# Git LFS pull request image reproduction

Environment for both captures: web client in Chromium, 1440 × 1000 CSS pixels, light theme, maximized pull request panel. Each commit ran its own dev server and fresh database with T3CODE_DEV_SEED=0 and --no-seed. The seed script registers a synthetic git project with the public repository as its remote; no agent turns or live database copies are used.

Base: [0cd6a73433](https://github.com/pingdotgg/t3code/commit/0cd6a73433c3b9c1e328560d591fd297e72d7ef9)
Head: [02623a089f](https://github.com/pingdotgg/t3code/commit/02623a089fbd1db46169ab4efc4319713efdd769)

1. Pair with the isolated dev server, open Pull requests and select [#35434](https://github.com/pingdotgg/t3code/pull/35434), using /pull-requests?repository=pingdotgg%2Ft3code&number=35434&state=all.
2. Maximize the pull request panel and click Edit description.
3. Replace the local draft with:

```markdown
LFS: ![lfs](https://github.com/IDEDARY/Bevypunk/blob/HEAD/art/movies/intro_frames/1_background.png?raw=true) plain: ![plain](https://github.com/pingdotgg/t3code/blob/main/apps/web/public/apple-touch-icon.png?raw=true)
```

4. Click Preview, wait for image requests to settle, and capture. Recordings start with the replacement draft already entered, before clicking Preview.
5. Cancel without saving. No forge writes are made.

Base visibly displays “Image unavailable · lfs”; the ordinary PNG decodes at 180 × 180. Head decodes the LFS image at 1920 × 1080 (the image is a black background) and the ordinary PNG at 180 × 180.

For each final still, two captures at least 500 ms apart were byte-identical. SHA256 of both base captures: bf1faf6c93afc23e145c66c041a81f87ce0ed61a00136db67becefe1e76f59b6. SHA256 of both head captures: 83d17eb54ca5293c8a9a46501ce69e496368ffb6b8746fb2192fe950b8786ad6.

The scenario runner was tried twice. Its attempts hit navigation/sign-in timing and a textbox-versus-combobox selector mismatch; the corrected head scenario passed, but no successful paired scenario proof was produced. The final evidence comes from repeating the same steps manually in both real apps through the shared preview browser.

AI-assisted; verified with the tests and tools listed above.
