For information only. Not advice, and not to be relied on. What that means, in full
Tools

Scan a model card

Paste any AI system’s documentation and see which of the seven governance dimensions it addresses and which it is silent on. This is the same automated scan that produces the card figures on the leaderboard, run against a document of your choosing instead of one of ours.

One documentpasted, not uploadedD1DocumentationD2Evaluation EvidenceD3SafetyD4FairnessD5PrivacyD6AccountabilityD7Regulatory MappabilitySeven questionsanswered, or silentT0reachableT1reachableT2reachableT3out of reachT4out of reacha document stops hereceiling 50/100
50 out of 100 is the top of this scale, not a bad score. A scan reads one document, and a document the vendor wrote can show they asserted something and published an artifact saying so. It cannot show that anyone independent reproduced or verified it — that needs somebody other than the vendor, which is what the upper half of the scale is for. So read a card result as how completely the document answers what a document can answer. It is not a measure of the system, and a low one is a statement about the page you pasted.
It runs in your browser and nothing is uploaded. Not a promise about our retention policy — a property of the page. There is no endpoint here that accepts a document, so there is nowhere for the text to go. Watch your network tab while you press the button.
0 characters

This runs in your browser. The text is not uploaded, because there is nothing here to upload it to — no request is made when you press the button, which you can confirm in your browser’s network tab.

What this measures, and what it does not

It reads a document, not a system

Every one of the 25 checks answers a question about the text — does it state prohibited uses? does it name an evaluation dataset? — and none of them measures whether the system is any good. A thorough card on a poor model scores well, and it should: what is being measured is what was written down.

It is capped at T2

A card is a vendor-published artifact by definition, so nothing read from one can reach T3 (independently reproducible) or T4 (independently verified). The composite therefore cannot exceed 50, and the percentage above is a percentage of that ceiling — not of a full assessment. A card scan and a GRI assessment are different scales and do not compare.

It scores under the general profile only

A GRI without its profile is meaningless, and weighting profiles live in the database, which a page running in your browser cannot reach. So this uses flat weighting — every dimension equal — which is what the leaderboard uses for card scans too. The method sets out the profiles that reorder a ranking.

A low score is often a real finding

Across the 7 dimensions, the cards in this registry are routinely silent on fairness and on regulatory mapping. If your document scores badly there, it is in the majority — which is the argument for fixing it, not for discounting the measurement.

What this number is. It measures the documentation located about this system at a stated source on a stated date, against a published method — not the system itself. A low score means the located documentation is thin; it is not a finding that anyone governs their models poorly, that a system is unsafe, or that any law or rule has been broken. Documentation held privately or published elsewhere is not counted, because this registry cannot see it. It is published for information only, describes documentation as it stood on a stated date, and is not advice — verify anything you intend to act on. Terms, corrections and right of reply →

If you want the version with a person behind it

This scan is automated and reads only what you paste. A full assessment locates the documents itself, records where each claim came from and when it was read, fingerprints the document it scored, and can reach evidence tiers a card never can. That is work somebody does, and it is priced.

Ask for an assessment

Which system, and what decision is it feeding into?

Your address is used to reply and nothing else. It is not published, not sold, and not added to a mailing list.

Scan a model card — GovernanceHub