{
  "schema_version": 1,
  "columns": [
    "family",
    "baseline",
    "comparison",
    "metric",
    "source_swap_p",
    "mcnemar_exact_independence_diagnostic_p",
    "interpretation",
    "holm_p_all_18_tests"
  ],
  "rows": [
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite MINIMAL corrected",
      "comparison": "Flash-Lite LOW",
      "metric": "species_accuracy",
      "source_swap_p": "0.29957004299570045",
      "mcnemar_exact_independence_diagnostic_p": "0.3018724904870079",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite MINIMAL corrected",
      "comparison": "Flash-Lite LOW",
      "metric": "adjusted_tis_pct",
      "source_swap_p": "0.8756124387561244",
      "mcnemar_exact_independence_diagnostic_p": "",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite MINIMAL corrected",
      "comparison": "Flash-Lite MEDIUM",
      "metric": "species_accuracy",
      "source_swap_p": "0.00029997000299970003",
      "mcnemar_exact_independence_diagnostic_p": "7.808423662658282e-05",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "0.0044995500449955"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite MINIMAL corrected",
      "comparison": "Flash-Lite MEDIUM",
      "metric": "adjusted_tis_pct",
      "source_swap_p": "0.2622737726227377",
      "mcnemar_exact_independence_diagnostic_p": "",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite MINIMAL corrected",
      "comparison": "Flash-Lite HIGH",
      "metric": "species_accuracy",
      "source_swap_p": "9.999000099990002e-05",
      "mcnemar_exact_independence_diagnostic_p": "2.736780589985052e-06",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "0.0017998200179982003"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite MINIMAL corrected",
      "comparison": "Flash-Lite HIGH",
      "metric": "adjusted_tis_pct",
      "source_swap_p": "0.9036096390360964",
      "mcnemar_exact_independence_diagnostic_p": "",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite LOW",
      "comparison": "Flash-Lite MEDIUM",
      "metric": "species_accuracy",
      "source_swap_p": "0.000999900009999",
      "mcnemar_exact_independence_diagnostic_p": "0.001016220951347889",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "0.012998700129987"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite LOW",
      "comparison": "Flash-Lite MEDIUM",
      "metric": "adjusted_tis_pct",
      "source_swap_p": "0.1792820717928207",
      "mcnemar_exact_independence_diagnostic_p": "",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite LOW",
      "comparison": "Flash-Lite HIGH",
      "metric": "species_accuracy",
      "source_swap_p": "0.00029997000299970003",
      "mcnemar_exact_independence_diagnostic_p": "0.00018314322488235366",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "0.0044995500449955"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite LOW",
      "comparison": "Flash-Lite HIGH",
      "metric": "adjusted_tis_pct",
      "source_swap_p": "0.7742225777422258",
      "mcnemar_exact_independence_diagnostic_p": "",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite MEDIUM",
      "comparison": "Flash-Lite HIGH",
      "metric": "species_accuracy",
      "source_swap_p": "0.48705129487051296",
      "mcnemar_exact_independence_diagnostic_p": "0.48278855684402133",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Flash-Lite",
      "baseline": "Flash-Lite MEDIUM",
      "comparison": "Flash-Lite HIGH",
      "metric": "adjusted_tis_pct",
      "source_swap_p": "0.265973402659734",
      "mcnemar_exact_independence_diagnostic_p": "",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Gemini 3.8",
      "baseline": "Gemini 3.8 HIGH media / MEDIUM thinking",
      "comparison": "Gemini 3.8 HIGH media / LOW thinking",
      "metric": "species_accuracy",
      "source_swap_p": "0.896010398960104",
      "mcnemar_exact_independence_diagnostic_p": "0.8974218269914307",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Gemini 3.8",
      "baseline": "Gemini 3.8 HIGH media / MEDIUM thinking",
      "comparison": "Gemini 3.8 HIGH media / LOW thinking",
      "metric": "adjusted_tis_pct",
      "source_swap_p": "0.8847115288471152",
      "mcnemar_exact_independence_diagnostic_p": "",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Gemini 3.8",
      "baseline": "Gemini 3.8 HIGH media / MEDIUM thinking",
      "comparison": "Gemini 3.8 HIGH media / HIGH thinking",
      "metric": "species_accuracy",
      "source_swap_p": "9.999000099990002e-05",
      "mcnemar_exact_independence_diagnostic_p": "1.4450157667486737e-10",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "0.0017998200179982003"
    },
    {
      "family": "Gemini 3.8",
      "baseline": "Gemini 3.8 HIGH media / MEDIUM thinking",
      "comparison": "Gemini 3.8 HIGH media / HIGH thinking",
      "metric": "adjusted_tis_pct",
      "source_swap_p": "0.42855714428557146",
      "mcnemar_exact_independence_diagnostic_p": "",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    },
    {
      "family": "Gemini 3.8",
      "baseline": "Gemini 3.8 HIGH media / LOW thinking",
      "comparison": "Gemini 3.8 HIGH media / HIGH thinking",
      "metric": "species_accuracy",
      "source_swap_p": "9.999000099990002e-05",
      "mcnemar_exact_independence_diagnostic_p": "1.1915233414863414e-09",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "0.0017998200179982003"
    },
    {
      "family": "Gemini 3.8",
      "baseline": "Gemini 3.8 HIGH media / LOW thinking",
      "comparison": "Gemini 3.8 HIGH media / HIGH thinking",
      "metric": "adjusted_tis_pct",
      "source_swap_p": "0.36976302369763026",
      "mcnemar_exact_independence_diagnostic_p": "",
      "interpretation": "exploratory; source-level model-label exchangeability required",
      "holm_p_all_18_tests": "1.0"
    }
  ],
  "unit_policy": {
    "table": "paired_tests.csv",
    "metric_units": {
      "adjusted_tis_pct": "percent_of_five_point_maximum",
      "species_accuracy": "proportion_0_to_1"
    },
    "counts": {},
    "interval": "ci_low and ci_high use the same units as estimate; pointwise 95% when interval_status is ok",
    "p_values": "probability_0_to_1; named Holm columns are adjusted p-values, not interval adjustments",
    "estimate": "not_applicable"
  },
  "note": "Saved aggregate candidate analysis; intervals are pointwise and exploratory tests have separate Holm-adjusted p-values."
}