# Evidence pass 2 — [#27798](https://git.cafe/pingdotgg/t3code/pull/27798)

**5/6 requested numbered rows captured (6/7 surface pairs); M12 unverified.**

Before [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251); after [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a). The before SHA is the actual output of `git fetch -q origin main && git merge-base cffa192a07 origin/main`: `d8a829e5ecd6626f8fdee23ef20ca9758e056251`. It differs from the abbreviated hint in the brief.

Fresh isolated per-SHA databases, T3CODE_DEV_SEED=0. Public microsoft/TypeScript reads use each server's own credential. Synthetic stopped thread; no provider run, live-database copy, or real PR mutation. Timing gates preserve actual payloads. This section supersedes only P2/L11, P3 Summary/Timeline/Code, P4, P6, M11 and M12; historical P1/P5/P3 Checks are indexed separately.

All accepted static frames are full surfaces, with pointer cleared, fonts ready and network settled after releasing gates. Each accepted frame must match a second frame at least500ms later. Diff0 maps every nonzero RGB-channel difference to255; DiffT maps values >8/255 to255. Differences refer to settled frames; recordings demonstrate loading. Rejected native frames are not presented as valid evidence.

## P3 — web1440×900 dark, Summary/Timeline/Code reload

Real Vite app, shared Preview Chrome156/Linux, no device emulation; viewport1440×900 CSS, DPR2, font100%, top scroll, whole-surface tool-scaled1280×800PNG. App Date frozen2026-11-03T00:00:00Z. Preview has no reduced-motion emulation.

t=0 reload with PR open. Both WebSocket PR answers and HTTP `/api/pull-requests/diff` answers are gated equally on both sides. Page-performance seconds before/after: inspect Summary57.182/57.075; Timeline69.086/69.084; Code73.354/73.432; expand code86.016/83.889. After's HTTP diff arrived71.822 and remained held until118.847; saved added code was visible before release. Before remained loading and Code was disabled, so it did not issue an HTTP read while its detail was held. The HTTP gate is installed on both sides. Return Summary after release and clear pointer before final stills.

Before recording: [P3-web-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c80a475c41bcff8b/P3-web-before.mp4) · After recording: [P3-web-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/af2e0b70293f3b78/P3-web-after.mp4)

Traces: [Before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a18d18cf74ffeaea/P3-web-before-trace.json) · [After](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5aae69787942dcb3/P3-web-after-trace.json)

| Before | After | Diff0 | DiffT >8/255 |
| --- | --- | --- | --- |
| [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) | [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) |
| ![P3-web before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/696f628d60184550/P3-web-before.png) | ![P3-web after](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f1884cd7245f7d14/P3-web-after.png) | ![P3-web diff0](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/58098cfea6e9daef/P3-web-diff0.png) | ![P3-web diff8](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8413c56f67bbf987/P3-web-diff8.png) |

![P3-web-sheet.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/79f78174e529967b/P3-web-sheet.png)

[P3-web-pixels.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/1e9fda653ce2a97e/P3-web-pixels.json)

What else changed:85 nonzero pixels, maximum6/255, on the small approval/review icon; no pixels >8/255. No layout, content, tooltip or surrounding-control difference in the settled pair.

## P3 — Electron1440×900 light, Summary/Timeline/Code reload

Real bundled Electron44.10.0 / Chromium152.0.7977.169, Linux headless-desktop/Xvfb, native1440×900, DPR1, font100%, light, top scroll, prefers-reduced-motion:reduce. ImageMagick import whole native window; ffmpeg x11grab15fps.

t=0 reload; page-performance seconds before/after: Summary2.961/4.149, Timeline5.241/6.692, Code7.508/9.109, expanded9.767/11.534. The same WS+HTTP gate is active. After's HTTP diff arrives7.136, stays held until11.912, and expanded saved code is already visible. Before is loading with Code disabled. Release, return Summary, move pointer away and dismiss tooltip before stills.

