Projects / high risk
Support Bot — Refund Policy
Customer-support assistant for a D2C brand. Answers refund and return questions from the policy document.
Targets — what's being tested
- Refund Support Prompt2 version(s)
Datasets — what good looks like
- Refund policy — core scenarios8 case(s)
Evaluators — the quality policy
- Required concepts presentrequired_concepts
- No forbidden contentforbidden_conceptsgate
- Valid JSON (json-tagged cases)json_valid
- JSON schema fields (json-tagged cases)json_schema
- Policy correctness (LLM judge)llm_judgegate
- Overall quality (pairwise)pairwise
Add evaluator
Runs
- random 1completed