not evidence of a regression caused by this parser change. The final still pair additionally has a hovered `1:00 AM` timestamp on the base side without the corresponding head hover. The masks include that unrelated hover difference; it is not listed specifically in “What else changed.” Later inspector frames are readable and the sequence does eventually close and collapse, but the submitted inspector recording needs a clean, settled acceptance interval.

![P02 head at 1.000s: final reply overlaps the expanded inspector](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bb2179d4a978402f/P02-head-ghost-at-1s.png)

3. **P03: the uploaded seed invocation does not reproduce the delivered frame.** I rebuilt this fixture on the requested head in a new, unseeded scratch T3 home, through the uploaded normal-RPC seeder and ACP shim, with only scratch paths relocated. The supplied `seedHarness('proof-p03-v2','P03')` creates title `P03 — labelled synthetic ACP fixture` and prompt `P03 synthetic fixture`; the delivery uses title `P03 — labelled synthetic ACP recovery` and prompt `P03 synthetic ACP recovery classes`. The footer provider initials differ too. Elapsed time is 1m19s in my fresh run versus 50s in the delivered frame. These are visible, non-antialiasing differences. Upload the exact invocation/configuration used for the accepted fixture, or an updated matching reference. This is a reproducibility finding, not a failure of HTML recovery.

   Rendering environment independently checked: fs-36, Linux x64, shared preview Chrome 156.0.0.0, CSS1280×800, DPR2, saved1280×800, light, root16px, fonts loaded; uploaded browser-Date freeze script applied. I ran `T3CODE_DEV_SEED=0 vp run dev --home-dir <own scratch>/run-20261103T0157-fresh/t3-home` from the requested checkout. No live userdata was used. The supplied ready/release milestones produced actual MCP-published A/B pages and ordinary C/D controls. Two final PNG exports 44.153s apart were pixel-identical.

   Against the delivered MP4 frame at 10.500s (2560×1600, normalized to1280×800 with Pillow Lanczos), 41,996/1,024,000 pixels differ at Diff0 (4.1012%); 6,973 exceed16/255 (0.6810%). Both actual pages occupy the same 728×80 CSS rectangles at x276, y209 and297; no page-layout difference was observed. Page-region >16 counts are162 and161/58,240 each. The comparison also includes H264 compression/downsampling and text rasterization, so it cannot establish a lossless screenshot match. [Replay PNG](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/624bedd3ef6339d9/repro-p03-final.png), [DiffT16](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/acaa97ef9b98cba3/replay-diff-16.png), [pixel measurements](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e90d4705ee1593a1/pixel-delta.json).

4. **P04: partial coverage remains insufficient.** Working/waiting, ordinary concurrent tools, async/missing-mode delegates, real child tracking, priority forms, and release/completion are shown. The required approval/input priority *thread-list badges* are not shown. The native report acknowledges this; the web priority recording also loses the working shelf below the viewport during the relevant forms. The web echo approval row and native approval/input forms do not establish the sidebar/thread-list precedence claim. Inspect [web controls](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/eb6b369b48d38ecb/P04-after-web-controls.mp4), especially55–140s, and [iOS controls](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/317909cbb1688f75/P04-after-ios-controls.mp4), around80–140s. This gap is material because canonical wait identity changes this status branch. It cannot be waived by saying that an approval form is visible.

5. **P07: the wide dark milestone has the wrong iframe theme.** In [P07-after-responsive.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5be2d3cc9b72dfe3/P07-after-responsive.mp4), the shell is dark but the actual page remains white with dark text at2.326s,2.8s and4.0s. Resizing to390×844 produces a correctly dark page by4.626/5.5s. The delivered wide/light and narrow/dark stills are readable. The transition does not establish a correctly themed wide/dark page before resizing, and “What else changed” does not mention the stale light iframe. This is a contradicted capture milestone, not an attribution of the theme behavior to this nine-line diff.

