PR #22529 real-environment verification, 2026-11-01 UTC Baseline: github/main 575579d68ac4bae09f8426a345d4a3eac1e90c67 Separate /home/user/main worktree; vp i; background vp run dev. Real Chrome + repository Playwright, fresh app-created project/threads, Codex gpt-6.1-sol through local proxy. Unmodified focused test: 1 passed. Temporary existing-test observation gate waited for queued run's persisted run.updated(status=starting) after worker.drain(). Original assertion then failed: AssertionError: expected [ 'sent' ] to deeply equal [ 'queued', 'sent' ] Observation-only edit removed. Real app queue-stop-send flow already worked on main. Exact PR head: 1e1b6c06df03011b8956048ba00c138ef9c49e88 vp i; background vp run dev --home-dir /home/user/t3code/.t3 (no seeded data). blacksmith-check: vp test run apps/server/src/orchestration-v2/SteeringCompletion.integration.test.ts Test Files 1 passed Tests 23 passed Real server event order for pending-stop case (UTC): 00:52:01.077 run 2 created queued, queuePosition=1 00:52:06.568 interrupt requested; run 2 queued, queueHeld=true 00:52:06.596 run 2 queued, queueHeld=false 00:52:06.597 run 3 created queued, queuePosition=2 00:52:06.649 run 1 interrupted 00:52:06.882 run 2 starting 00:52:11.543 Codex: QUEUED_COMPLETE 00:52:11.735 run 2 completed 00:52:12.113 run 3 starting 00:52:15.163 Codex: SENT_COMPLETE 00:52:15.201 run 3 completed UI created and queued the messages and clicked Stop. To hit the short pending-interrupt window, the plain send was sent immediately after Stop over the actual web client's authenticated WebSocket with the same message.dispatch payload. No provider delays, mocks, fixtures, or synthetic provider events. Nearby real-provider checks: - Ordinary UI steering: message entered active run 1 at 00:52:42.777; Codex returned STEER_COMPLETE at 00:53:07.981; no extra run created. - Stop without send: queueHeld=true at 00:53:34.616; remained queued after active turn interrupted. UI Send now released hold at 00:53:44.369; Codex returned HELD_COMPLETE at 00:53:47.146. No source fixes or pushes. CI snapshot: 32/32 successful. Not verified: native desktop/mobile, other providers, remote/relay/tunnel modes. This PR changes test synchronization only; the UI has no visual behavior delta.