Recipe: score a free-form chat output

Recipe — score chat output. Pointer to the canonical SDK text-evaluation sample.

The canonical sample lives in the SDK: samples/modalities/text_evaluation.py — score text completions across safety, relevance, helpfulness, factual accuracy, and compliance.

Pattern

Build judges with client.judges.create(name=, evaluation_goal=) for the dimensions you care about, then run trace evaluations via client.trace_evaluations.create(trace_id=, judge_id=).

For most chat use-cases, start from the SDK sample and adapt the rubrics to your team's bar. Use Bootstrap a judge before GEPA for the week-1 setup.

See also

• SDK sample: text_evaluation.py
• Tutorial 2: Build your first judge
• Concept: Judges