nx_evidence_verdict.nx

nx_evidence_verdict.nx -- THE EVIDENCE LAW, shared. Operator directive 2026-07-30: "we dont have evidence for most of our systems that real actual proven sota ... triangulation between an llm, mechanistic, and human validated with proven outputs not just navel gazing inputs." ROOT CAUSE THIS LIBRARY EXISTS TO KILL (all three MEASURED on the live system 2026-07-30): D1 QUORUM nx_swcompare_evidence.nx:480 read `liar_green = (green >= 1)`. ONE green gate out of five stamped the whole domain PROVEN. Four RED gates were invisible to the headline number. D2 EXPIRY nx_sota_status.nx read the stamp's ok=1 unconditionally; `epoch=` was parsed for DISPLAY only and never gated the verdict. Oldest live stamp measured at 1183616s = 13.7 days. D3 PROVENANCE ev_stamp wrote no host and no artifact digest, so a verdict produced where the gate binaries exist (the laptop tree) is byte-indistinguishable from one produced where they do not (the NAS, where 0 of the declared medbilling/warden gate elfs resolve). Each defect is the same shape: a VERDICT that outlived, or never contained, the EVIDENCE that earned it. The fix is not a TTL constant bolted onto one reader -- it is ONE law every reader imports, so the optimism cannot be re-introduced per-instrument. Sibling of nx_honesty_grade_lib.nx (which grades what a claim's PROSE earns); this grades what a claim's EXECUTION earns. Import both. THE BAR (a claim is PROVEN only if EVERY clause holds -- fail-closed, downgrades dominate): 1 QUORUM every declared gate ran and every one came back GREEN. Not "at least one". 2 FRESHNESS the stamp is younger than the policy TTL. A verdict has an expiry date or it is a rumor. 3 PROVENANCE the stamp names WHERE it ran and the DIGEST of what ran (the SLSA/in-toto lesson: provenance without builder identity + artifact digest is unverifiable by construction). 4 NON-VACUITY each gate has a RECORDED RED observation, i.e. it has been proven able to FAIL. A gate never observed failing measures nothing. (2026 mutation-testing result: suites at 100% line/branch coverage scoring 4% mutation score -- coverage is not detection.) 5 TRIANGULATION at least TWO distinct method classes agree, and at least one is MECHANISTIC. An LLM-only verdict can NEVER prove: 2026 judge studies measure run-to-run self- agreement at coin-flip rates, and judges over-credit incomplete work by silently supplying the missing premise. A model may CORROBORATE; it may not CONVICT. DISAGREEMENT IS NOT AVERAGED. Two classes that disagree return RED, never a mean. A contradiction is a finding, not a rounding error -- averaging it is how a real defect gets laundered into a passing score. TTL and the required class count are PARAMETERS, never literals here (rule 11: thresholds live in config, not code). The caller reads knowledge/evidence_policy.conf and passes them in. No main, no syscalls -> any grader/census/judge/gate imports it. license_tier: ORIGINAL