nx_swarm_tend.nx

nx_swarm_tend.nx -- ACTIVE QUEUE TENDER (operator 2026-07-16: "whoever in the nishi RACI is part of the queue -- use it WELL, not fire-and-forget letting things block and get traffic jams; SOTA"). THE GAP: nx_swarm_queue reaps dead holders PASSIVELY (only when a waiter polls sq_try_grant). If a holder dies with NO waiters polling, its slot LEAKS forever; starvation + jams are invisible. My own 1.5B job queued 45min and timed out unnoticed -- the fire-and-forget failure. THE FIX: an active tender the CONDUCTOR (RACI schedule_resources|R = Responsible-for-draining) runs on a loop / on demand: * REAP: proactively DONE any HOLD whose pid is DEAD -> slots free even with no waiter polling. * DETECT: a waiter starved past a jam threshold = JAMMED (the traffic-jam alarm), the oldest wait-age reported so the PM sees pressure building BEFORE a timeout. * VERDICT: OK / DRAINING (waiters progressing) / JAMMED (starvation) -- health, not silence. tend reap + scan + report + verdict-exit (0 OK / 1 DRAINING / 2 JAMMED) Sovereign store (knowledge/store/swarm_sched) is the SSOT; PURE tend_verdict is gate-testable. license_tier: ORIGINAL