About AtWhatMileage
This site has not finished being set up. The publisher name and governing jurisdiction are missing from the site's configuration, so the pages that depend on them — this one, the privacy policy and the terms of use — cannot name who is responsible for the site or whose law applies to it.
To fix it, set legal.owner and
legal.jurisdiction in app/config/config.php, then clear
the page cache. Until then these pages are incomplete, and they should not be
relied on by a reader or submitted to anyone for review.
This site answers one question that the car industry, the used-car listings and the review sites all leave alone: at what mileage does a given car actually start to break, and with what? Not whether it is reliable in the abstract, and not what it scored in a survey — at what odometer reading the people who owned one reported the water pump, the torque converter, the ABS module or the timing chain letting go.
- What this is
- Who publishes it
- Where the data comes from
- Independence and funding
- Corrections
- How the site is built
What this is
Every complaint filed with the US National Highway Traffic Safety Administration carries an odometer reading taken at the moment the fault happened. That single field is unusual: safety databases normally record what broke and when it was reported, not how far the car had travelled. NHTSA publishes the raw records, millions of them, as flat files that almost nobody reads. Nothing else on the public internet turns those readings into distributions — a median, a spread, a curve — for a specific make, model and model year.
That is the whole of the idea. The database behind this site holds 1,943,536 owner complaints, of which 1,092,729 carry a usable mileage, spread across 66,828 make-model-year combinations from 1,691 manufacturers, alongside 134,500 recall campaigns. Complaint narratives are sorted into named failure modes, the mileages behind each mode are turned into percentiles, and the result is published as one page per vehicle, one page per failure mode within a vehicle, and a lookup tool that takes your current odometer reading and tells you what tends to go next.
What the site deliberately does not do is rank cars as good or bad. A risk score here is a rank against other cars of the same model year, because a 2009 model has had fifteen more years to accumulate complaints than a 2024 one and comparing them directly would be arithmetic dressed up as judgement. Every number is documented on the methodology page, which also lists, at some length, the things these figures cannot tell you.
Who publishes it
The publisher of this site has not been recorded in its configuration. Until that is corrected, this page cannot tell you who is responsible for it, which is a reasonable reason not to rely on it.
There is no newsroom here and no editorial staff. The site is a piece of software: it downloads the federal data, classifies it, computes the statistics and writes the pages. Nobody types a verdict about a car into a database. That is a limitation as much as a virtue — software cannot tell you that a particular failure mode was fixed by a mid-year part revision — and the pages say so where it matters.
Where the data comes from
Three public NHTSA datasets, all free to use and to republish: the Office of Defects Investigation complaint file, the recall campaign file, and the vehicle catalogue used to organise everything by make, model and year. They are re-downloaded on a schedule and the statistics are recomputed from scratch each time, so a figure on this site is never a stale copy of an older calculation.
Source data last imported 2026-08-22 Statistics last recomputed 2026-08-20
Two things about the source are worth knowing before you read a number anywhere on this site. Complaints are voluntary: nobody is required to file one, most failures never are, and so every count here is a floor rather than a census. And an owner-reported odometer reading is a person typing a round number from memory into a federal form, which is why mileages are presented rounded to the precision the sample can carry rather than to the nearest mile.
Independence and funding
This site currently carries no advertising and no sponsorship, and it sells nothing. If that changes, it will change visibly: an ad slot is not something a reader can miss, and the privacy policy is updated in the same deployment as the ad code rather than afterwards.
What will not change is the separation between the money and the numbers. No manufacturer, dealer, parts retailer or warranty seller has any input into what appears on a vehicle page. Nothing is removed, softened or reordered on request from a company whose product the data reflects badly, and there is no arrangement under which that could be bought — the statistics are generated by code from public records, so there is nothing to negotiate over. A request to change a figure is only ever treated as a bug report, and it is answered the same way whether it comes from an owner or from a manufacturer's legal department: show which step of the arithmetic is wrong.
Corrections
Mistakes here fall into three kinds, and they are fixed differently. A classification error — a complaint filed under the wrong failure mode, or a failure mode with a misleading name — is a fault in this site's catalogue and gets corrected in the code. A computation error — a figure that does not follow from the formulas on the methodology page — is a bug and gets a fix. An error in the underlying federal record cannot be fixed here at all, because this site is downstream of it; the right place for that is NHTSA, and we will say so rather than quietly overwrite what the source says.
Every page on this site can be checked. The methodology page has the arithmetic in full, and every computed page is also available as JSON, so a figure can be recomputed from the source data rather than taken on trust. If something is wrong, tell us — the contact page explains what to include so that a correction can actually be verified.
How the site is built
Plainly, and on purpose. There are no third-party trackers, no analytics scripts, no social widgets, no web fonts and no cookies set by this site. Pages are rendered once and served as static files, which is why they load quickly on a phone in a car park — the situation the mileage lookup was built for. The figures are published under CC BY 4.0: take them, cite them, link to the page you took them from. The terms of use set out the licence and the limits, and the most important limit is short enough to repeat here — nothing on this site is mechanical advice, and no statistic about a fleet predicts what your car will do.
How every number is computed Report a correction What breaks at my mileage