{
  "acceptance_version": "0.1",
  "object_path": "examples/american-rivers-dam-removal-database",
  "manifest": "agent/tools.json",
  "published_results_path": "results.json",
  "notes": "Every published unit is reproduced at the defaults, cross-linked to its recipe. The off-reference calls exercise the two data-handling choices main() hard-codes and never tested -- dropping zero-mile removals, and reading numbers out of free text, which copies a project's total miles onto every dam in it -- plus the bootstrap and winsorization settings. The height null (claim 1) and the height half of claim 3 hold under every alternative. Claim 2's 'survives Holm-Bonferroni correction' does not: withholding only the 11 records whose outcome is a total shared between two to four dams takes length's Holm p from 0.018 to 0.074 (raw p 0.037), and also counting the zero-mile removals takes the raw p to 0.077 with a bootstrap CI that spans zero. The length association is positive under every handling, but its significance after correction rests on counting a shared project total once per dam, which claim 2 should disclose.",
  "calls": [
    {
      "call_id": "records-published",
      "tool": "dataset_summary",
      "args": {},
      "select": "/n_records_total",
      "expect": {
        "kind": "published_pointer",
        "result_path": "/meta/n_records_total"
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "recipe_id": "american-rivers-dam-removal-records",
      "enabled": true
    },
    {
      "call_id": "variable-summary-published",
      "tool": "dataset_summary",
      "args": {},
      "select": "/variable_summary",
      "expect": {
        "kind": "published_pointer",
        "result_path": "/variable_summary"
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "tbl-3. main() builds this in an inline loop; the tool re-implements it and this call proves the re-implementation."
    },
    {
      "call_id": "data-handling-exposure",
      "tool": "dataset_summary",
      "args": {},
      "select": "/data_handling",
      "expect": {
        "kind": "literal",
        "value": {
          "n_zero_miles_records": 5,
          "zero_miles_records_admitted": {
            "height_vs_us": 5,
            "length_vs_us": 4
          },
          "cells_read_from_text": {
            "Dam_Height_ft": 36,
            "Dam_Length_ft": 17,
            "Miles_Restored_US": 24
          },
          "population_records_read_from_text": {
            "height_vs_us": 24,
            "length_vs_us": 26
          },
          "outcome_cells_shared_between_dams": 12,
          "population_records_shared_total": {
            "height_vs_us": 11,
            "length_vs_us": 11
          }
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "The counts behind the two exposed parameters, so the notes below are checked rather than asserted. 5 removals are recorded as reconnecting zero miles, which the outcome > 0 rule drops; include_zero_miles admits all 5 to the height population and 4 to the length population (the fifth has no recorded length). analysis.to_num reads a number out of free text in 36 height, 17 length and 24 outcome cells; inside the published populations 24 of 420 height records and 26 of 404 length records carry such a cell, and those are what bare_only withholds. 12 of the 24 outcome cells state a total shared between dams or projects -- '400 total between 4 dams' (the four Klamath dams), '7 miles total for all 4 dams' (four dams), '175 (total between both projects)' (two) and '10 (between the two removals)' (two) -- and to_num copies each onto every dam that carries it. 11 of those records sit in each published population (the twelfth has no recorded height or length), and those are what no_shared_totals withholds."
