AnswerLoop
How FinePrint organizes legal intelligence for OpenLegalLM.
- 01FinePrint runs the work.
- 02AnswerLoop supplies judgment at the edge.
- 03Every reviewed edge case makes the next answer better.
What it is
The lawyer answers once. The model answers forever.
Legal AI has one problem that a bigger model does not solve. Law is local, it changes, and most of it has never been written down as a rule a model could learn. Somebody has to answer the question the first time.
AnswerLoop is how FinePrint does that. When the counsel meets a question it has not seen before, it does not guess. It stops, sends the question to a licensed lawyer, and pays for the answer. The answer clears your matter, and it goes into OpenLegalLM. The next company that asks does not need the lawyer.
Every question is answered once by a lawyer and forever by the model.
One question, end to end.
Five steps between a question FinePrint has not seen and a question it never needs to ask again.
- Not seen before. An endpoint reaches a question outside what OpenLegalLM has learned. The matter stops. Nothing runs.
- A Review Request. The question goes to a licensed lawyer in the right state and practice area, with the file already built: the record, the documents, the issue.
- An answer. The lawyer answers it: accept, correct, redline, or stop the work. Every answer names the governing reason. The lawyer is paid the same whatever the answer is.
- Back to your matter. The answer clears the matter and the endpoint finishes. You see the attorney, the state, the practice area and the scope. There is no fee on that card.
- Into OpenLegalLM. The answer is tested, versioned and released into the model. Next time, FinePrint knows.
Every answer moves work one lane to the right.
Red to Yellow to Green. The share of work the agents run keeps growing. The share that reaches a lawyer keeps falling.
Red — a lawyer leads.
A financing, a dispute, a regulated question. A specialist takes the matter with the file already built. Some matters stay Red no matter how many times the model has seen one.
Yellow — a lawyer confirms.
The question is outside what OpenLegalLM has learned, so it stops. A licensed attorney answers before anything moves. Included in your plan. That answer is how AnswerLoop teaches the model.
Green — agents run it.
Standard work runs end to end, with your approval. Lawyers score samples of the output. A question that needed a specialist becomes a question a lawyer confirms, then a question the agents run on their own.
Both numbers are measured on real matters and publish together. The flag rate and the novelty rate
The lawyers
Licensed, conflict-checked, paid per answer.
The lawyers who answer are a network of licensed legal professionals. Every one is admitted where the question lives, conflict-checked before a matter is offered, and paid per answer, not per hour.
Three things keep the answers honest
- Pay never depends on the answer. A lawyer who says “stop, this needs a human” is paid the same as one who says “this is fine.”
- The product cannot overrule a lawyer. A flag is cleared by a licensed attorney or not at all.
- The standards are written. Every endpoint carries a review standard a lawyer can read before accepting anything.
And they share in what it earns
- Paid per Review Request. A published rate, set by the request before it is accepted.
- Revenue share on knowledge. Playbooks, positions, clause patterns. When OpenLegalLM uses them, the lawyer who contributed them shares in what it earns.
- Their judgment is the model. Every answer is what OpenLegalLM is made of. Their name builds the trust.
AnswerLoop learns from lawyers’ answers, never from your record.
Sampled review of Green work happens on anonymized artifacts. The company, the people, the counterparties and the amounts come off before a reviewer sees anything.
A Yellow or Red matter is different. The attorney you engage sees your matter, scoped to it, and the engagement closes with it. Your documents are never training material for a shared model.
The questions worth asking.
Does the lawyer answer my question or the model’s?
Yours. The question comes from your matter, and the answer clears it. What the model learns is the general rule, tested and released, without your company in it.
Do I pay for the answer?
No. The lawyer is paid by FinePrint out of the subscription. Your card shows the attorney and the scope, and no fee.
Who decides a question has not been seen before?
OpenLegalLM does, against what it has learned, and the endpoint’s own rules. Some matters go to a lawyer no matter how many times the model has seen one: a financing, a dispute, a regulated question. Those are Red, and they stay Red.
Can I see it working?
Yes. The share of matters that reach a lawyer, and how often lawyers correct the model, publish together with the evaluation that produces them. The numbers
FinePrint runs the work. AnswerLoop supplies judgment at the edge.
Every question is answered once by a lawyer and forever by the model.
Thirty minutes, live. One legal workflow, start to finish, including the moment a lawyer answers.
Book a demoThe model the answers train: seven layers, how an answer becomes a release, and the numbers we publish.
Inside OpenLegalLM