Data Center

How it's built: The reliability score is a TRANSPARENT heuristic on a 1.0 (worst) – 5.0 (best) scale (the site shows it ×2, out of 10). It starts at 5.0 and subtracts log10(complaints + 1) × 0.55, 0.30 per critical recall, and 0.10 per severe recall. It is a complaint-and-recall-VOLUME signal, NOT a per-capita failure rate: we do not yet have sales/registration denominators, so higher-volume models can score lower simply because more of them are on the road. "top_component" is the component category with the most complaints for that vehicle. Alleged death/fire counts are sums of unverified consumer allegations across the vehicle’s complaints. Motorcycles are excluded (this site is cars-only).

How it's built: Each row is one (vehicle, component-category) cluster. Mileage is parsed from the NHTSA ODI bulk export and validated to the 1,000–400,000 mi band (data-entry artifacts outside that range are dropped). Quartiles are published only when a cluster has ≥8 mileage-bearing complaints. Complaints are unverified consumer reports.

How it's built: Subset of the failure-mileage dataset where the median failure mileage exceeds 60,000 mi (the most common US powertrain term, 5yr/60k) and the cluster has ≥8 mileage-bearing complaints. "miles_past_warranty" is median minus 60,000. Complaints are unverified consumer reports.

How it's built: Each row is a (vehicle, component-category) cluster carrying ≥1 complaint flagged by NHTSA for crash, fire, injury, or death. Counts are sums of those NHTSA flags across the cluster’s complaints. NHTSA complaints are submitted by consumers and are not verified by NHTSA or ProblemsByVin; treat as allegations.

How it's built: Families are the editorial set powering /engines/. For each family we sum complaints/recalls/harm flags across every matching (year, make, model) in the DB. "complaints_per_modelyear" = total complaints ÷ model-year entries present. Complaints are unverified consumer reports.

How it's built: Families are the editorial set powering /transmissions/. NHTSA does not expose a standalone "transmission" complaint category — transmission complaints bucket under "powertrain" — so this dataset aggregates by curated family (matching year/make/model lists), not a category filter. Complaints are unverified consumer reports.

How it's built: Candidates are (vehicle, component) clusters with no recall for that component in our DB and ≥ the complaint threshold. Each candidate vehicle is then queried against the live NHTSA recalls-by-vehicle API; any cluster whose component is in fact covered by a real NHTSA recall is REMOVED. Survivors are genuine recall gaps as of the verification date. Complaints are unverified consumer reports; absence of a recall is per NHTSA records on the verification date.

How it's built: Parsed from NHTSA’s ODI investigations flat file (FLAT_INV) and matched to our vehicles by year/make/model. One row per NHTSA action number. "status" is open when the federal record carries no close date; "recall_campaign" holds the NHTSA campaign number when the investigation led to a recall. An investigation is NHTSA examining a pattern — not a confirmed defect.

How it's built: Built from NHTSA's technical service bulletin flat file (manufacturer-to-dealer communications NHTSA collects). "tsb_count" is the distinct bulletins mapped to that (year, make, model); "top_category" is the NHTSA component area with the most. A TSB documents a known condition and a repair procedure — it is NOT a recall, and bulletin types range from major driveability fixes to minor informational or warranty-extension notes (Service Bulletin/Repair Instructions, Service Campaign, Warranty Program/Extension, Emissions, Over-The-Air). A high count reflects DOCUMENTATION VOLUME, not severity, and partly tracks a model's sales volume and age. Dates are bulletin communication dates. complaint_count and recall_count come from the NHTSA owner-complaint and recall record, for context. Motorcycles excluded.