    },
    {
      "call_id": "t1-height-published",
      "tool": "primary_tests",
      "args": {},
      "select": "/height_vs_us",
      "expect": {
        "kind": "published_pointer",
        "result_path": "/primary/height_vs_us"
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "recipe_id": "american-rivers-dam-removal-height-vs-restored",
      "enabled": true
    },
    {
      "call_id": "t2-length-published",
      "tool": "primary_tests",
      "args": {},
      "select": "/length_vs_us",
      "expect": {
        "kind": "published_pointer",
        "result_path": "/primary/length_vs_us"
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "recipe_id": "american-rivers-dam-removal-length-vs-restored",
      "enabled": true
    },
    {
      "call_id": "multiplicity-published",
      "tool": "primary_tests",
      "args": {},
      "select": "/multiple_comparison",
      "expect": {
        "kind": "published_pointer",
        "result_path": "/primary/multiple_comparison"
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "recipe_id": "american-rivers-dam-removal-multiplicity",
      "enabled": true
    },
    {
      "call_id": "both-data-handling-off-reference",
      "tool": "primary_tests",
      "args": {
        "include_zero_miles": true,
        "coercion": "bare_only"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "height_vs_us": {
            "test": "Spearman rank correlation",
            "predictor": "Dam_Height_ft",
            "outcome": "Miles_Restored_US",
            "rho": 0.029242855191400085,
            "p_raw": 0.5592979681527466,
            "p_holm": 0.5592979681527466,
            "ci95": [
              -0.0705175898836526,
              0.12944578581452304
            ],
            "n": 401
          },
          "length_vs_us": {
            "test": "Spearman rank correlation",
            "predictor": "Dam_Length_ft",
            "outcome": "Miles_Restored_US",
            "rho": 0.0986097139432256,
            "p_raw": 0.05414326257261831,
            "p_holm": 0.10828652514523662,
            "ci95": [
              1.1209541698832036e-05,
              0.19742234553373778
            ],
            "n": 382
          },
          "multiple_comparison": "Holm-Bonferroni across the two primary tests"
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "Zero-mile removals counted and bare_only coercion together: length rho = 0.099, raw p = 0.054, Holm p = 0.108, n = 382, with the 95% bootstrap CI's lower bound at 0.00001. bare_only is the strictest coercion, not the targeted one: beyond the 11 shared project totals it withholds bounds and 9 ordinary single-dam values written with a unit, '~' or annotation, so this call bounds how far text handling can move the result rather than grounding the qualification of claim 2 -- the shared-totals calls do that. Height is null (rho = 0.029, p = 0.56), so claim 1 is unaffected."
    },
    {
      "call_id": "include-zero-miles-off-reference",
      "tool": "primary_tests",
      "args": {
        "include_zero_miles": true
      },
      "expect": {
        "kind": "literal",
        "value": {
          "height_vs_us": {
            "test": "Spearman rank correlation",
            "predictor": "Dam_Height_ft",
            "outcome": "Miles_Restored_US",
            "rho": 0.05133779258080823,
            "p_raw": 0.2909978690714106,
            "p_holm": 0.2909978690714106,
            "ci95": [
              -0.04580701446410111,
              0.14905697068825394
            ],
            "n": 425
          },
          "length_vs_us": {
            "test": "Spearman rank correlation",
            "predictor": "Dam_Length_ft",
            "outcome": "Miles_Restored_US",
            "rho": 0.11343149734858371,
            "p_raw": 0.0219301779017342,
            "p_holm": 0.0438603558034684,
            "ci95": [
              0.01524372163114479,
              0.21081668999104608
            ],
            "n": 408
          },
          "multiple_comparison": "Holm-Bonferroni across the two primary tests"
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "Requiring Miles_Restored_US > 0 selects on the outcome. Of the 5 zero-mile removals, 4 have a recorded length; counting them moves length from rho = 0.129 (Holm p = 0.018, n = 404) to 0.113 (raw p = 0.022, Holm p = 0.044, n = 408). It still survives correction, but four records take most of the margin under 0.05. Height, which admits all 5 (n = 425), stays null (rho = 0.051, p = 0.29)."
    },
    {
      "call_id": "bare-only-coercion-off-reference",
      "tool": "primary_tests",
      "args": {
        "coercion": "bare_only"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "height_vs_us": {
            "test": "Spearman rank correlation",
            "predictor": "Dam_Height_ft",
            "outcome": "Miles_Restored_US",
            "rho": 0.023862928530455338,
            "p_raw": 0.635904593062911,
            "p_holm": 0.635904593062911,
            "ci95": [
              -0.07652490413818676,
              0.12397020331579882
            ],
            "n": 396
          },
          "length_vs_us": {
            "test": "Spearman rank correlation",
            "predictor": "Dam_Length_ft",
            "outcome": "Miles_Restored_US",
            "rho": 0.11628722262066181,
            "p_raw": 0.02375491233316584,
            "p_holm": 0.04750982466633168,
            "ci95": [
              0.015731619286582597,
              0.21480711878846054
            ],
            "n": 378
          },
          "multiple_comparison": "Holm-Bonferroni across the two primary tests"
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "Admitting only cells that are a plain number withholds the 24 height-population and 26 length-population records whose values to_num picked out of text. Those are the 11 shared project totals plus, in the length population, bounds ('<1' three times, '>20 river miles', '50+'), one single dam's total and 9 ordinary single-dam values such as '1.43 miles', '~70' and '89 (mainstem only)'. Length: rho = 0.116, raw p = 0.024, Holm p = 0.0475, n = 378 -- under 0.05 by 0.0025. That is a smaller move than the targeted no_shared_totals makes (Holm p = 0.074, n = 393, a population containing all 378 of these): the 15 further records bare_only withholds pull the coefficient back up, from 0.105 to 0.116. Height moves toward zero (rho = 0.024, p = 0.64)."
