Skip to main content
POST
Create an evaluator, or change one of your project’s. Try it on a logged call with test_evaluator, then use its id in evaluators_impacting_benchmark or evaluators_for_reference of set_benchmark.

Parameters

Passed in the request body.
integer
The evaluator to change. Omit it to create one. A built-in evaluator cannot be changed.
string
The evaluator’s name. Required to create one.
string
“judge” (an AI compares the candidate’s answer with the baseline’s by following your instruction in words) or “deterministic” (Python code checks the answers). Required to create one.
string
A judge’s instruction in words, or a deterministic check’s Python code. Required to create one.
boolean
Add the evaluator to your library, which the Evaluators page lists and every benchmark can pick from. Defaults to false on creation: the person adds it from the dashboard once they have checked it.
string
The email of the member of your organization this is for, recorded as the evaluator’s author when it is created. Defaults to the email of the person who created the API key, or for a key without one of the project’s creator. The response’s recorded_for_email gives the email recorded: tell the person, and send this to record another.
The response also carries blockers (what prevents the usual next step), in plain words. It is empty when the usual workflow applies.