nx_ioadmit_lib.nx

nx_ioadmit_lib.nx -- THE I/O-STORM ADMISSION RULER, importable: ONE ruler for every process that is about to ADD uninterruptible I/O to a shared array (torrent workers, crawlers, bulk writers) -- the same witness nx_build_admit already uses to refuse a compiler fork, so a spawner and the build gate can never disagree about whether the box is in an I/O storm. WHY (measured 2026-08-19): the box sat at load 12-17 on 8 CPUs for days with procs_blocked=8, and 5 of the 8 blocked pids were nx_torrent_get workers -- the torrent daemon resumed EVERY active torrent at once (no budget, no resource awareness), and that alone pushed procs_blocked to ncpu, where the build admission's I/O-storm witness correctly refused every lane's builds. The cure is not a higher ceiling; it is that an I/O spawner asks the same question the build gate asks BEFORE it forks. THE RULER (unchanged from nx_build_admit.ba_verdict, 2026-08-17): procs_blocked >= per_cpu x ncpu is an I/O storm. ioa_spawn_budget answers "how many MORE blocked processes may I add and stay under that line, leaving reserve slots for everyone else" -- a number DERIVED from the box, never chosen. A third state is first-class: /proc/stat unreadable or truncated -> IOA_UNREADABLE, never a guess (the caller owns its fallback and must NAME it). license_tier: ORIGINAL No hw writes (Rule 26). lib (no main)