Changelog

• 2026-09-29 — new pages. Every model now has a page for each build year (its failure rate at each age against the same model's norm at that age, with the failure groups of that year's cars), a page for each MOT failure group where the model has at least 30 failed items, its postcode areas, and its record in other countries' inspections where the data exists. Eleven failure-group overviews and 118 postcode-area pages rank every model on the same figures. Nine more models are published: BMW 1, 4, 5, 6, 7 and 8 Series, MG ZS, Toyota Prius+ and Vauxhall Combo Life. Correction (fixed on 26 September): the area figures on model pages compared an area's 2019–2024 failure rate with the model's 2024 rate; both now cover the same years. No percentage is published from fewer than 30 tests.
• 2026-09-26 — correction: build-year figures counted MOT tests from 2019 to 2023 twice since 21 September (a pipeline error). Each test is now counted once: sample sizes on the build-year tables are smaller, the intervals wider, and 667 build-year verdicts changed, most of them from "better" or "worse" to "no different"; build years that no longer reach the published minimum were removed. The fleet block on model pages named ages 5 and 10 even where the data held other ages; it now names the ages it shows. Definitions on the methodology page now quote DVSA's own wording, and the licence notice for every outside source is at/data/ATTRIBUTION.md. The downloadable dataset is now version 2024-v5 (DOI 10.5281/zenodo.22975051), which corrects the build-year file; every other file carries the same numbers as v4.
• 2026-09-25 (evening) — data and dataset: mileage is now compared inside the same age group (3–5, 6–9, 10–14 and 15+ years), so high-mileage cars are no longer also the older ones, and a mileage effect is stated only when the intervals of the lowest and highest bands do not overlap; build-year verdicts now follow the one-decimal bound printed beside them (a year whose interval touches 100.0 reads "no different"); Netherlands inspections were refreshed and the Vauxhall Viva is matched to the Opel Karl; Mercedes-Benz model names are corrected. Model pairs show where the two diverge across every failure group, by age band, and with recalls, other countries and repair costs side by side. The downloadable dataset is now version 2024-v4 (DOI10.5281/zenodo.22966354): the same family and defect-group numbers as v3, plus outside sources, mileage within age groups and model years.
• 2026-09-25 (later) — data: recall campaigns now reach every model the DVSA register names (Volkswagen, Fiat and Honda campaigns were filed under other spellings and had been missed), with all campaigns listed and sorted by date; inspection results from Norway now cover the whole of 2024 (the first quarter only before), and Norway, the Netherlands and Finland are matched across the full model list, a Vauxhall as the Opel of the same name; a parsing error that gave the Mercedes-Benz C-, A- and E-Class only their AMG versions in Finland is fixed; US owner complaints and bulletins cover many more models, each with its US name and model years. Small samples are now published as rough estimates with their interval instead of being hidden (see methodology). The worst and most reliable cars rankings show 20 models with a chart of their intervals, and brand pages show where each brand's models fail above other cars of the same age.
• 2026-09-25 — site: a new page layout that puts the answer first, with a table behind every figure and a CSV download, on half of model pages and half of comparison pages (the other half keeps the previous layout while we measure the difference), on all brand pages, and on the worst and most reliable cars rankings. The rankings now lead with mainstream cars (100,000+ first MOTs); vans and pickups, and models sold as both van and car, are listed separately. Definitions and a citation block were added to the methodology. No figure changed.
• 2026-09-24 — model pages compare failure groups and dangerous defects with the national rate for the same years (2019–2024) rather than with 2024 alone.
• 2026-09-24 — dvsa-2024-v2: correction of every 2024 figure. The DVSA 2024 source files repeat many test records: each monthly file is several export chunks joined together, and the same test appears two to eight times (66,857,355 rows hold 42,598,624 distinct tests). Release dvsa-2024-v1 counted the repeats. Passes were repeated more often than failures and some models far more than others, so v1 understated 2024 failure rates (national class-4 initial failure rate 26.01% → 28.28%), overstated sample sizes about 1.6-fold (33,012,215 class-4 initial tests, not 52,245,925) and made confidence intervals too narrow. Every page is recomputed from one record per test ID; the method is described in the methodology. Model indices moved by a median of 4.8 points; 46 of 755 models moved across 100, almost all of them within a few points of normal; the overall order held (rank correlation 0.98; 17 of the 20 most reliable and 19 of the 20 worst models are unchanged). Also in this release: the BMW 3 Series family now includes badge-code spellings (320D, 330E and similar), which v1 missed. The v1 data capsule stays archived; a corrected capsule will be published as a new version.
• 2026-08-03 — new report: MOT failure by fuel type. Electric, petrol, diesel and petrol-hybrid compared on the same age-standardised index, with the failure categories that separate them. Same release dvsa-2024-v1, no change to any previously published figure.
• 2026-07-30 — dvsa-2024-v1: initial release. 25 model families computed from 52,245,925 class-4 initial tests (66,857,355 total test records, calendar 2024; both counts include repeated records, corrected on 2026-09-24). Frozen national age baseline published. Layers: identity & sample, age-adjusted index with CI, component failure and advisory anatomy, age curve, powertrain splits, mileage percentiles.

How each model's index moved from v1 to v2

Baselines are frozen per release: statistics computed against dvsa-2024-v2 remain reproducible after later releases are published.