Updated 25 July 2026
We publish data. Data has errors. When we find one, or you do, we fix it and log it here with the date and what changed, rather than quietly editing. Our methodology explains how the figures are built.
Spotted something wrong, a superseded tax rate, a miscalculation, a misread source? Email cars@aitmultiverse.com or use the contact page, with the page and the correct figure. Corrections from tax authorities, regulators and industry bodies are especially welcome.
Log
Every entry below was found by counting what is actually on the page rather than trusting the sentence that introduces it: the same method we apply to recall records. It is the only reason these are on a public page instead of in a private fix list.
| When | What was wrong | How we found it | What changed | Status |
|---|---|---|---|---|
| 14 Aug 2026 | Our central claim was not true. The register said we hold the wording a recall carried before it was revised. For 7 of the 11 changes recorded at that point (6 remedy rewrites and 1 summary rewrite) we had not stored the earlier text at all. | Counting the stored fields rather than trusting the sentence above them. | The register, the homepage and one study were corrected the same day to say exactly which fields we keep a before-and-after for, and which we do not. | Fixed |
| 14 Aug 2026 | The change log contained things that were not changes. 14 entries recorded a list of affected models that the authority had merely reordered, and 1 recorded the authority’s own record-update date moving. | Comparing the before and after values of every model-list entry as sets. | The detector now compares model lists as sets, so ordering alone can no longer register as a change. The old entries were removed. Re-checked 16 August: none remain. | Fixed |
| 14 Aug 2026 | Two studies were published before we checked prior art. One implied a first that was not a first; one overstated how an authority handles amended documents. | Checking what already existed: after publishing, which was the mistake. | Both were corrected the same day, with the correction visible on the page rather than made silently. | Fixed |
| 15–16 Aug 2026 | The register’s tables stopped rendering. A maintenance script damaged the code that publishes them. The table structure was lost and for about a day the data appeared as run-together text. | A reader could have seen it. We found it by re-reading our own page rather than the page’s cache. | The publishing code was repaired and the tables rebuilt. The data itself was never affected: collection continued throughout. | Fixed |
| 16 Aug 2026 | The register conflated first sightings with amendments. The change table was introduced as “every such move we have observed”, meaning a recall changing after publication. Of 38 rows, 8 were amendments; 30 were baselines. | Counting the rows by event type instead of reading the sentence above the table. | The table now explains which entries are amendments and which are baselines, and why we publish both. | Fixed |
| 17 Aug 2026 | The weekly digest contradicted itself. A table headed “Recalls that were changed after publication” was filled mostly with first sightings: 24 of 25 rows. | The same count, applied to the digest. | The table was split in two, so the heading now describes what is underneath it. | Fixed |
| 17 Aug 2026 | Every “Skip to content” link on the site pointed nowhere. All 158 pages carried a link to an anchor that did not exist, so keyboard and screen-reader users who used it went nowhere. | Checking all 487 in-page anchors across the site, not just the visible links. | The missing target was added to every page. Re-checked: 157 of 157 pages now resolve it, with no duplicates. | Fixed |
| 17 Aug 2026 | Our Canadian source citation pointed at a service that had shut down. The link resolved, which is why no automated checker flagged it, but the page it loaded announced that the service closed on 29 September 2023. | Following our own citations by hand instead of trusting the status code. | Being repointed at Transport Canada’s current recalls page. Where our data is actually fetched from was never affected. | In progress |
| 17 Aug 2026 | Our own accessibility fix broke the thing it was fixing. Adding the missing main landmark produced four of them on every Elementor page, including the homepage. Four main landmarks are worse than none: invalid, and confusing to exactly the screen readers the fix was for. | Counting the elements after the change instead of checking that one had appeared. | Corrected within the hour. Every page now carries exactly one main landmark, one skip target and one H1: checked across nine pages on both page templates. | Fixed |
| 17 Aug 2026 | We published a coverage claim that was out of date within two hours. A new section said the wording history tracked “the 100 most recently published US campaigns rather than all 1,397”. The 100 was not a design decision: it was the limit of the public feed the engine happened to read, and we described the limit as if we had chosen it. | Checking why the number was 100 instead of accepting it. | The engine now reads the register’s own index and tracks the entire US holding, 1,445 campaigns. The sentence on the page was rewritten the same afternoon to say what actually happened. | Fixed |
25 July 2026, Driving Licence Cost Index (in progress). Following reader analysis, the Australia row will be re-based on realistic paid-instruction hours under the NSW/Victoria 120-hour supervised-driving regimes, and the wording around best-case assumptions revised (an underlying source assumes all tests are passed first try). Credit: readers on r/dataisbeautiful. Also under review for the same update: Canada (share-of-salary basis), Spain (typical real-world cost including retakes) and Germany (gross vs net salary presentation).
21 July 2026, Car Insurance Cost study (transparency fix). The insurance page now discloses that the TCO Index uses clearly labelled modelled estimates for Finland (≈€950/year) and Indonesia (≈€600/year), for which no comparable published premium exists. No figures changed.
The standard we correct against
This site changed its standard during the summer of twenty twenty six, and the correction log above spans both sides of that change. It is worth stating plainly what the standard is now, because it decides what counts as an error.
We no longer publish prices, tax rates, registration or inspection fees, insurance premiums, incentive amounts, fuel or electricity tariffs, or model years, and we no longer rank cars. Where a figure matters to a decision, we name the body entitled to publish it and send you there: the tax authority, the registration office, the roadworthiness inspection body, the vehicle safety regulator, or the manufacturer national organisation for recall status against a vehicle identification number.
That policy exists because of what the log above records. Every batch of figures we checked contained numbers that were not merely stale but wrong, and a reader who acts on a wrong number pays for it while we publish a correction weeks later. Removing the class of content that generates those errors is a better fix than getting faster at correcting them.
Pages that carried figures or rankings before the change now carry a published note at the top saying what was removed and why. Those notes stay. We would rather show the change than hide it.
What now counts as an error
- A statement of fact about a document, an authority or a procedure that is not true.
- A physical or mechanical claim that is wrong, misleading, or unsafe to act on.
- A residual figure, price, tax rate or model year that survived the change and should not have.
- A named brand, dealer, garage, insurer or finance provider presented as a recommendation.
- A broken link, a page that fails to load, or a tool that produces a wrong result from correct inputs.
The third and fourth items are the ones we most want reported. A change of this size leaves residue, and readers find it faster than we do.
How to report one
Send us the address of the page, the sentence quoted, and what you believe the correct position is. Where the point turns on a fact rather than on wording, tell us which body publishes it. You do not need to be certain, and you do not need to be polite about it.
Corrections go ahead of other correspondence, because a wrong page is doing damage for as long as it is up. If we cannot verify a claim at a source we are willing to name, we take the claim down rather than leave it standing with a hedge attached. Where a change is material we say on the page what changed. You can reach us through the contact page, and the wider standard is set out in our editorial policy.
Corrections are for numbers that were wrong. For numbers that are merely old (a study whose scheduled rebuild failed, or whose source stopped answering) see the status page, which reports the freshness of every study on this site in public.