Before recording: [P3-desktop-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/aa98671d45841ab5/P3-desktop-before.mp4) · After recording: [P3-desktop-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/815b02909bad0e9e/P3-desktop-after.mp4)

Traces: [Before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b92d94356657ed81/P3-desktop-before-trace.json) · [After](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b4739cc328369b17/P3-desktop-after-trace.json)

| Before | After | Diff0 | DiffT >8/255 |
| --- | --- | --- | --- |
| [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) | [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) |
| ![P3-desktop before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/cb58a202b9d42065/P3-desktop-before.png) | ![P3-desktop after](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/52a0e7dae782e400/P3-desktop-after.png) | ![P3-desktop diff0](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/66d1ea28755c5d1c/P3-desktop-diff0.png) | ![P3-desktop diff8](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/790cb09f030980d3/P3-desktop-diff8.png) |

![P3-desktop-sheet.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/283d8d170cde9da6/P3-desktop-sheet.png)

[P3-desktop-pixels.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/aa44e92d07cd4c50/P3-desktop-pixels.json)

What else changed:10 nonzero pixels, maximum2/255, sparse glyph/edge antialiasing; no pixels >8/255. Recording includes a transient copy tooltip during interaction; final stills do not.

## P2 + L11 — web1440×900 dark, dwell cost and real pointer pass

Real Vite app, Playwright-driven Chromium148/Linux; trusted physical browser pointer moves, no synthetic events, no device emulation; viewport1440×900 CSS, DPR2, font100%, dark, reduced-motion:reduce. Whole2880×1800PNG; recording1440×900. App Date frozen2026-11-03T00:00:00Z.

Fresh browser contexts and cold rows; same author filter and newest ordering. Page-performance seconds are reported here, independently of the media boot lead-in. Before: quick pass19.316–19.903 across three rows, per-row86.1/103.0/79.6ms; unused cold-row dwell19.945–20.570 (625ms), then leave without click; clicked-row dwell23.679, click26.721. Detail starts26.732, after the click. After: quick pass17.927–18.504, per-row85.0/68.9/102.3ms; unused dwell18.545–19.161 (616ms), detail starts18.710; leave without click. Clicked-row dwell22.245, prefetch22.400, click25.286. All pointer events are trusted. Quick pass sends no detail read on either side. Unused dwell costs exactly one core detail RPC on after, zero on before. Associated workflow/review reads are separate secondary calls; this is not a claim of only one total HTTP request. The after panel has summary immediately at click; before paints the gray ghost until its request returns.

Quick rows: [#65122](https://github.com/microsoft/TypeScript/pull/65122), [#65121](https://github.com/microsoft/TypeScript/pull/65121), [#65073](https://github.com/microsoft/TypeScript/pull/65073). Unused [#65064](https://github.com/microsoft/TypeScript/pull/65064); clicked [#65044](https://github.com/microsoft/TypeScript/pull/65044).

Before recording: [P2-physical-before.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ed81674aba6b2f97/P2-physical-before.webm) · After recording: [P2-physical-after.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2ced79ea23fb1de1/P2-physical-after.webm)

RPC/pointer traces: [Before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ef5c394f90acf9ba/P2-physical-before-trace.json) · [After](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0fde709a071e3e60/P2-physical-after-trace.json). Server HTTP/GraphQL spans: [P2-physical-http-traces.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/69467198d42ec90a/P2-physical-http-traces.json).

| Before | After | Diff0 | DiffT >8/255 |
| --- | --- | --- | --- |
| [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) | [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) |
| ![P2-physical before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8969dbf27d0219f1/P2-physical-before.png) | ![P2-physical after](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6194a86cbd09c24d/P2-physical-after.png) | ![P2-physical diff0](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6b964080bb04245a/P2-physical-diff0.png) | ![P2-physical diff8](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2c360c9f850b8093/P2-physical-diff8.png) |

![P2-physical-sheet.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/296632711f35f9fb/P2-physical-sheet.png)

[P2-physical-pixels.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8e81e03452f61428/P2-physical-pixels.json)

What else changed:zero RGB pixels in the settled pair. Nothing else changed in these static frames. Startup/network durations differ in recordings; action and request clocks are in the traces.

