How the numbers are made

Safety Signal

Safety Signal starts every model-year at 100 and subtracts points per open NHTSA recall campaign by severity: critical 15 (fire risk, park-outside or do-not-drive advisories), high 8 (brakes, steering, airbags, loss of drive power, structure, battery), moderate 6, minor 3 (software, labels, lighting fixed by an update). The most serious recall counts in full and each further one counts 85% of the one before, because a vehicle sold in millions is recalled more often than a rare one without being more dangerous — and because a flat subtraction bottomed the scale out, scoring a vehicle with 22 recalls within a point of one with six. A ≥ 90, B ≥ 80, C ≥ 70, D ≥ 60, otherwise F. Read it as what it counts: recall campaigns on the federal record, weighted by severity. The damping reduces the sales bias by roughly 40% but does not remove it: like our complaint counts, this is not normalized by how many were sold, so a best-seller still scores somewhat worse than an obscure model with the same real risk. NHTSA crash ratings, do-not-drive and park-outside advisories, and the crash, fire, injury and death counts in owner complaints are reported separately on every page and never blended into the letter — a vehicle can hold five stars and a poor letter at once, and where that happens we say so rather than leave you to notice. Because the scale subtracts from a clean sheet, a model year too new to have been recalled, complained about or bulletined scores full marks on an empty record; we label those grades provisional rather than present them as earned.

Nameplates and years

A "vehicle" on this site is one nameplate in one model year as defined by the EPA's base-model field, so gas, hybrid and plug-in versions of the same nameplate share a page and their recalls are shown together with a powertrain tag where NHTSA distinguishes them. NHTSA and EPA spell model names differently; we match them automatically and publish the leftovers in our build report so the alias table keeps improving.

Complaints

Counts are unique NHTSA complaint numbers per model year. They are not normalized by sales, so best-sellers look worse than they are; we show them as a pattern signal and never blend them into the grade.

Trim specs

Trim-level specifications, features and MSRP are transcribed from the manufacturer's official press kit or newsroom spec sheet by a model-assisted pipeline validated against hand-curated ground truth at roughly 95% cell accuracy; each value carries its source URL and a confidence score in the data file.

Diagnostic trouble codes

Codes are not a separate source. They are a WhichTrim dataset derived from the federal record, and the counts on a code page are documented references in that record — not failure rates, and not an estimate of how often a vehicle sets the code. Codes whose second character is 1 or 3 are defined by the manufacturer rather than by the OBD-II standard, so we publish no definition for those at all rather than guess at one.

Refresh

Nightly: the whole site is rebuilt from the current federal files every night, and manufacturer figures are picked up on the next build after a press kit is published. This page was generated 2026-10-26 10:34 UTC.