## Evidence
Captured **3 of 3 requested rows**, at the requested revisions, for [#32840](https://git.cafe/pingdotgg/t3code/pull/32840) ([GitHub mirror](https://github.com/pingdotgg/t3code/pull/32840)). Before: [fa2f4ff2a2](https://github.com/pingdotgg/t3code/commit/fa2f4ff2a2bac5ac92e07749f6b83b0003d642a0). After: [d6e7212c10](https://github.com/pingdotgg/t3code/commit/d6e7212c10d0af56a48b54ad7ed5b8bd5388c515).
**Rendering environment for every capture:** Simlock iPhone Duo, iOS 27.2, folded closed at 0° using the outer display, portrait, light, Large system font (standard 1.0 scaling), 466×678 pt. Whole-screen stills are 466×678 px at 1×; whole-screen recordings are 1398×2034 px at native 3×, variable frame rate (30 fps requested). No cropping, masking, status-bar substitution, or video trimming. The same leased simulator was used throughout the accepted captures.
Both revisions ran from their own checkouts and their own fresh servers with `T3CODE_DEV_SEED=0`. The same RPC driver seeded Guide and Demo with three stopped synthetic threads: Review guide, Update examples, Check layout. One connected environment, All environments/All projects, list at the top, no active draft, search resting. No production database was read or copied; no agent turns ran. Each checkout's mobile installer installed the matching development client. Both revisions share the native fingerprint; the accepted client was built with the iOS 27.1 SDK.
### C1 — Folded Chats, resting after cold launch · iPhone Duo / light
Each side was terminated and relaunched, then settled on Chats. For each accepted still, the driver took two whole-screen screenshots with a requested 500 ms pause; ImageMagick AE was **0** within each pair. Network command overhead makes the actual gap longer. Accepted stills and both diff images were opened and visually inspected. No overlay, skeleton, spinner, duplicate content, or clipped text is present in either accepted still.
| Before | After | Diff0 | DiffT |
|---|---|---|---|
| 
[fa2f4ff2a2](https://github.com/pingdotgg/t3code/commit/fa2f4ff2a2bac5ac92e07749f6b83b0003d642a0) | 
[d6e7212c10](https://github.com/pingdotgg/t3code/commit/d6e7212c10d0af56a48b54ad7ed5b8bd5388c515) | 
[fa2f4ff2a2](https://github.com/pingdotgg/t3code/commit/fa2f4ff2a2bac5ac92e07749f6b83b0003d642a0) → [d6e7212c10](https://github.com/pingdotgg/t3code/commit/d6e7212c10d0af56a48b54ad7ed5b8bd5388c515) | 
[fa2f4ff2a2](https://github.com/pingdotgg/t3code/commit/fa2f4ff2a2bac5ac92e07749f6b83b0003d642a0) → [d6e7212c10](https://github.com/pingdotgg/t3code/commit/d6e7212c10d0af56a48b54ad7ed5b8bd5388c515) |
Diff0: ImageMagick `compare`, 0% fuzz, every differing pixel amplified to magenta; **799 / 315,948 pixels (0.253%)**. DiffT: identical comparison at **2% fuzz**, **719 pixels (0.228%)**. Dark pixels are unchanged. No alignment, resize, or excluded region. The shared [diff.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5212d5dc19758ee9/diff.py) records the exact commands.
**What else changed (the PR intends no visual change):** the status clock reads **4:33 → 4:52**, and all three elapsed thread ages read **53m → 1h**. Both diffs isolate those four text regions. No other visible still difference was found; toolbar icons, list content, geometry, colors, and scroll position match.