## P4 — web1440×900 light, thread PR card reload

Real Vite app, shared Preview Chrome156/Linux, no device emulation; viewport1440×900 CSS, DPR2, font100%, top scroll, whole-surface tool-scaled1280×800PNG. App Date frozen2026-11-03T00:00:00Z. Preview has no reduced-motion emulation.

t=0 reload synthetic stopped thread. Only PR replies are held. The after saved full PR card appears around9s in media; before retains its fallback badge until release around43s. Gate markers in page-performance clock: before44.674, after48.935. Release PR replies and take settled stills. Git controls are allowed to finish their own reads equally; the shared project fixture is clean main at [042e0fb957](https://github.com/microsoft/TypeScript/commit/042e0fb9574980bd67964165df295e629f834fe3), origin/main at the same revision, ahead0/behind0, no working changes. Thus neither side has a Pull/Changes control caused by checkout drift. `fixture.sh` pins and asserts that exact state; it is in the script archive.

Before recording: [P4-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bca81b71116d18b6/P4-before.mp4) · After recording: [P4-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/36fc7b1fc6fa4365/P4-after.mp4)

Traces: [Before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/56a64d95f54f4661/P4-before-trace.json) · [After](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/298209004e76e44e/P4-after-trace.json)

| Before | After | Diff0 | DiffT >8/255 |
| --- | --- | --- | --- |
| [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) | [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) |
| ![P4 before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/37d3ce8f7a8aa632/P4-before.png) | ![P4 after](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/452f0a6df8f4297f/P4-after.png) | ![P4 diff0](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9667d03364278676/P4-diff0.png) | ![P4 diff8](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3eddc1e568e24657/P4-diff8.png) |

![P4-sheet.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9bf4a2e6289eeba9/P4-sheet.png)

[P4-pixels.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/88e76aa0648fce36/P4-pixels.json)

What else changed:zero RGB pixels in settled frames; nothing else changed in those frames. During the controlled reload the PR card is the visible feature difference. Startup timings differ.

## P6 — iPhone17Pro402×874 dark, thread-card navigation

Real iOS27.2 iPhone17Pro simulator dev client, 402×874points, DPR3, full1206×2622PNG, dark, Dynamic Type medium, Final stills have Reduce Motion and Reduce Transparency enabled; recordings precede verified switch changes and use default native motion/transparency. Same native fingerprint887f84d7e8ee2e1063658bf884c9d17f60d6cbb3; independent Metro/backend at each SHA. App Date frozen; detail RPC replies delayed6000ms equally. Pool exposes no arbitrary simctl/status_bar command: persistent OS9:41 override unavailable. No crop.

Both runs open the same thread branch sheet and tap its linked [#65119](https://github.com/microsoft/TypeScript/pull/65119) card. The source tap is at driver/media-relative5.330s before and0.699s after; published clips trim4.700s/0.100s from their starts, respectively, retaining full surface and the navigation. Approximate first-second tap, then before full-screen spinner for the6000ms detail gate versus after known title/state/author/branches immediately, with secondary sections loading. At about7–8s both show fetched detail. Edit metadata provides exact source offsets. Native recording clock is6:39/6:40 before and6:30 after; the clock difference is outside PR scope and not a claimed effect. A late temporary status-bar normalization was excluded from the published clips.

