Core concepts
Evaluations & scoring
Attach a rubric to a persona and every conversation is scored against it, with evidence quoted from the transcript.
Scores range 1–4 per criterion. Each score is accompanied by the transcript excerpts that justify it, so reviewers can audit every number. Scoring runs on Claude by default.
json
"rubric": ["Structure", "Evidence & metrics", "Ownership", "Clarity"]Evaluations support decisions made by people. Don't use them as the sole basis for consequential decisions.