Evaluate your simulation.
Jean tests whether your model can stand in for people. Send its predictions for the items in a suite and get back a score for how well it simulates an audience, a customer base, or one person at a time, measured with open industry methods and our own. Detect the weaknesses and gaps in your model before your users do.
Every score is computed against ground truth the model has never seen: held-out human data that Jean keeps and never returns, read against baselines so the number means something. The result carries the score, the count it was computed over, and a trace of how it got there.
Quickstart
From no account to a score, in five curl commands.
Evals
What a suite is, what the number means, and how it is pinned.
API reference
Authentication, errors, limits, and every endpoint.
OpenAPI
The contract itself. These pages are generated from it.
Beyond the score
A number tells you where a model falls short. For more comprehensive evals, expert grading and validation, or data to fill the gaps a score reveals, talk to us.