## Section 1 — Liveness (both sides, always-on)

**I-1. Every process must prove liveness or alert.**
Any automated process (watcher, watchdog, comms reader, cron) that stops — for any reason —
must emit an alarm to its operator within 1 heartbeat cycle.
- PBS implementation: launchd liveness-reaper + heartbeat.py per service.
- Czak: per Czak principle "Push, not pull" — operator gets pinged on process death.
- *This rule exists because:* Czak's comms reader died silently on 2026-07-23. PBS watchdog
  fired without awareness of HITL reversal. Both caused today's dispute.

**I-2. A silent process is a dead process.**
No process may be assumed alive because it was started. Liveness = last heartbeat < 2× interval.
Any component checking a peer's status must also verify its own watcher is alive before acting.

**I-3. Watcher restart is automatic, not manual.**
If a watcher dies: auto-restart (launchd `KeepAlive`, systemd, or equivalent). If restart fails
3× in 30 min: alert operator and halt further automated actions that depend on that watcher.

---