Visual reproduction bundle

manual.py prepare <checkout> <fresh-run-directory> creates the synthetic repository and fresh T3 settings. Run forge.py <fresh-run-directory> in a separate terminal. Start that checkout with T3CODE_DEV_SEED=0, T3CODE_AUTO_BOOTSTRAP_PROJECT_FROM_CWD=0, GH_CONFIG_DIR=<fresh-run-directory>/gh, GH_TOKEN=scenario-fixture-token, and node scripts/dev-runner.ts dev --home-dir <fresh-run-directory>/home --no-seed. After it is ready, run manual.py seed <checkout> <fresh-run-directory>. Pair the browser normally.

Use the same dark 1280x800 browser session throughout each run. Open the synthetic pull request scenario/review-guide number 1. Select Guide, Write guide, and wait until its excerpt paints PREVIOUS_PUSH. Run manual.py push <checkout> <fresh-run-directory>. In the PR More menu select Refresh. Select Rewrite, wait for completion, then open Code. In the PR More menu select Refresh again. Expand all files and confirm CURRENT_PUSH. Return to Guide. Base retains PREVIOUS_PUSH; head displays CURRENT_PUSH. Take each still after two byte-identical screenshots at least 500ms apart.

The GitHub API and guide-generation model are deterministic local subprocess fixtures. Git commits, git fetch, T3 server persistence, guide generation service, client cache and diff rendering are real. No live GitHub repository was changed. The token literal is synthetic, not a credential.

rewritten-guide-new-push.e2e.ts is the attempted automated scenario. Its base run completed and reproduced the bug; its head run stopped because a text locator matched the hidden Guide copy while Code was visible. Final evidence comes from the manual browser procedure above, independently repeated on both commits.
