nx_arbitration.nx

nx_arbitration.nx -- MANHEIM-BUILD-L3 (assurance/trust): the ARBITRATION ELIGIBILITY engine (Manheim / NAAA-class dispute resolution), the rules that decide whether a buyer's post-sale claim is arbitrable. Sovereign (nx_cc->nxasm, no gcc), deterministic. WHY IT MATTERS: arbitration disputes are a major friction + cost in wholesale. The failure mode of an opaque / single-factor process is INCONSISTENCY and OVER-ADMISSION -- a claim that is expensive but time-barred, sold as-is, or in a non-arbitrable category gets let through by a "is it expensive?" heuristic, creating cost + unfairness. Correct eligibility is CONJUNCTIVE: every condition must hold. ENGINE (auditable -- returns the precise failing reason): ELIGIBLE iff ALL: repair_cost >= threshold AND filed within window AND arbitrable category AND NOT as-is. reason codes: 0=eligible 1=below_threshold 2=window_expired 3=non_arbitrable_category 4=as_is_sale naive baseline = checks ONLY the dollar threshold (the over-admitting heuristic). ===== S-CLASS EXCEED, MEASURED (no-wave law) ===== main() benches ours vs the naive dollar-only arbiter over 4 claims (deterministic, non-circular): A genuine claim -> both ELIGIBLE (ours admits real claims). B expensive-but-LATE + C expensive-but-AS-IS -> naive OVER-ADMITS (wrong), ours REJECTS with the exact reason. D below-threshold -> both reject. MEASURED: naive over-admits 2 of 4 (the time-barred + as-is claims); ours over-admits 0. NEG-CONTROL is built in: D shows ours still rejects on the dollar rule (not a blanket-reject), and A shows ours admits genuine claims (not a blanket-admit) -> the discrimination is real. HONEST SCOPE (no-overclaim): exceed is conjunctive-correctness + determinism vs the naive single-factor baseline -- NOT a claim to match a specific real NAAA/Manheim arbitration policy's exact thresholds, windows, or category list (those are policy params = config / operator). Evidence -> knowledge/status/arbitration.log. license_tier: ORIGINAL