nx_sensor_gap_census.nx

nx_sensor_gap_census.nx -- THE live 4-axis SENSOR-DEPTH risk map across every gate in runtime/_hdl_build. Answers the operator's "get the gap in sensors fixed across the nishi environment": for the WHOLE gate universe, what fraction carries each of the four SENSORS that make a verdict trustworthy -- 1 LIAR-KILLER (neg-control / physics-invariant / evidence-grounding) -> can't pass a GAMED number 2 REGRESSION BASE (stored baseline + no-backslide) -> can't SILENTLY backslide 3 HANDOFF TELEMETRY (per-stage quality/bytes/time/fail) -> can LOCALIZE where it breaks 4 EXTERNAL-COMP (named industry oracle: x264/MSU, CUDA/WARP, V8, ...) -> EXTERNALLY judged not self-graded Per [[feedback-handoff-chain-sota-gate-doctrine]] #1 rule a self-judged verdict is worthless; axis 4 is deepest. Axis-4 here is a fast per-gate marker scan; the AUTHORITATIVE per-workstream WIRED/UNWIRED status lives in nx_benchmark_suite_registry -- the two reconcile (this flags coverage; that names the suite). Matching is CASE-INSENSITIVE (catches CUDA/X264/Baseline) with WORD-boundary mode for short ambiguous tokens (so "red" does NOT match inside "REQUIRED"/"STRUCTURED" -- a universe-inflation gaming vector, liar-killed below). This organ is ITSELF liar-killed: POS fixture classifies 5/5, NEG fixture 0/5, a REQUIRED/STRUCTURED neg-control must NOT register as a verdict-emitter, classification is deterministic, and an UNREADABLE gate is counted BLIND (never fabricated as covered). It carries a no-backslide regression gate on its OWN coverage (knowledge/status/sensor_gap_baseline.tsv). It PRINTS the honest risk map REGARDLESS; GREEN = classifier TRUSTWORTHY and coverage did NOT backslide (GREEN != "coverage good" -- the blind gates ARE the work). Sovereign nx_cc->nxasm (no gcc, no sh). Run with CWD = nxc2 root. expect_exit: 0 license_tier: ORIGINAL