### I1 — New chat → new-task project flow → Back to Chats · iPhone Duo / light
Same scripted steps: start on Chats; request New chat at **2.000 s** on both sides; wait for Choose project; hold; request native Back at **9.905 s before / 9.516 s after**; wait for Review guide and hold on Chats. These are driver request timestamps relative to record-start acknowledgment; remote input arrives later. In the videos the project flow appears around **4 s**, and Chats returns around **14 s before / 13.5 s after**. The native return transition briefly reveals New thread before Chats on both sides. No task was submitted. Durations: **15.130 s before / 14.648 s after**.
| Before | After | Diff0 | DiffT |
|---|---|---|---|
| [I1-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/631aa734d8c4e3e3/I1-before.mp4)
[fa2f4ff2a2](https://github.com/pingdotgg/t3code/commit/fa2f4ff2a2bac5ac92e07749f6b83b0003d642a0) | [I1-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c45624612a86d188/I1-after.mp4)
[d6e7212c10](https://github.com/pingdotgg/t3code/commit/d6e7212c10d0af56a48b54ad7ed5b8bd5388c515) | N/A — recording
[fa2f4ff2a2](https://github.com/pingdotgg/t3code/commit/fa2f4ff2a2bac5ac92e07749f6b83b0003d642a0) → [d6e7212c10](https://github.com/pingdotgg/t3code/commit/d6e7212c10d0af56a48b54ad7ed5b8bd5388c515) | N/A — recording
[fa2f4ff2a2](https://github.com/pingdotgg/t3code/commit/fa2f4ff2a2bac5ac92e07749f6b83b0003d642a0) → [d6e7212c10](https://github.com/pingdotgg/t3code/commit/d6e7212c10d0af56a48b54ad7ed5b8bd5388c515) |
### I2 — Chats → PRs → Chats · iPhone Duo / light
Same scripted steps: start on Chats; request PRs at **2.000 s before / 2.001 s after**; hold on the empty PR list; request Chats at **8.137 s before / 7.971 s after**; verify the same three threads and restored Compose/Search. Timestamps use the same driver clock as I1 (before I2 reconstructed from logged command completion minus command duration). Visible transitions occur around **4 s to PRs / 10 s back to Chats** on both sides. Durations: **173.912 s before / 11.562 s after**. The before export failed during a pool outage and was recovered later, leaving a long, unedited resting-Chats tail; the interaction steps are the same.
| Before | After | Diff0 | DiffT |
|---|---|---|---|
| [I2-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b5607f0274d8d0ee/I2-before.mp4)
[fa2f4ff2a2](https://github.com/pingdotgg/t3code/commit/fa2f4ff2a2bac5ac92e07749f6b83b0003d642a0) | [I2-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c653e599870ac685/I2-after.mp4)
[d6e7212c10](https://github.com/pingdotgg/t3code/commit/d6e7212c10d0af56a48b54ad7ed5b8bd5388c515) | N/A — recording
[fa2f4ff2a2](https://github.com/pingdotgg/t3code/commit/fa2f4ff2a2bac5ac92e07749f6b83b0003d642a0) → [d6e7212c10](https://github.com/pingdotgg/t3code/commit/d6e7212c10d0af56a48b54ad7ed5b8bd5388c515) | N/A — recording
[fa2f4ff2a2](https://github.com/pingdotgg/t3code/commit/fa2f4ff2a2bac5ac92e07749f6b83b0003d642a0) → [d6e7212c10](https://github.com/pingdotgg/t3code/commit/d6e7212c10d0af56a48b54ad7ed5b8bd5388c515) |
The recordings were opened through decoded contact sheets and transition/end frames. In addition to the C1 clock/age differences, the recordings have slightly different remote-input and native-transition timings; before I2 also contains the idle tail described above. No persistent visual difference was observed at the matched endpoints. Transient native navigation animation and PR loading are visible in motion.
### Reproduction artifacts
The same seed, pairing, capture, and diff drivers were used for both revisions, with timestamped run directories: [seed.mjs](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5f90f0acfa3c15cb/seed.mjs), [pair-client.mjs](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6ea8ce2d164fdc73/pair-client.mjs), [capture.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9bb40c23226ab2b7/capture.py), [diff.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5212d5dc19758ee9/diff.py). [capture-proof.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c4ad09ae392c1642/capture-proof.json) includes file hashes, settled-pair hashes and timestamps, dimensions, driver steps, and ImageMagick metrics. All public artifacts contain synthetic data only.
**Unverified requested rows: none.** X1 was explicitly outside this assignment and was not investigated or reproduced. This evidence does not establish the cause of the reported startup failure. No PR checks or reviews were assessed in this capture task.
AI-assisted; verified with the tests and tools listed above.