nx_ready_census.nx

nx_ready_census.nx -- LIVENESS-vs-READINESS CENSUS across the whole supervised fleet (seq1314 oracle). WHY: on 2026-07-29 nx_tools_api_serve was PROCESS-ALIVE with its LISTEN SOCKET GONE; every liveness-only guard read it as healthy and the agent-facing MCP surface stayed dark for an hour. seq1299 fixed the detector and wired it into ONE guard. This organ answers the fleet-wide question that fix raised: "which OTHER services are currently claiming UP while nothing is listening?" METHOD: mgmt_snap.json is the supervisor's OWN per-poll claim (SVC <name> <port> <state> ...). We take each claim and CONNECT to the port -- connect-only, never writing a byte, so this can never disturb a working daemon (the retired /healthz serving-probe crash-looped redirect.elf by writing to it). A DISAGREEMENT (claim=UP, socket=REFUSED) is the exact 07-29 wound, caught by measurement. ★It is ALSO the empirical port oracle for seq1314: the supervisor's guard CALL-SITE COMMENTS are stale (they say the docportal admin daemon is :8456; it is really :18456 -- wiring a guard from the comment would kill a HEALTHY daemon every poll). mgmt_snap.json agrees with the live listen table; comments do not. nx_ready_census [snapfile] exit 0 = all claims honest · 3 = disagreement · 2 = snapshot unreadable license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0