2026-11-01, disposable Linux Boat VM
baseline: github/main 575579d68ac4bae09f8426a345d4a3eac1e90c67
PR: ad0943e71af21ab27dd58bd68594cade4bd4d2d7

$ git worktree add ../main github/main
$ (cd ../main && timeout 600 vp i)
$ setsid nohup timeout 7000 vp run dev > /tmp/main-dev.log 2>&1 &
[dev-runner] serverPort=16733 webPort=8693 baseDir=/home/user/main/.t3

Playwright from repo dependencies, /usr/bin/google-chrome:
Added /home/user/shell-verification through Add project > Local folder.
Created 34 threads through New thread; submitted real Codex gpt-6.1-sol turns.
Copied this app-created state for the PR, with the baseline server stopped.
No mocks, fixtures, Replay provider, synthetic shell events or database-created threads.

Two signed-in browser clients; selected these same 34 threads > Settle (34) > undo.
Chrome debugger conditional probes observed the real client-runtime functions.
main: 69 thread.updated items in 10 received batches; 69 applyShellStreamEvent calls.
PR: 70 thread.updated items in 10 received batches; 10 applyShellStreamEvents calls,
    0 single-event reducer calls; delta batch sizes 5,3,7,14,7,6,7,7,7,7.
The captures include read/metadata updates in addition to the 68 bulk settle/undo updates.
PR added one empty default server project row; original 35 shell rows match exactly
between baseline and PR after undo, including order, titles and settle/snooze state.
No elapsed-time or large-list speedup claim is made from these debugger probes.

$ git switch --detach ad0943e71af21ab27dd58bd68594cade4bd4d2d7
$ timeout 600 vp i
$ setsid nohup timeout 6000 vp run dev --home-dir /home/user/t3code/.t3 > /tmp/dev.log 2>&1 &
[dev-runner] serverPort=13773 webPort=5733 baseDir=/home/user/t3code/.t3

PR web client > existing thread > real Codex gpt-6.1-sol turn:
  user: Reply with exactly SHELL-CHECK-AFTER. Do not use tools or change files.
  assistant: SHELL-CHECK-AFTER
Single-thread Snooze > In 1 hour: Snoozed shelf count 1; undo: no Snoozed shelf.
Network offline/online: snapshot, snapshot, synchronized; 36 rows and 36 unique ids.
Reconnect used full snapshots on both revisions, not historical delta replay.

$ blacksmith-check run --id [task-owned worker] "vp i && vp test run packages/client-runtime/src/state/shellReducer.test.ts packages/client-runtime/src/state/shell-sync.test.ts"
2 files passed; 30 tests passed.
$ blacksmith-check run --id [same worker] "git fetch --depth=2 https://github.com/saphid/t3code.git t3code/sweep/shell-batch-merge && vp run check:changed 79f1fadfc6020b820d541ca4691d2feebb15dd1b"
pass format, lint, knip (2), typecheck (7 packages), test (2 files, 30 tests).

Not exercised: native mobile/Electron, relay/tunnel, historical replay delta chunks,
other providers, thousands-of-threads rendering or timing.