Evals Scorer Example

{
  "id": "scorer_faithfulness_v2",
  "name": "faithfulness",
  "description": "Measures whether the answer is grounded in the retrieved context (RAG faithfulness). Implemented as an LLM-as-a-judge call against a strict rubric.",
  "type": "llm_judge",
  "implementation": "judge:judge_groundedness_v1",
  "judge_model": "gpt-5",
  "judge_prompt": "Score how well the answer is grounded in the retrieved context on a 0.0-1.0 scale. Penalize unsupported claims.",
  "scale": {
    "min": 0,
    "max": 1,
    "step": 0.01,
    "kind": "continuous"
  },
  "threshold": 0.8,
  "rubric": "1.0 = every claim supported. 0.5 = mostly supported with minor invented additions. 0.0 = contradicts or invents content.",
  "tags": ["rag", "groundedness", "faithfulness"]
}