PR 30381 runtime verification, Linux x86_64, 2026-11-03 UTC
main: 7c22e24e3fa1a3bb3a95f27aea000067f198fe7f
PR:   187f84f9abba078a6fe393d00bcf9e44f8fc4f8d

Separate main worktree; vp i on both; one background dev server at a time.
Compiled each checkout's native monitor with cargo build --locked --release for runtime use.
Playwright 1.60.0 from repo dependencies, /usr/bin/google-chrome.
Empty isolated state. Project /home/user/t3code created through UI. No mocks or fixtures.

Real Codex gpt-6.1-sol turns through web composer on both builds:
  git rev-parse --show-toplevel -> /home/user/t3code
  printf "provider-path-ok\n" -> provider-path-ok

Real terminal sleep 90; monitor CPU measured from /proc/<monitor-pid>/stat:
  main PID 22821: 30010 ms elapsed, 25 ticks = 250 ms CPU, 266 host processes.
  PR   PID 35011: 30031 ms elapsed,  5 ticks =  50 ms CPU, 263 host processes.
Whole-monitor windows, one per build. No tracing during these measurements.
No universal speedup claimed. Main includes a separate idle-polling optimization;
both windows kept a child running to exercise processTable at the active cadence.

Separate 12-second syscall traces of the monitors spawned by the real servers:
  main: 11 processTable replies, 2908 stat opens, 2908 status opens.
  PR:   11 processTable replies, 2969 stat opens,    0 status opens.

Terminal activity delivered through the actual browser's WebSocket:
  main 23:09:59.314Z hasRunningSubprocess=true, label=sleep
  main 23:11:00.511Z hasRunningSubprocess=false, label=Terminal 1 (after ctrl+c)
  PR   23:15:08.080Z hasRunningSubprocess=true, label=sleep
  PR   23:17:59.790Z hasRunningSubprocess=true, label=a) b (c)
  PR   23:18:07.837Z hasRunningSubprocess=false, label=Terminal 1 (natural exit)

Nearby behavior: ctrl+c clears running child, printf still returns output.
Running-child close warning says sleep; Cancel preserves terminal.
Close idle terminal, confirm, reopen, printf -> reopened-terminal-ok.

Blacksmith: vp run test:resource-monitor (cargo test --locked): 20 passed.
Blacksmith: vp run check:changed e5d07108760777a8b366c727d01c582cddf27dbc: passed.
Worker needed a depth=3 fetch of PR history to resolve its comparison base.
GitHub checks: all reported runnable checks passed; native/mobile/web deployment jobs skipped where not applicable.

Not verified: native Electron/iOS/Android clients; macOS/Windows/Android collector paths;
remote/tunnel clients; the original 480-process benchmark magnitude.