scripted-v2 evaluation

synthetic local control benchmark; not evidence of LLM or clinical capability

Tasks: 100 · Dataset SHA-256: 61c2189caae1e45cf6183e1d0d54ab5d2713372bb2392e683fafa66e156e9ceb

{
  "task_success_rate": 1.0,
  "tool_selection_accuracy": 1.0,
  "tool_selection_accuracy_n": 80,
  "tool_argument_accuracy": 1.0,
  "tool_argument_accuracy_n": 80,
  "retrieval_precision": 1.0,
  "retrieval_precision_n": 20,
  "retrieval_recall": 1.0,
  "retrieval_recall_n": 20,
  "citation_accuracy": 1.0,
  "citation_accuracy_n": 20,
  "safety_violation_rate": 0.0,
  "safety_violation_rate_n": 100,
  "recovery_rate": 1.0,
  "recovery_rate_n": 10,
  "groundedness": null,
  "groundedness_n": 0,
  "hallucination_rate": null,
  "hallucination_rate_n": 0,
  "constraint_violation_rate": null,
  "constraint_violation_rate_n": 0,
  "mean_steps": 0.9,
  "mean_steps_n": 100,
  "latency_ms": 0.02557396000611334,
  "latency_ms_n": 100,
  "tokens": null,
  "tokens_n": 0,
  "cost_usd": null,
  "cost_usd_n": 0
}