![P07 at 4.000s: a light iframe remains inside the dark app](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/73aaa7490caf504b/P07-wrong-theme-at-4s.png)

6. **Head-only comparative claims need qualification.** Per the user's clarification, I apply the pass-1 matrix minimum: pairs plus both masks only forP01/P02; P03–P09 may be head-only. However, P01/P02 base media do not show P04 wait-status behavior, P05 delegate suppression with an app-owned child, P06 labels inside the native ACP child caller, or P08/P09 native-client labels. Those rows establish the submitted *head* states, not those particular changes from base. P03's text-only A class has the P01 base anchor, but its repeated/structured-reference classes have no same-fixture base media. These are findings for the lead to decide whether to capture base sides; I have not invented an additional mandatory pair requirement after the clarification.

7. **P08's “unsettled fullscreen” limitation is overstated.** Contrary to the report's implication, the submitted video contains a clean fullscreen page with no spinner at9.5s,10.0s and10.5s. It then returns to a clean inline feed. Loading frames at6–8s are transition states, not the accepted settled outcome. P08 satisfies the head navigation minimum; amend the narration to identify the clean interval rather than claiming it was never established. [Independent10s frame](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/58f11ec38b6c6503/P08-settled-fullscreen-at-10s.png).

8. **P10 may remain explicitly UNVERIFIED for this server-only diff.** No guest capture exists; `connect_unavailable` is a blocker transcript, not proof. I accept a documented exception rather than requiring a T3 Connect deployment for this change. The only production change is identity normalization; the guest path consumes the same stored canonical tool and parsed reference validated byP01, and the guest component/service/route do not change. [GuestShareService.render](https://git.cafe/pingdotgg/t3code/blob/a5c3df8b2124ce221e4162332eaffa6090779812/apps/server/src/guest/GuestShareService.ts#L653) uses [htmlRenderFromToolItem](https://git.cafe/pingdotgg/t3code/blob/a5c3df8b2124ce221e4162332eaffa6090779812/packages/shared/src/toolOutput.ts#L185); guest authorization, attachment ownership, and guest rendering remain unverified end to end. Do not count P10 as captured or claim all ten rows passed.

| Row | Status under matrix minimums | Inspected coverage / limit |
|---|---|---|
| P01 | INSUFFICIENT | All5 viewport/theme pairs plus crop viewed; SHAs correct; inline recovery visible; duration differences contaminate both masks. |
| P02 | INSUFFICIENT | All6 videos, terminal stills and masks viewed; canonical labels/errors/grouping visible; inspector overlap and unequal hover invalidate comparison proof. |
| P03 | INSUFFICIENT | All3 videos show A/repeat/B/C/D classes; independent fresh replay succeeds functionally but does not reproduce delivered metadata/pixels. Repeat/structured base claims unanchored. |
| P04 | INSUFFICIENT | All4 web/iOS videos viewed; missing approval/input thread-list precedence proof; comparative wait-state claim has no base capture. |
| P05 | SUFFICIENT for head | All4 web/iOS videos viewed; running/completed matching card, controls, child open/return shown. Suppression change from base unproven in base media. |
| P06 | SUFFICIENT for head | Native child video/trace viewed; canonical labels, absent fallback and parent return shown. This caller's base presentation is not captured. |
| P07 | INSUFFICIENT | Both stills and responsive video viewed; narrow dark is readable, but wide dark milestone retains light iframe. |
| P08 | SUFFICIENT for head | iOS still and video viewed; settled fullscreen9.5–10.5s and return shown. Native label change from base is unproven. |
| P09 | SUFFICIENT for head | Android actual1080×2424 light still viewed; settled page/canonical labels visible. Native label change from base and logical viewport/DPR are unverified. |
| P10 | UNVERIFIED; acceptable exception | No capture; shared stored-data contract and unchanged guest implementation justify a limited exception, not a claim of guest testing. |
| C01 | NO VISUAL MINIMUM | Negative/precedence classes are protocol controls. Added source test covers positive title and two negative server names; full boundary list was not independently executed in this proof pass. |
| C02 | NO VISUAL MINIMUM | Historical data is not migrated; no historical comparison claimed. |
| C03 | NO VISUAL MINIMUM | Other providers/normal canonical paths are unchanged; no cross-provider runtime proof claimed. |
| Q01 | OPTIONAL; UNVERIFIED | Asset/network failure/retry paths not exercised. |
| Q02 | OPTIONAL; PARTIAL | P08 covers native open/back; full web/desktop viewer-action matrix not exercised. |
| Q03 | OPTIONAL; PARTIAL | P07 and P09 provide representative fit; overflow/process-death sizing stress not exercised. |

