01
Data.
Where the legal knowledge came from, whether we hold the rights to it, and whether it is current.
The Legal AI Trust Score
We measure trust before we ask you to rely on it.
Every kind of legal work FinePrint does is scored by independent lawyers whose pay never depends on the score. The scores publish with their method. Like a credit rating, for legal AI.
Three things it measures.
02
Generation.
How often independent lawyers correct the AI’s work on real matters (the flag rate), and how often the AI meets a question it has not seen and stops (the novelty rate).
03
Compliance.
Whether the work stayed inside the playbook, the jurisdiction’s rules and the approval path.
Who sets it
Who sets it.
Lawyers in AnswerLoop, on real matters. Paid for the answer, never for approving it, and never for the score. Reviewer standing is weighted by practice area and jurisdiction. Disagreement goes to a second lawyer. A flag stands until a licensed attorney clears it; the software cannot override it.
How it is published
How it is published.
Per endpoint, with the sampling method, the categories and jurisdictions in scope, how reviewer standing is weighted, and how disagreement is resolved. We do not publish private benchmarks or exam scores. The first scores publish with the evaluation that produces them.
Built to rise
Built to rise.
Every lawyer’s correction feeds the model. A new release has to beat the last one on every score, or it does not ship. A candidate that drafts better and escalates worse does not ship.
Trusted skills
What a trusted skill is.
An endpoint becomes a trusted skill when it clears its Trust Score. 41 are built. The catalog is 101.
Inside FinePrint Enterprise
Inside FinePrint Enterprise.
On your matters, your own lawyers’ scores stay inside your boundary. They measure your model, FinePrint LM Enterprise, against your playbook.