nx_coverbeat.nx

nx_coverbeat.nx -- RUN THE COVERAGE INSTRUMENTS AND PROVE IT RAN THEM. RUNG R3 of the SOTA plan (nishifamily.com/code/sotaplan): the instruments exist and nothing runs them. Every finding on 2026-07-31 existed only because someone happened to look. A gate nobody runs is itself uncovered -- and that recursion has to bottom out in a beat. WHY THIS IS NOT JUST A SHELL LOOP. Two failure modes have to be impossible by construction: 1. THE SILENT NO-OP. A beat that reports fail=0 without having executed anything is the quietest failure in the system -- observed live in the learnings rail, where a harvest reported fail=0 for weeks while moving zero learnings. So this beat counts what it ATTEMPTED and what it ACTUALLY REAPED, prints both, and goes RED when they disagree. `ran` is evidence; `fail=0` is not. 2. THE FLATTERING AVERAGE. A rollup that means-averages its gates lets a healthy plane hide a dead one. The HEADLINE IS THE WORST VERDICT, never the average -- if any instrument is RED the beat is RED, and the log line names which. SCHEDULING: this is a ONESHOT. Launch it from cron exactly like the existing sovereign beats, e.g. */30 * * * * elderwesto /volume1/homes/elderwesto/nishihost/nx_coverbeat.elf >> logs/coverbeat.log 2>&1 It is deliberately idempotent and side-effect-free apart from its own log line, so running it twice in a minute is harmless -- a beat that cannot be safely re-run cannot be safely scheduled. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0