    },
    {
      "call_id": "shared-totals-off-reference",
      "tool": "primary_tests",
      "args": {
        "coercion": "no_shared_totals"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "height_vs_us": {
            "test": "Spearman rank correlation",
            "predictor": "Dam_Height_ft",
            "outcome": "Miles_Restored_US",
            "rho": 0.026289598978138165,
            "p_raw": 0.5960157951192058,
            "p_holm": 0.5960157951192058,
            "ci95": [
              -0.07325650196779959,
              0.12613912085133527
            ],
            "n": 409
          },
          "length_vs_us": {
            "test": "Spearman rank correlation",
            "predictor": "Dam_Length_ft",
            "outcome": "Miles_Restored_US",
            "rho": 0.10533150469052831,
            "p_raw": 0.03686338188701506,
            "p_holm": 0.07372676377403012,
            "ci95": [
              0.006637075268843713,
              0.2042611114357016
            ],
            "n": 393
          },
          "multiple_comparison": "Holm-Bonferroni across the two primary tests"
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "The targeted fix for to_num's one clear error: withhold only the 11 records per population whose outcome cell is a total shared between two to four dams (each of the four Klamath dams carries '400 total between 4 dams'), keeping every other value to_num reads. Length falls to rho = 0.105, raw p = 0.037, Holm p = 0.074, n = 393. Its 95% CI [0.007, 0.204] still excludes zero, but claim 2's 'survives Holm-Bonferroni correction' does not hold under this handling: 11 records, 2.7% of the length population and four distinct project totals, are enough to take it over 0.05. Height stays null (rho = 0.026, p = 0.60)."
    },
    {
      "call_id": "shared-totals-zero-miles-off-reference",
      "tool": "primary_tests",
      "args": {
        "include_zero_miles": true,
        "coercion": "no_shared_totals"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "height_vs_us": {
            "test": "Spearman rank correlation",
            "predictor": "Dam_Height_ft",
            "outcome": "Miles_Restored_US",
            "rho": 0.03172541014332166,
            "p_raw": 0.519752528644156,
            "p_holm": 0.519752528644156,
            "ci95": [
              -0.06716001461088054,
              0.12892992610106344
            ],
            "n": 414
          },
          "length_vs_us": {
            "test": "Spearman rank correlation",
            "predictor": "Dam_Length_ft",
            "outcome": "Miles_Restored_US",
            "rho": 0.08899800254253637,
            "p_raw": 0.0765278958475327,
            "p_holm": 0.1530557916950654,
            "ci95": [
              -0.009299696673369287,
              0.1856403795457669
            ],
            "n": 397
          },
          "multiple_comparison": "Holm-Bonferroni across the two primary tests"
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "The two targeted corrections together -- shared totals withheld, zero-mile removals counted: length rho = 0.089, raw p = 0.077, Holm p = 0.153, n = 397, and the 95% CI [-0.009, 0.186] spans zero. The association is still positive but is not distinguishable from zero at the 5% level even before correction. Height stays null (rho = 0.032, p = 0.52)."
    },
    {
      "call_id": "ci99-off-reference",
      "tool": "primary_tests",
      "args": {
        "ci_level": 0.99
      },
      "select": "/length_vs_us/ci99",
      "expect": {
        "kind": "literal",
        "value": [
          0.00350476794759058,
          0.2547970280370891
        ]
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "At 99% the length interval is [0.0035, 0.255]: it still excludes zero, by 0.0035, consistent with a raw p of 0.0092 sitting just under 0.01."
