nx_edge_watchdog.nx

nx_edge_watchdog.nx -- OFF-BOX LIVENESS WATCHDOG for the family's public surfaces (F620). BORN FROM THE 2026-07-20 NAS OUTAGE: every monitor we had (vqoe gate, cron-watch, deploy-ready, reconciler, supervisor, self-heal) runs ON THE NAS -- so when the NAS died, the monitoring died WITH it and NOTHING alerted; the outage was noticed only because a human's tool calls started timing out. ★LAW: a monitor that shares a failure domain with its subject is not a monitor. This organ therefore runs on the LAPTOP (or any host that is not the NAS) and probes the PUBLIC edge exactly as a family member's browser would -- WAN-side, no LAN shortcuts, no NAS-side dependency. CONTRACT (data-driven, rule 11): targets live in a conf, one per line "<label> <https-url> <needle>". A target is UP iff the fetch returns 200 AND the body contains <needle> (content proof, not port-open -- the "wedged but alive" class the recovery doctrine calls out). State is a tiny append-only log plus a LATCHED alert file that a human/automation can watch; consecutive failures must reach `fail_min` before alerting, so one flaky fetch never cries wolf, and recovery CLEARS the latch with an explicit RECOVERED line (an alert that cannot clear itself is noise). usage: nx_edge_watchdog [conf] [statelog] [alertfile] defaults: knowledge/registry/edge_watch.conf · knowledge/status/edge_watch.log · knowledge/status/EDGE_ALERT exit: 0 all UP · 1 one or more DOWN (so a scheduler/pulse can branch on it) license_tier: ORIGINAL No hw writes (Rule 26).