nx_condition_report.nx

nx_condition_report.nx -- MANHEIM-BUILD-L3 (assurance/trust): the CONDITION REPORT GRADER (Manheim Condition Report / AutoGrade-class), the inspection grade every wholesale lot carries. Sovereign (nx_cc->nxasm, no gcc), integer-exact + deterministic. WHY GRADE INTEGRITY IS THE WHOLE GAME: the grade drives the price AND it is the #1 source of ARBITRATION disputes. A naive grader that AVERAGES category scores produces DANGEROUSLY WRONG grades for structurally-compromised cars -- a frame-damaged car with pristine cosmetics averages out to "excellent", which a buyer will arbitrate (and which then poisons any valuation built on it). The fix: a CRITICAL DEFECT must CAP the grade, never be washed out by good panels. ENGINE: cr_naive = weighted category average / 2 (ext 30 / int 20 / mech 30 / tires 10 / glass 10, sum 100) cr_grade = cr_naive, but CAPPED at the structural ceiling (2.0 = 20) if ANY critical flag is set (frame damage / flood / airbag-deployed / salvage brand). grade scale 0..50 (= 0.0..5.0). ===== S-CLASS EXCEED, MEASURED (no-wave law) ===== main() is an EXCEED BENCH vs the naive averaging grader on grade INTEGRITY (deterministic, non-circular): A (frame damage + pristine cosmetics): naive overgrades to 47 ("4.7 excellent" -- WRONG/dangerous); ours CAPS at 20 ("2.0 structural") -> a 27-point integrity gap the naive grader gets wrong. NEG-CONTROL B (same pristine cosmetics, NO critical flag): ours == naive == 47 -> ours does NOT always-cap, it tracks real condition (proves the cap is defect-triggered, not a blanket clamp). C (monotonicity): a clearly-worse car grades strictly lower. + determinism. HONEST SCOPE (no-overclaim): the measured exceed is grade INTEGRITY vs the naive-average baseline -- NOT a claim to reproduce a specific human inspector's grade or Manheim's exact proprietary AutoGrade algorithm (those need real inspection data = operator-host). Evidence -> knowledge/status/condition_report.log. license_tier: ORIGINAL