nx_gate_mutation_probe.nx

nx_gate_mutation_probe.nx -- SOVEREIGN MUTATION TESTING of gate sensor power (the SOTA ruler). Grounding (fetched live 2026-07-14 via nx_https_get, en.wikipedia.org/wiki/Mutation_testing): "the mutation score is the number of mutants killed / total number of mutants"; goal #1 = "identify weakly tested pieces of code (mutants not killed)". Lineage: Lipton 1971, DeMillo/Lipton/Sayward 1978, Budd 1980. RULER = industry-standard; TOOL = GREENFIELD on NishiLang (PIT=JVM, mutmut=Py, Mull=LLVM -- none run here) => sovereign-built, guarded by internal liar-killers per [[feedback-3rd-party-proving-for-nonnovel]]. TWO MODES: (no args) CURATED v1: fixed pair (runtime/nx_handoff_gate.nx x its proof gate), 6 hand-designed killable mutants -> GREEN iff 6/6 killed. This is the REGRESSION GATE on the sensor stack's power (baseline-wrapped). <subject> <gate> [budget] GENERIC v2: ANY lib x gate pair (naming law nx_X / nx_X_gate). Auto-generates mutants with classic operators (ROR: < > <= >= == != ; AOR: + -) at sites OUTSIDE comments/strings (region mask -- mutating a comment is a guaranteed FALSE survivor), evenly spread over the file. Survivors are reported as FINDINGS (op + line) = the liar-killer worklist; score is DATA (verdict = pipeline soundness, not score). ⚠kills assume the gate is idempotent (re-run-stable). LIAR-KILLERS ON THE PROBE ITSELF (both modes): * IDENTITY CONTROL -- an unmutated copy MUST stay GREEN (pipeline introduces no fault; models the equivalent-mutant honesty case). Identity RED => all results VOID. * NEEDLE/SITE-NOT-FOUND = RED -- never silently skip. * import needle is BUILT AT RUNTIME as `import "<subject>.nx"` (quote bytes composed) so a comment mentioning the filename can never eat the splice (the v1 trap, killed generally). * signal-aware rc decode (128+sig). NEVER-BRICK: originals READ-ONLY; only __mutprobe_* copies written, then ARCHIVED to knowledge/mutprobe/. Sovereign nx_cc->nxasm; CWD=nxc2 root. license_tier: ORIGINAL