Skip to main content
Evals grade definition wording and agent behaviour against written rules. Code settles what it can; models judge the rest, and a case no tier settles goes to you. See How it works.
  • Evals: the rules and behaviours that define good definitions and agent work, and how they are graded.
  • Evaluation Report: figures from recorded runs.
  • Methodology: how evals are built, run, and reported.