nx_gate_mutation_sweep.nx

nx_gate_mutation_sweep.nx -- BATCH mutation-scoring across the gate universe (the liar-killer worklist engine for the sensor-gap program, [[project-nishi-sensor-gap-census-2026-07-14]]). usage: nx_gate_mutation_sweep <pair-budget> [mutants-per-pair] 0 = DRY RUN (list pairs only) DISCOVERY (mechanical, evidence-based -- the confirmed naming law): for every runtime/_hdl_build/ nx_X_gate.nx, the pair (subject=nx_X, gate=nx_X_gate) qualifies iff: * subject nx_X.nx exists (in _hdl_build or runtime/) * the gate actually IMPORTS the subject (runtime-built needle `import "nx_X.nx"`, quote bytes composed) * the gate is not RISKY (listeners/network/bench markers), not build-only, not infra, not known-broken (gatebase_*.tsv.refused marker) For each qualifying pair (budget-bounded, resumable): fork the installed probe _offc/nx_gate_mutation_probe.elf <subject> <gate> <mutants> -- it runs the identity control + evenly spread ROR/AOR mutants and prints the mutation score. This sweep captures the probe output, parses the LAST THREE ints (killed / total / survived), and BANKS: knowledge/status/mutscore_<subject>.tsv subject \t gate \t killed \t total \t survived knowledge/status/mutscore_<subject>.out the full probe transcript (survivor op+line = the worklist) Ledger existence = resume-skip. SURVIVORS never fail the sweep -- they ARE the product (ranked liar-killer worklist). Pipeline-RED pairs (identity fail etc.) are tallied as findings. Never-brick: writes only mutscore_* ledgers/transcripts (+the probe's own archived copies). Sovereign nx_cc->nxasm; CWD=nxc2 root. license_tier: ORIGINAL