Evals

Evals are structured tests that measure how well an AI model or agent performs a task, run repeatedly to track quality across versions and prompts.

A good eval suite pairs representative tasks with a grading method — exact match, a rubric, or an LLM judge — so changes can be compared on the same basis.

WorkForce evaluations are a formal, sealed-bank form of evals whose results are published as an AQO score with a confidence interval.