baseline.json

{
  "from_run": "2026-07-12T05:01:17+00:00",
  "mode": "full",
  "offline": false,
  "answer_model": "us.anthropic.claude-haiku-4-5-20251001-v1:0",
  "provenance": {
    "prompt_versions": {
      "system": "v10 2026-07-11 (v9 added two-path disambiguation; v10 requires the exact published criterion before process details on attribute-based eligibility questions and adds explicit Tagalog language/date guidance)",
      "answer_user": "v7 2026-07-11 (v6 bound proof/contact details to the exact class/product; v7 adds the canonical Tagalog snapshot-date disclosure)",
      "judge_groundedness": "v2 2026-07-02 (v1 2026-06-11; v2 admits prior conversation turns as context for multi-turn cases \u2014 conv-004)",
      "judge_helpfulness": "v3 2026-07-02 (v1 2026-06-11; v2 2026-06-12 stopped second-guessing documented policy \u2014 ml-010, fresh-007; v3 threads prior conversation turns and the case rationale so multi-turn answers are not graded blind \u2014 conv-004)"
    },
    "corpus_version": "0938fff0539a"
  },
  "suites": {
    "conversation": {
      "passed": 8,
      "total": 10,
      "pass_rate": 80.0
    },
    "cross_agency": {
      "passed": 3,
      "total": 3,
      "pass_rate": 100.0
    },
    "edge_cases": {
      "passed": 46,
      "total": 48,
      "pass_rate": 95.8
    },
    "freshness": {
      "passed": 9,
      "total": 10,
      "pass_rate": 90.0
    },
    "groundedness": {
      "passed": 27,
      "total": 29,
      "pass_rate": 93.1
    },
    "multilingual": {
      "passed": 22,
      "total": 22,
      "pass_rate": 100.0
    },
    "refusal": {
      "passed": 34,
      "total": 34,
      "pass_rate": 100.0
    },
    "sensitivity": {
      "passed": 28,
      "total": 30,
      "pass_rate": 93.3,
      "pairs_passed": 13,
      "pairs_total": 15
    },
    "stretch_tagalog": {
      "passed": 15,
      "total": 15,
      "pass_rate": 100.0
    }
  }
}