Before recording: [P6-final-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/97c52aa917028dcd/P6-final-before.mp4) · After recording: [P6-final-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9900324df7fdf9ac/P6-final-after.mp4)

[Recording edit metadata](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/46e7ed90ac8e038a/P6-final-edit.json). Contacts: [Before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f5b83c45fafefd3e/P6-final-before-contact.png) · [After](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/616045a99173e67c/P6-final-after-contact.png).

| Before | After | Diff0 | DiffT >8/255 |
| --- | --- | --- | --- |
| [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) | [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) |
| ![P6-accessible before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/42ff4da37593b896/P6-accessible-before.png) | ![P6-accessible after](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/93b48ff198ee94d4/P6-accessible-after.png) | ![P6-accessible diff0](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9d9babebe418ad9c/P6-accessible-diff0.png) | ![P6-accessible diff8](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9bf7b033798d8e61/P6-accessible-diff8.png) |

![P6-accessible-sheet.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6ddef559fda619ef/P6-accessible-sheet.png)

[P6-accessible-pixels.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b1b40a04906709c6/P6-accessible-pixels.json)

Final stills use verified Reduce Motion/Reduce Transparency and matching natural OS minute7:28. Warm re-entry after detail has loaded makes native chrome comparable; both images are full surface. What else changed:25 pixels, maximum1/255, sparse native button-edge antialiasing; zero pixels >8/255. Settings observations: [native-settings-observations.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/295fd3b345c0958c/native-settings-observations.json).

## M11 — iPhone17Pro402×874 dark, HARNESS stack navigation

Real iOS27.2 iPhone17Pro simulator dev client, 402×874points, DPR3, full1206×2622PNG, dark, Dynamic Type medium, Final stills have Reduce Motion and Reduce Transparency enabled; recordings precede verified switch changes and use default native motion/transparency. Same native fingerprint887f84d7e8ee2e1063658bf884c9d17f60d6cbb3; independent Metro/backend at each SHA. App Date frozen; detail RPC replies delayed6000ms equally. Pool exposes no arbitrary simctl/status_bar command: persistent OS9:41 override unavailable. No crop.

HARNESS: `seed-stack.js` injects a schema-valid native stack through `thread.pull-request-link.sync` in each isolated server, with actual public summaries for [#65119](https://github.com/microsoft/TypeScript/pull/65119) and [#65017](https://github.com/microsoft/TypeScript/pull/65017). No forge write occurs. Stack replies are held600000ms equally so background host refresh cannot remove the synthetic stack; detail replies retain the6000ms gate. t≈0 begin on stack card, t≈0.8–0.9 tap the other layer. t≈1.8 before shows full-screen spinner; after paints that layer's known Open/title/+75 summary. Fetched detail first appears at t≈9.16 s on the after recording (corrected from t≈7–8 by the proof audit, from the same footage); until then after shows the seeded summary with secondary placeholders. Unknown author and secondary placeholders in after's initial seed are expected from the deliberately minimal schema-valid snapshot. OS recording clocks6:48 before and6:57 after differ; no chrome comparison claim.

Before recording: [M11-final-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bb4e80aaf5aadfb7/M11-final-before.mp4) · After recording: [M11-final-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f5da1f87ecfd51bb/M11-final-after.mp4)

Steps: [Before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a47ef75a85fc6df6/M11-final-before-steps.json) · [After](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/69938ce87d99e588/M11-final-after-steps.json). Contacts: [Before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/be1ab4f471b7be52/M11-final-before-contact.png) · [After](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f2a6425e45d8ded5/M11-final-after-contact.png).

| Before | After | Diff0 | DiffT >8/255 |
| --- | --- | --- | --- |
| [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) | [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) | [d8a829e5ec](https://git.cafe/pingdotgg/t3code/commit/d8a829e5ecd6626f8fdee23ef20ca9758e056251) → [cffa192a07](https://git.cafe/pingdotgg/t3code/commit/cffa192a07c19b904aea4c22a5f16276159cb56a) |
| ![M11-accessible before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d0134e77876e73a4/M11-accessible-before.png) | ![M11-accessible after](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4f4b3c89db00467d/M11-accessible-after.png) | ![M11-accessible diff0](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e51d5a826ce2f6a1/M11-accessible-diff0.png) | ![M11-accessible diff8](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6c1b0a7a1c452f7d/M11-accessible-diff8.png) |

![M11-accessible-sheet.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/757869543d5dc13d/M11-accessible-sheet.png)

[M11-accessible-pixels.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6065996ecee87d8b/M11-accessible-pixels.json)

Settled target after the HARNESS stack is removed: same actual public detail, full surface, verified accessibility settings, matching natural OS minute7:30. What else changed:24 native edge pixels, maximum1/255; zero pixels >8/255. The stack-specific navigation proof is the recording above; this static pair checks the settled destination.

