The burn loop today: cron tick → stale takeover → the same poisoned state crashes → next
tick takes over again — the only backstop is running out of usd. Record a stale-takeover
count; the same Loop taken over N times inside a window refuses the takeover and parks the
Loop as needs-human.
Evidence: hermes production incident (SIGTERM loop every ~10s, human had to intervene) and
its restart_loop_guard.
The burn loop today: cron tick → stale takeover → the same poisoned state crashes → next
tick takes over again — the only backstop is running out of
usd. Record a stale-takeovercount; the same Loop taken over N times inside a window refuses the takeover and parks the
Loop as needs-human.
Evidence: hermes production incident (SIGTERM loop every ~10s, human had to intervene) and
its restart_loop_guard.