    },
    {
      "call_id": "monte-carlo-off-reference",
      "tool": "primary_tests",
      "args": {
        "seed": 1,
        "n_boot": 20000
      },
      "select": "/length_vs_us/ci95",
      "expect": {
        "kind": "literal",
        "value": [
          0.03168558920792304,
          0.22529684607852354
        ]
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "A different seed and twice the resamples move the length CI from the published [0.0324, 0.2256] to [0.0317, 0.2253]. Monte Carlo error is under 0.001 at both bounds, so 'the CI excludes zero' is not a seed artefact."
    },
    {
      "call_id": "rho-published",
      "tool": "rank_variance_explained",
      "args": {},
      "select": "/rho",
      "expect": {
        "kind": "published_pointer",
        "result_path": "/primary/length_vs_us/rho"
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true
    },
    {
      "call_id": "claim5-rho-squared",
      "tool": "rank_variance_explained",
      "args": {},
      "select": "/rho_squared",
      "expect": {
        "kind": "literal",
        "value": 0.0167
      },
      "rel_tolerance": 0.0,
      "abs_tolerance": 5e-05,
      "enabled": true,
      "notes": "Claim 5's reported rho-squared of 0.0167, checked at its reported precision (half the last decimal place). It is a claim value, not a results.json value, so it is a literal; it is at the published defaults and is not an off-reference call."
    },
    {
      "call_id": "claim5-interval-published-settings",
      "tool": "rank_variance_explained",
      "args": {},
      "select": "/rho_squared_ci95",
      "expect": {
        "kind": "literal",
        "value": [
          0.001047014571264144,
          0.050903544260790086
        ]
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "Claim 5 reads the point estimate ('under 2% of rank variance'). The 95% bootstrap interval admits rho-squared from 0.001 to 0.051, so length could account for up to about 5% of rank variance. Still small -- the claim's reading survives -- but its honest bound is the interval's, not the point's."
    },
    {
      "call_id": "height-off-reference",
      "tool": "rank_variance_explained",
      "args": {
        "predictor": "Dam_Height_ft"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "predictor": "Dam_Height_ft",
          "n": 420,
          "rho": 0.046629112344296085,
          "ci95": [
            -0.05224294234171982,
            0.14548577856920683
          ],
          "rho_squared": 0.0021742741180169854,
          "rho_squared_ci95": [
            0.0,
            0.02116611176588828
          ]
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "Claim 5's subject is dam size, height and length. Height's rho-squared is 0.002; its CI spans zero, so the interval runs from 0 to 0.021."
    },
    {
      "call_id": "rank-variance-both-data-handling-off-reference",
      "tool": "rank_variance_explained",
      "args": {
        "include_zero_miles": true,
        "coercion": "bare_only"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "predictor": "Dam_Length_ft",
          "n": 382,
          "rho": 0.0986097139432256,
          "ci95": [
            1.1209541698832036e-05,
            0.19742234553373778
          ],
          "rho_squared": 0.00972387568396478,
          "rho_squared_ci95": [
            1.2565382509785423e-10,
            0.03897558251604255
          ]
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "With zero-mile removals counted and bare_only coercion, length's rho-squared falls to 0.0097, with an interval reaching 0.039. Every alternative handling makes claim 5's 'minor determinant' reading stronger, not weaker."
    },
    {
      "call_id": "r1-partial-published",
      "tool": "partial_association",
      "args": {},
      "expect": {
        "kind": "published_pointer",
        "result_path": "/robustness/partial_spearman"
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "recipe_id": "american-rivers-dam-removal-partial-spearman",
      "enabled": true
    },
    {
      "call_id": "partial-both-data-handling-off-reference",
      "tool": "partial_association",
      "args": {
        "include_zero_miles": true,
        "coercion": "bare_only"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "n": 380,
          "height_vs_us_given_length": {
            "partial_rho": -0.02859901851171766,
            "p": 0.5783649906013797
          },
          "length_vs_us_given_height": {
            "partial_rho": 0.09425112058345075,
            "p": 0.06645723450642839
          },
          "note": "rank-residualization (rank all; OLS-residualize x,y on control; Pearson)"
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "Claim 3's length-given-height association (published p = 0.015) goes to p = 0.044 with zero-mile removals counted, 0.036 with shared totals withheld, 0.022 with bare numbers only (the next three calls), 0.095 with zero-mile removals counted and shared totals withheld, and 0.066 here, with zero-mile removals counted and bare numbers only. The height partial stays null throughout (p >= 0.387). So the half of claim 3 that says height adds nothing once length is controlled is robust; the half that says length holds survives each single alternative at p < 0.05 and neither combination."
