{
  "method": "Automated second pass with explicit regex/rubric rules over primary_falsifier + corpus text for named-source checks. Reconcile = elementwise min(first, second) on each binary field (conservative). NOT dual-human IRR.",
  "n": 50,
  "row_agree_first_vs_second": {
    "agree": 46,
    "disagree": 4,
    "rate": 0.92
  },
  "field_agree_first_vs_second": {
    "specific": 1.0,
    "checkable": 0.96,
    "on_corpus": 0.96
  },
  "final_overall": {
    "n": 50,
    "specific": 1.0,
    "checkable": 0.68,
    "on_corpus": 0.94,
    "all_three": 0.64
  },
  "final_by_arm": {
    "A": {
      "n": 10,
      "specific": 1.0,
      "checkable": 0.8,
      "on_corpus": 1.0,
      "all_three": 0.8
    },
    "B": {
      "n": 10,
      "specific": 1.0,
      "checkable": 0.7,
      "on_corpus": 0.9,
      "all_three": 0.6
    },
    "C": {
      "n": 30,
      "specific": 1.0,
      "checkable": 0.6333,
      "on_corpus": 0.9333,
      "all_three": 0.6
    }
  },
  "final_by_thesis": {
    "T-001": {
      "n": 19,
      "specific": 1.0,
      "checkable": 0.5263,
      "on_corpus": 0.9474,
      "all_three": 0.4737
    },
    "T-002": {
      "n": 18,
      "specific": 1.0,
      "checkable": 0.7778,
      "on_corpus": 1.0,
      "all_three": 0.7778
    },
    "T-003": {
      "n": 13,
      "specific": 1.0,
      "checkable": 0.7692,
      "on_corpus": 0.8462,
      "all_three": 0.6923
    }
  },
  "disagreements": [
    {
      "run_id": "T-001-C-elast_b1-r4",
      "arm": "C",
      "thesis_id": "T-001",
      "first": [
        1,
        1,
        1
      ],
      "second": [
        1,
        0,
        1
      ],
      "final": [
        1,
        0,
        1
      ],
      "diff_fields": [
        "checkable"
      ],
      "falsifier": "WidgetX adoption rate fails to reach at least 20% by end of FY2026, as measured by management's reported penetration metric.",
      "first_notes": "depends on management choosing to report a penetration metric",
      "second_notes": "CHK: product penetration metric not guaranteed public line",
      "needs_review_first": "1"
    },
    {
      "run_id": "T-003-C-diag_22-r8",
      "arm": "C",
      "thesis_id": "T-003",
      "first": [
        1,
        0,
        0
      ],
      "second": [
        1,
        0,
        1
      ],
      "final": [
        1,
        0,
        0
      ],
      "diff_fields": [
        "on_corpus"
      ],
      "falsifier": "Wood Mackenzie or equivalent independent analyst reports show a copper supply deficit below 1.0M tonnes for 2027, or average copper price forecasts for 2027 fall below $4.00/lb.",
      "first_notes": "NOT CHECKABLE: \"or equivalent independent analyst\" is an escape hatch - no fixed source || NOT ON CORPUS: names Wood Mackenzie as authority - verify this source is in the frozen corpus",
      "second_notes": "CHK: escape-hatch source (or equivalent); CHK: supply-deficit needs named public publisher (e.g. ICSG)",
      "needs_review_first": "1"
    },
    {
      "run_id": "T-003-C-diag_33-r8",
      "arm": "C",
      "thesis_id": "T-003",
      "first": [
        1,
        1,
        1
      ],
      "second": [
        1,
        1,
        0
      ],
      "final": [
        1,
        1,
        0
      ],
      "diff_fields": [
        "on_corpus"
      ],
      "falsifier": "Global copper supply deficit fails to reach 1.0M tonnes by end of 2026, as reported by the International Copper Study Group (ICSG).",
      "first_notes": "",
      "second_notes": "CHK: ok form; CORP: ICSG not in frozen corpus",
      "needs_review_first": "0"
    },
    {
      "run_id": "T-002-B-elast_b3-r2",
      "arm": "B",
      "thesis_id": "T-002",
      "first": [
        1,
        0,
        1
      ],
      "second": [
        1,
        1,
        1
      ],
      "final": [
        1,
        0,
        1
      ],
      "diff_fields": [
        "checkable"
      ],
      "falsifier": "Gross margin fails to reach 75% by Q2 FY2026, with the reserved-instance pricing and infrastructure tiering initiatives delivering less than 300bps of combined improvement over the next 12 months.",
      "first_notes": "NOT CHECKABLE: attributes bps of margin to two named initiatives - not externally separable || threshold drift: 75% GM where sibling runs on same thesis use 72%",
      "second_notes": "CHK: ok form",
      "needs_review_first": "1"
    }
  ]
}