{
  "acceptance_version": "0.1",
  "object_path": "examples/invacost-invasion-costs",
  "manifest": "agent/tools.json",
  "published_results_path": "results.json",
  "notes": "The three published comparisons are three calls to one tool, which is the point: the object reports three applications of a comparison that generalises to any two levels of any column.",
  "calls": [
    {
      "call_id": "dataset-published",
      "tool": "dataset_summary",
      "args": {},
      "expect": {"kind": "published_pointer", "result_path": "/dataset"},
      "rel_tolerance": 1e-6,
      "abs_tolerance": 1e-9,
      "enabled": true
    },
    {
      "call_id": "t1-observed-vs-potential-published",
      "tool": "compare_groups",
      "args": {"column": "Implementation", "group1": "Observed", "group2": "Potential", "label": "Observed vs Potential"},
      "expect": {"kind": "published_pointer", "result_path": "/T1_observed_vs_potential"},
      "rel_tolerance": 1e-6,
      "abs_tolerance": 1e-9,
      "enabled": true
    },
    {
      "call_id": "t2-terrestrial-vs-aquatic-published",
      "tool": "compare_groups",
      "args": {"column": "Environment", "group1": "Terrestrial", "group2": "Aquatic", "label": "Terrestrial vs Aquatic"},
      "expect": {"kind": "published_pointer", "result_path": "/T2_terrestrial_vs_aquatic"},
      "rel_tolerance": 1e-6,
      "abs_tolerance": 1e-9,
      "enabled": true
    },
    {
      "call_id": "t3-damage-vs-management-published",
      "tool": "compare_groups",
      "args": {"column": "Type_of_cost_merged", "group1": "Damage", "group2": "Management", "label": "Damage vs Management"},
      "expect": {"kind": "published_pointer", "result_path": "/T3_damage_vs_management"},
      "rel_tolerance": 1e-6,
      "abs_tolerance": 1e-9,
      "enabled": true
    },
    {
      "call_id": "reliability-confound-off-reference",
      "tool": "compare_groups",
      "args": {"column": "Method_reliability", "group1": "High", "group2": "Low"},
      "expect": {
        "kind": "literal",
        "value": {
          "test": "High vs Low",
          "group1": "High",
          "group2": "Low",
          "n1": 12268,
          "n2": 1076,
          "U": 3702078.0,
          "p": 1.918720514251824e-126,
          "rank_biserial_r": 0.43909472826818163,
          "median1_usd": 65712.99870835856,
          "median2_usd": 2624671.1351623368,
          "median_ratio_1_over_2": 0.025036659956369806,
          "median_ratio_ci95": [0.017470352355541324, 0.033979438072641754]
        }
      },
      "rel_tolerance": 1e-6,
      "abs_tolerance": 1e-9,
      "enabled": true,
      "notes": "A comparison the published run never makes, and the most informative call in this file. High-reliability records carry a median cost 40x LOWER than low-reliability ones (ratio 0.025), with the same effect-size magnitude (r = 0.44) as the headline observed-vs-potential result (r = 0.45). Cost magnitude tracks record reliability about as strongly as it tracks whether the cost was observed, so the T1 contrast is partly a data-quality contrast. This qualifies claim 1 and belongs in Usage notes."
    },
    {
      "call_id": "bootstrap-stability-off-reference",
      "tool": "compare_groups",
      "args": {"column": "Implementation", "group1": "Observed", "group2": "Potential", "label": "Observed vs Potential", "bootstrap_n": 12000, "seed": 7},
      "select": "/median_ratio_ci95",
      "expect": {"kind": "literal", "value": [0.013474096697916755, 0.020847237439229185]},
      "rel_tolerance": 1e-6,
      "abs_tolerance": 1e-9,
      "enabled": true,
      "notes": "Four times the resamples on a different seed moves the published CI [0.01337, 0.02075] by under 1% of its width. The interval is a property of the data, not of the 3000-resample default."
    },
    {
      "call_id": "multiplicity-published",
      "tool": "multiplicity",
      "args": {},
      "expect": {"kind": "published_pointer", "result_path": "/multiplicity"},
      "rel_tolerance": 1e-6,
      "abs_tolerance": 1e-9,
      "enabled": true
    },
    {
      "call_id": "multiplicity-divergent-family-off-reference",
      "tool": "multiplicity",
      "args": {"p_values": [0.01, 0.04, 0.2]},
      "select": "/p2",
      "expect": {"kind": "literal", "value": {"raw_p": 0.04, "bh_p": 0.06, "holm_p": 0.08}},
      "rel_tolerance": 1e-6,
      "abs_tolerance": 1e-9,
      "enabled": true,
      "notes": "A family where BH and Holm diverge. The published family is so extreme that both corrections are indistinguishable from the raw p-values, so it never exercises the difference."
    }
  ]
}