Evals Eval Run Pairwise Example

{
  "id": "run_pairwise_chatbot_arena_8421",
  "suite_id": "suite_pairwise_helpfulness_v1",
  "case_id": "case_pairwise_8421",
  "experiment_id": "exp_2026_05_22_pairwise_opus_vs_gpt5",
  "model": {
    "provider": "anthropic",
    "name": "claude-opus-4-7"
  },
  "prompt": "Explain the difference between supervised and reinforcement learning to a high-school student.",
  "output": "Supervised learning is like studying with an answer key — for every question, you already know the correct answer and the model learns to match it. Reinforcement learning is like learning a sport — the model tries things, gets rewards or penalties, and gradually figures out what works.",
  "scorer": {
    "id": "scorer_pairwise_helpfulness",
    "name": "pairwise-helpfulness",
    "type": "pairwise"
  },
  "score": 1.0,
  "label": "A_BETTER",
  "evidence": {
    "rationale": "Output A provides a clearer concrete analogy and is more age-appropriate than Output B, which uses unexplained jargon like 'policy gradients.'",
    "judge_model": "gpt-5",
    "trace_id": "trace_pairwise_8421"
  },
  "metrics": {
    "latency_ms": 2210,
    "cost_usd": 0.0117
  },
  "tags": ["pairwise", "helpfulness", "education"],
  "timestamp": "2026-05-22T15:51:38Z"
}