Expectations before answers
What did you expect the answer to be?
No AI telling you you're right. No AI telling you the other AI is right. You set the expectations. The interpreter checks the work.
No AI grading AI. The interpreter does the math.
Did you get what you asked for? Now you know.
Receipts catches what went wrong. Agreements defines what right looks like.
The supervisor writes the expectation. The junior staff collects the evidence. The system checks the work. The receipt shows what passed.
It judges the AI, not you. Every failure in an agreement is the model's mistake, verified against the expectation you set. The agreement protects the person doing the work, not the system that produced it.
Describe what you need. Review the agreement. Submit the evidence. Read the decision.
Describe what you need
In plain language. What are you checking? What does right look like? Who's responsible? You don't write code — you describe your requirements.
Review the agreement
We assemble the Liminate agreement from your description. You read each clause — the syntax and a plain-language explanation — and confirm it's right before anything runs.
Submit the evidence
The junior staff, the analyst, or the AI agent provides the numbers, the citations, the source. They don't need to interpret policy from scratch. The readable rule is on screen.
Read the decision
Passed. Blocked. Needs review. The interpreter ran the agreement against the evidence. Every clause has a gate. The receipt is already generated.
Pay per decision
A decision is a verified outcome — evidence checked against your agreement by the interpreter. Save decisions to keep the proof.
Receipts catches what went wrong. Agreements defines what right looks like — before the first answer comes back.