How Vigilot scores a VIN.
Every trust score is the same formula: start at 100 and subtract the weight of every signal that fires. This page lists every signal, its weight, and the threshold the engine uses to decide when it fires — so dealers can read it once and learn the rule.
The formula. Score = 100 minus the sum of per-hit weights, clamped to 0–100. Weights and thresholds live in src/lib/business/reconcile.ts so a tweak flows to the engine, the per-VIN detail view, and this rubric in one edit.
The sources. NMVTIS (state titling brands), NHTSA SaferCar (recalls), three auction-history wires (Manheim, OPENLANE, ACV), and Vigilot's proprietary AI-cloned-site detector that watches for cross-lot storefront duplicates.
The per-VIN view. The same four weights fuel the per-VIN "Why this score?" panel on the dashboard. If a dealer drills into a VIN, the explanation they see is the same rules in this table.
| Signal | Weight (of 100) | Trigger threshold | Sample VIN |
|---|---|---|---|
| Odometer rollback | 40% | 5% drop, ≥ 1,000 mi over an older reading | 5NPE34AF··0····· |
| Branded title | 25% | Any NMVTIS row with a non-Clean brand label | 1C4NJCB00GD57··K |
| Open NHTSA recall | 20% | One or more NHTSA rows with repaired = false | 2T1BURHE··JC0···· |
| Auction-history contradictions | 15%(shared weight) | Distinct cross-source contradiction between auction and NMVTIS odometer readings | 5NPE34AF··0····· |
| Cross-lot storefront VIN conflict | 15%(shared weight) | Same VIN on multiple dealer storefronts where at least one disagrees on odometer or brand | 3FA6P0H7··HR····· |
Odometer rollback
A newer reading whose odometer is more than 5% lower (and at least 1,000 mi lower) than an older reading from a distinct source.
Sources compared: NMVTIS rows vs. auction pickup-mile records. Same-source pairs only count when the dates differ.
A rollback also increments the cross-source conflict counter — the same event fires two rules.
Branded title
Three tiers of branded-title severity, from a single non-Clean NMVTIS brand on file to a disagreement between distinct brand labels.
Tier 1 — any non-Clean NMVTIS brand on file → brand-title flag fires.
Tier 2 — distinct brand labels across NMVTIS rows → flag fires AND AI conflict resolution is attached.
Open NHTSA recall
Single-source flag against NHTSA SaferCar. The recall row does not require cross-source disambiguation — the flag itself is the meaning.
Source is NHTSA SaferCar only. No AI step is invoked for single-source flags.
If every NHTSA row reports repaired = true, no flag fires — the VIN is recall-clean.
Auction-history contradictions
A NMVTIS-vs-auction odometer contradiction on file. The same event is what triggers the rollback rule — the conflict is logged separately and counts toward the cross-source conflict weight.
Both rules apply the same weight; the engine only flags a row when a cross-source contradiction is on file.
Auction-history data comes from Manheim, OPENLANE, and ACV pickup-mile feeds.
Cross-lot storefront VIN conflict
Vigilot maintains a proprietary index of cross-lot listings. If the same VIN appears on more than one storefront with disagreeing odometer or brand, we log a cross-source conflict.
Both rules apply the same weight; the engine only flags a row when a cross-source contradiction is on file.
The cross-lot layer feeds the AI-cloned-site detector surfaced on the landing page.
weights update as the model retrains.