Grading

How grading works

Verifiers, result rows, and the one score shape every grader emits

Verifiers, result rows, and the one score shape every grader emits.

This page is a stub

Scaffolded as part of the documentation build-out. Content to be written in a later phase — see docs/ASSUMPTIONS.md for the product gaps some pages depend on.

Connecting a model

Point agent-env at any OpenAI-compatible endpoint, provider, or proxy

Deterministic grading

Assert on environment state and run tests inside the container, with no LLM