## M12 — iPad Pro13-inch split columns, remembered PR tab

**UNVERIFIED.** Real iPadPro13-inch(M5), iOS27.2, split columns,1032×1376points, DPR2, font100%/Dynamic Type medium, light, whole2064×2752 recording; no emulation or crop. Both idle-expiry attempts lost their pool lease before capture. Restart discarded the remembered selection. Warm/pending-read tab-return attempts did not reproduce a before spinner: both sides showed detail. Do not treat the following diagnostic pair as proof of the requested change. OS clocks7:33/7:34 differ; movie PTS also compressed the elapsed command interval on this simulator, so use the UTC action timestamps only. No accepted M12 static pair/diffs/sheet.

Before recording: [M12-relaunch20-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f88e622d1cc9aa6e/M12-relaunch20-before.mp4) · After recording: [M12-relaunch20-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/af2118c43671a740/M12-relaunch20-after.mp4)

Diagnostic steps: [Before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b5a36bd9242e7393/M12-relaunch20-before-steps.json) · [After](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/211b60c07e29700c/M12-relaunch20-after-steps.json). Contacts: [Before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ca3081807d2f65b4/M12-relaunch20-before-contact.png) · [After](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/056db66580c5d85d/M12-relaunch20-after-contact.png).

## Scope and reproducibility

[All seed/driver scripts](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/77f1d875c95ecb5a/scripts-pass2.tar.gz) · [Reproduce notes](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0c4eb636799d813d/README-pass2.md) · [Pinned git fixture](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8dc85988016aff10/fixture.sh) · [Friction log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/035cb23f70ffbf2c/FRICTION.md).

Remaining limits: M12 lacks the requested proof. P6/M11 recordings use default native motion/transparency and different OS clock minutes; their final stills have verified accessibility settings and matching minutes. The pool does not expose a persistent OS-clock override. Synthetic M11 validates known-summary stack navigation, not a host integration. No stale overlays, skeletons or spinner are accepted as settled stills.

FRICTION:
- tool | Preview hover/drag could not reliably make a trusted pointer pass under150ms per row; two attempts failed that condition. Used installed Playwright Chromium for that row. | Precise real pointer move support in Preview.
- runbook | Pool lease belonged to repo-root cwd; mobile install from detached checkout returned DEVICE_IN_USE. Exact checkout native-client script run from lease cwd installed successfully. | Document lease cwd ownership beside mobile:install.
- infra | Long-running dev/Metro processes exited; Offline/Expo-error attempts were discarded and own processes restarted. | Durable tracked process sessions with failure reporting.
- tool | iPad lease ended twice before the cold-tab capture, including only five minutes after the last UI action. Renewed after one-minute retry, reinstalled exact client and re-paired; switched to the requested restart path. | Lease-expiry notification and renewal visibility.
- tool | Pool has no arbitrary simctl/status_bar command. Temporary screenshot normalization animated battery chrome. | Expose persistent9:41 OS-clock override through the pool.
- model | Earlier accessibility taps did not toggle the actual switches. Direct AX inspection caught this; final phone stills retaken after verified switch selection. | Role-specific selectors and read-back immediately after changing a setting.
- tool | Spinner-first versus seeded-first native entry left differing Liquid Glass edges even after data settled. Re-entering the already-loaded route made the chrome comparable. | Stable native compositor capture state.


No repository changes were committed, pushed or merged. Historical sufficient rows are retained only at their original SHAs. The linked updated index distinguishes them from this pass.