nx_market_insights.nx

nx_market_insights.nx -- MANHEIM-BUILD-L2 (DATA MOAT): the USED VEHICLE VALUE INDEX (UVVI), the Nishi answer to Cox Automotive's flagship public Manheim Used Vehicle Value Index -- a macro index tracking wholesale price LEVEL over time. Sovereign (nx_cc->nxasm, no gcc), integer-exact + deterministic. THE METHODOLOGY PROBLEM a good index must solve: month-to-month the MIX of what sells changes (more trucks one month, more compacts the next). A naive "average sale price this period / base period" index is CONFOUNDED -- it reports phantom inflation when the mix shifts toward pricier classes even if every model's price is flat. A proper index (CPI-style Laspeyres) holds a FIXED BASKET (base-period quantities) and reprices it, isolating TRUE price change from mix change. mi_naive = avg_price(period) * 1000 / avg_price(base) (confounded by mix) mi_basket = sum(base_qty * period_price) * 1000 / sum(base_qty * base_price) (fixed basket, robust) index base = 1000 (per-mille; 1000 = "same as base period"). ===== S-CLASS EXCEED, MEASURED (no-wave law) ===== main() is an EXCEED BENCH vs the naive average-price index, on a deterministic, non-circular axis where the naive baseline PROVABLY FAILS (Referee high-TNR): period T: the MIX shifts toward trucks but every matched model's price is FLAT. truth = no price change (index should read 1000). naive reads 1285 (+285 permil PHANTOM inflation); ours reads exactly 1000. -> ours is correct, naive emits a false signal. period U: NO mix shift but all prices rise 10%. BOTH read 1100 (+100 permil) -> proves ours TRACKS a real price move (not trivially clamped to 1000) and naive is only wrong WHEN the mix shifts. NEG-CONTROL: under T the naive index must DEVIATE from the truth 1000 (proves the bench discriminates). HONEST SCOPE (no-overclaim): the measured exceed is vs the NAIVE AVERAGE-PRICE INDEX on mix-shift robustness + determinism -- NOT a claim to beat the real Manheim UVVI's proprietary methodology on real market data (that needs the real series = operator-host head-to-head). NOT CLAIMED. Evidence -> knowledge/status/market_insights.log. license_tier: ORIGINAL