Every study on this site is built by a scheduled job that reads a public register: NHTSA in the United States, the national vehicle register in the Netherlands, and the national registers we are adding country by country. Nobody types these numbers in by hand.
Which means they can go wrong without anyone noticing. A job can fail. An authority’s API can stop answering. A number that was true on Tuesday can sit on the page all week looking exactly as confident as it did when it was correct.
So this page exists to catch us at it. It reads every engine on the site and reports when each study last rebuilt, how often it is supposed to rebuild, and (where the engine can tell) whether the register it depends on was actually answering. If something is stale, it says stale. If an authority’s API is failing, it names the authority.
Every study on this site rebuilt on schedule and every source answered.
| Study | Last rebuilt | Rebuilds | State |
|---|---|---|---|
| The dated wording archive What every tracked recall notice said, hashed and dated | 1 minutes ago | every 6 hours | current |
| The notification delay study How long between an American recall and the Dutch one | 2 hours ago | every 3 hours | current |
| The cross-border join The same defect, matched across two countries by campaign code | 4 hours ago | daily | current |
| The completion study Recalled cars with no repair recorded against them | 18 hours ago | daily | current |
| The severity index How strongly the same defect is described in each country | 16 hours ago | daily | current |
| The American record, read in bulk NHTSA's whole recall file, read on this server instead of one API call at a time | 14 minutes ago | hourly | current |
| The coverage gap The same model, recalled more often in one country than the other | 44 minutes ago | hourly | current NHTSA responding normally |
Every study here is rebuilt by a scheduled job that reads a public register. This is when each one last rebuilt, and whether its source was answering. A study marked stale is stale: do not treat its numbers as current.
Machine-readable: /wp-json/cmvst/v1/status. Checked 2026-09-17T10:04:52+00:00.
Every surface, checked against one figure
The same fact is published here in several places at once: on this site, in the file an assistant reads, through the public API, and to any program that calls the record as a tool. Every hour each of them is read back and compared with the one value the engine holds. A surface rebuilt on a timer may be an hour behind and is shown catching up; a dated snapshot is meant to be behind. Anything else is a defect, and fails this site’s own audit.
| Surface | Figure | It says | The record holds | |
|---|---|---|---|---|
| The front page a reader lands on | campaigns | 15,259 | 15,259 | agrees |
| The front page a reader lands on | campaigns reread | 1,485 | 1,485 | agrees |
| The front page a reader lands on | rewrites held | 102 | 102 | agrees |
| The front page a reader lands on | campaigns rewritten | 89 | 89 | agrees |
| The dataset page a reader sees | crossing | 2,058 | 2,058 | agrees |
| The dataset page a reader sees | all three | 109 | 109 | agrees |
| llms.txt, what an assistant reads | campaigns | 15,259 | 15,259 | agrees |
| llms.txt, what an assistant reads | notices rewritten | 101 | 101 | agrees |
| llms.txt, what an assistant reads | campaigns reread | 1,485 | 1,485 | agrees |
| llms.txt, what an assistant reads | rewrites held | 102 | 102 | agrees |
| The record as a callable tool | campaigns | 15,259 | 15,259 | agrees |
| The record as a callable tool | uk recalls | 8,965 | 8,965 | agrees |
| The record as a callable tool | crossing | 2,058 | 2,058 | agrees |
| The dated crosswalk snapshot on GitHub | crossing | 2,058 | 2,058 | agrees |
| The dated crosswalk snapshot on GitHub | built | 2026-09-17 (0.4 days old) | a dated snapshot | by design |
Last checked 2026-09-17T09:27:38+00:00. The same check, as JSON.
How to read this
Current means the study rebuilt within four times its promised interval. We are deliberately generous with that threshold: a single missed run is not a scandal, and we would rather under-claim a problem than cry wolf at every hiccup.
Stale means the numbers on that study’s page are old. Treat them as a snapshot of whenever they last rebuilt, not as today’s figure.
A source note (for example that NHTSA is not answering normally) means the engine tried to read the register and could not. When that happens the engine does not record a zero and move on. It stops, leaves the previous reading in place, and waits. An outage that gets written into a dataset as “no recalls found” is worse than no dataset at all, and it is invisible from the outside. That is precisely the failure this page is built to make visible.
Still being read means an engine is part-way through its first pass over a list. Those studies do not publish partial rows (a half-counted table is worse than none), so the page stays unpublished until the count is complete.
Why publish this at all
Most sites that republish public data show you the number and stop. You have no way of knowing whether it was read this morning or last spring, or whether the read succeeded. We think that is the wrong way round: if the entire value of a record is that it is checkable, then the health of the record has to be checkable too.
The same information is available as JSON at /wp-json/cmvst/v1/status, so anyone monitoring this site (a journalist, a regulator, a rival) can watch it without asking us.
If you find a study whose numbers do not match the register it cites, we want to hear about it. Corrections are published on the corrections page, with dates.