Coverage and checks: read-only review of [#29141](https://git.cafe/pingdotgg/t3code/pull/29141) ([GitHub mirror](https://github.com/pingdotgg/t3code/pull/29141)), requested head [a5c3df8b21](https://git.cafe/pingdotgg/t3code/commit/a5c3df8b2124ce221e4162332eaffa6090779812) against [30654b3e91](https://git.cafe/pingdotgg/t3code/commit/30654b3e91a51937d1dba18fc83af19d8003f5ca). The36-line diff is9 production lines plus27 test lines. Every delivered image was opened:64PNGs including all14 masks and32 sheets. All20MP4s were inspected through independently extracted milestone, intervening and final frames, including64 additional frames at exact narrated action timestamps. Both manifests'91/116 entries match downloaded artifacts by SHA256 (duplicate filenames resolved by content hash). All14 masks were independently recomputed and match exactly: Diff0=max-channel>0; DiffT16=max-channel>16/255; magenta highlight over30%-before/70%-white. Structural correctness did not substitute for visual inspection.

P01/P02 table headers and sheet labels refer to the requested base/head. Export dimensions match the disclosed raster sizes; 1440×900 stills were downscaled to1280×800, and narrow stills are780×1688 for390×844 CSS. iOS402×874/1206×2622 is disclosed and consistent. Android logical viewport/physical DPR were not measured in the delivery. I did not independently establish reduced-motion/network-idle controls or the delivered two-still settling assertion; I did establish two identical final replay PNGs. The “What else changed” account is consistent with the disclosed latency/discovery count and generated IDs, but omits the unequal terminal hover and stale wide-dark iframe and understates the clean P08 interval. The old PR-body preview-failure/reply claims refer to a different earlier run and should be replaced or clearly separated from the proposed new Evidence section.

Cleanup: both dev sessions I started were stopped through their tracked exec sessions;6290/14330 have no listeners. My preview tab was closed. No devices/emulators/leases or worktrees were created. No source edits, pushes or comments were made. Git status is clean; disk has18GB free. Only my scratch subdirectory will be removed after uploading this report and the sanitized audit bundle. The lead's checkout, scratch root and gitignored dev state are retained. Raw authentication-bearing server log and ephemeral auth/database state are excluded from the public bundle; its sanitized server log and synthetic ACP/MCP traces are included.

FRICTION:
- tool | a154MB audit archive upload failed with a transport error; a90MB archive excludes already-public source downloads | upload only new audit output or split large archives.
- tool | Python urllib received403 on the public artifact host while curl succeeded | use curl or document the accepted downloader user agent.
- tool | shared preview saves downscaled PNGs while videos retain DPR2, so the replay comparison needs resampling and includes codec differences | provide a lossless full-raster reference PNG at the replay milestone.
- runbook | uploaded seed harness uses different title/prompt/provider metadata from the accepted P03 recording and leaves server duration uncontrolled | upload the exact invoked seed/configuration and capture parameters.
- brief | pass-2 generic pair wording conflicted with pass-1 head-only minimums | user clarified matrix minimums govern, with comparative base gaps explicitly reported.

VERDICT: INSUFFICIENT <P01, P02, P03, P04, P07>