What the MOT data cannot tell you

This site is built on 111,079,869 MOT tests — every test recorded in England, Scotland and Wales for the vehicles we cover. It is a census of inspections, not of cars, roads or drivers, and the difference matters: five of the questions people most want answered are not in the data at all.

What is actually in a row

Each test record carries the vehicle's make, model, first-use date and fuel, the odometer reading, the result, the tester's wording for every defect, the defect's category, and the keeper's postcode area. That is a rich row. It is also a closed list, and everything below is on the other side of it.

1. Roads

There is no road-condition field, no surface type, and no location finer than the keeper's postcode area. So the pothole piece cannot say how many failures a pothole causes — it can only show which parts fail most and let the engineering do the talking. Anyone who puts a number on "potholes cause X% of failures" from this dataset is inventing a variable.

2. The driver, and the use

No age of driver, no annual mileage between tests, no distinction between a taxi and a second car. A high-mileage van and a low-mileage one of the same model-year are pooled until the mileage band separates them, and even then the band is a snapshot of the odometer at one test rather than the life the car has had.

This is the single biggest confounder in every pass rate on the site, which is why they are adjusted for the model-year and mileage mix before any comparison is made.

3. The garage

The anonymised results have no test-station field. The station directory comes from the DVSA's separate published list of active stations, and the two datasets cannot be joined: we can tell you a station exists, not how it tests. Anyone ranking garages from MOT data is ranking postcode areas.

4. Money

The extract contains no prices. Every bracket on this site is our estimate — parts-weighted, built from the part type in the tester's wording — and the only real prices here are the ones readers reported for particular repairs. That is the reason the price list labels a bracket as an estimate and the calculator totals ranges rather than a figure.

5. The cars that never arrived

Cars are exempt from their first MOT until three years old, so young vehicles appear rarely and their pass rates are noise. At the other end, a car scrapped before its next test simply stops appearing — the dataset is a record of vehicles that turned up, not of vehicles that existed. A model with a high pass rate may be reliable, or its bad examples may have been scrapped young.

Why 15% of what we hold is not published

We hold 12,797 model-years. 1,885 of them are deliberately not indexed: 1,752 carry fewer than 100 tests, and 164 are unnamed motorhome filings ("Suzuki Motorhome"), which are real tests in a bucket with no manufacturer to compare. Both would render as a page of numbers that cannot support a claim, so they exist behind a noindex note instead. The same discipline sets the 500-test floor for model hubs and the eight-station floor for town pages.

It is a strange thing for a site with 12,797 pages to write down, but the alternative is worse: a page that looks like evidence and is not.

What it can answer

• Is this model-year worse than its age and mileage predict? That is the adjusted comparison, and it is the one figure here that has been through a model rather than a mean.
• What actually fails on it? The defect mix is recorded, itemised, and the largest categories are consistent enough to fix before the test rather than after.
• What will the repair cost? As a bracket, for the items on your sheet — which is the question that decides whether the car is worth fixing.

Where a piece here reasons rather than measures, it says so. Where a number is an estimate, it is labelled one. If you find a figure you think is wrong, the methodology says how it was built and the address is open.