PR #28282 real development verification, 2026-11-01 UTC
main: 955e2bfbb5c092c54ad795c8d379fec99702a6b5
PR:   85fc128b58df6271fcab30186ea98c9bf5dca365 (unchanged)

Installed each checkout with timeout 600 vp i.
Started one background dev server at a time with isolated empty state and --no-seed.
main ports: server 16733 / web 8693. Stopped captured process group 15581 before PR startup.
PR ports: server 13773 / web 5733.
Browser: repo Playwright 1.60.0, /usr/bin/google-chrome, 1440x1000.
Created a project at /home/user/main through Add project > Local folder.
Selected Codex gpt-6.1-sol in the composer and submitted identical first prompts.
No mocks, fixtures, replay provider or demo data.

Counted SELECTs on orchestration_v2_projection_messages beneath each
OrchestratorMcpService.loadCaller span, following parentSpanId links in existing
server.trace.ndjson. No instrumentation or source modifications.
                                  main  PR
orchestrator_capabilities             1   0
t3_thread_list                       1   0
list_scheduled_tasks                 1   0
t3_thread_read caller lookup         0   0
All four tools succeeded on both heads. Both turns confirmed the exact marker
'audit-m016-real-run' in the original untruncated user prompt.

PR second real Codex turn, with existing prompt history:
capabilities succeeded; scheduleTask created disabled current-thread-bound task;
list confirmed disabled state and weekdays 09:00 schedule; update renamed it;
list confirmed new title, still disabled, never run; delete removed exact task;
list returned []; readThread returned both prompts untruncated.
All 12 loadCaller spans in both PR turns succeeded and selected no caller messages.
Read-only SQLite: SELECT count(*) AS remaining_scheduled_tasks FROM scheduled_tasks
=> 0.

Blacksmith: vp i && vp test run apps/server/src/mcp/OrchestratorMcpToolkit.integration.test.ts
=> 1 test file passed, 2 tests passed.
PR CI: 31 successful, 0 failing, 0 pending at inspected head.

No source changes, commits or pushes.
No latency/speedup claim. Not verified: native desktop/mobile, other providers,
remote/relay/tunnel, send/wait/interrupt, scheduled execution, large-thread timing.