Endpoints

/data/models.json: All 35 models with every field, each carrying its own source_url, quote, confidence and checked date. CORS is open, so you can call it from a browser.
/data/models.csv: The same data flattened for a spreadsheet. Each claim column is followed by its own __source and __checked columns, so a pasted row still shows where each answer came from.
/rss.xml: Changes only — new publications, withdrawals and revisions as they are recorded.

Using it without misrepresenting it

Two rules, and they are the same ones we hold ourselves to.

Keep the date with the claim. Every value is a reading taken on a particular day. Stripped of checked, a dated finding becomes a permanent unqualified assertion about a named company, which is not what we published.

Do not read a negative as non-compliance. A false means we did not find the document at the sources we monitor. Many models are inside a transitional period and are not yet required to publish anything, and some sources refuse us outright — which is why every record carries blocked and model_level_sources. If both suggest we could not see properly, the honest reading is that we do not know.

Stability

Field keys and slugs are stable; we add fields rather than rename them. If something must change incompatibly, the old shape stays served alongside the new one and the change is announced in the changelog first.

The files rebuild whenever the register does — currently weekly, and immediately after any correction.

Licence

CC BY 4.0. Use it commercially, redistribute it, build on it. Attribute trainedon.org and keep the check dates. If you are doing something interesting with it we would like to hear about it, and if you find us wrong we would much rather hear about that: hello@trainedon.org.

No rate limit is applied to these files. They are static and cached at the edge; please do not poll them more than once a minute, because there is nothing to gain — the data changes weekly.