PR https://git.cafe/pingdotgg/t3code/pulls/19288
Host: Linux x86_64, Node v24.22.1, /usr/bin/google-chrome, repo Playwright 1.60.0.
Baseline: github/main 4914cdd98f, /home/user/main, vp i.
PR: 6a787619a26ea414fb27fbb7fb65b499c8235684, /home/user/t3code, vp i.
Both servers started in background with setsid nohup vp run dev; PR uses explicit --home-dir /home/user/t3code/.t3.
Main ports: web 8693, server 16733. PR ports: web 5733, server 13773.
Created git-console-verification projects through Add project > New project.
Selected Codex > GPT-6.1-Sol in the actual composer and submitted real turns.
Main: initial file creation, git status --short => ?? greeting.txt.
Main: local commit 8e52e94, tracked file edit.
PR: local commit 665d437, tracked file edit.
Both: git status --short => M greeting.txt.
Both: git diff => index ce01362..3b18e51; -hello / +hello world.
Both clients: branch main; Changes +1 -1.
PR client: opened Changes, expanded greeting.txt, refreshed diff, switched split/stacked layout. Diff retained -hello / +hello world.
No remote was contacted by the provider turns.
These are Linux regression observations, not proof of the Windows optimization.
Windows conhost.exe counts, native spawn flags, cancellation/crash cleanup and latency cannot be reproduced on this VM.
CI: 6 success status contexts; GitHub head workflow runs conclude action_required. Full CI is not established.
No source edits, commits, pushes, label changes, merge, close, rebase or force-push.