    },
    {
      "call_id": "partial-zero-miles-off-reference",
      "tool": "partial_association",
      "args": {
        "include_zero_miles": true
      },
      "expect": {
        "kind": "literal",
        "value": {
          "n": 404,
          "height_vs_us_given_length": {
            "partial_rho": -0.013588760837401744,
            "p": 0.7853936646752621
          },
          "length_vs_us_given_height": {
            "partial_rho": 0.10021180567765599,
            "p": 0.04410818635688451
          },
          "note": "rank-residualization (rank all; OLS-residualize x,y on control; Pearson)"
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true
    },
    {
      "call_id": "partial-shared-totals-off-reference",
      "tool": "partial_association",
      "args": {
        "coercion": "no_shared_totals"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "n": 389,
          "height_vs_us_given_length": {
            "partial_rho": -0.03697233877115312,
            "p": 0.4671582976782145
          },
          "length_vs_us_given_height": {
            "partial_rho": 0.10619055057366325,
            "p": 0.03629671319899331
          },
          "note": "rank-residualization (rank all; OLS-residualize x,y on control; Pearson)"
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "Shared project totals withheld: length given height p = 0.036 (partial rho = 0.106, n = 389); height given length stays null (p = 0.47)."
    },
    {
      "call_id": "partial-bare-only-off-reference",
      "tool": "partial_association",
      "args": {
        "coercion": "bare_only"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "n": 376,
          "height_vs_us_given_length": {
            "partial_rho": -0.04473525367111056,
            "p": 0.3870422797312438
          },
          "length_vs_us_given_height": {
            "partial_rho": 0.11786332590230292,
            "p": 0.022264919582666562
          },
          "note": "rank-residualization (rank all; OLS-residualize x,y on control; Pearson)"
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true
    },
    {
      "call_id": "partial-shared-totals-zero-miles-off-reference",
      "tool": "partial_association",
      "args": {
        "include_zero_miles": true,
        "coercion": "no_shared_totals"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "n": 393,
          "height_vs_us_given_length": {
            "partial_rho": -0.021852843907031655,
            "p": 0.6658232549817958
          },
          "length_vs_us_given_height": {
            "partial_rho": 0.0842756097125514,
            "p": 0.09524701267585428
          },
          "note": "rank-residualization (rank all; OLS-residualize x,y on control; Pearson)"
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "Shared totals withheld and zero-mile removals counted: length given height p = 0.095 (partial rho = 0.084, n = 393); height given length stays null (p = 0.67)."
    },
    {
      "call_id": "r2-winsorized-published",
      "tool": "winsorized_tests",
      "args": {},
      "expect": {
        "kind": "published_pointer",
        "result_path": "/robustness/winsorized_us"
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "recipe_id": "american-rivers-dam-removal-winsorized",
      "enabled": true
    },
    {
      "call_id": "winsor-p75-off-reference",
      "tool": "winsorized_tests",
      "args": {
        "winsor_quantile": 0.75
      },
      "expect": {
        "kind": "literal",
        "value": {
          "winsor_cap_miles": 18.450000000000003,
          "height_vs_us": {
            "rho": 0.04180597236537626,
            "p": 0.39277941573945035,
            "n": 420
          },
          "length_vs_us": {
            "rho": 0.11111476094930925,
            "p": 0.02552396119257339,
            "n": 404
          }
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "Spearman is invariant to monotone transforms, so a cap moves rho only through the ties it creates; the published 99th-percentile cap (614.45 miles) touches only the 5 largest of 478 outcome values, 3 of them inside each analysis population, and could barely have moved either coefficient. Capping the whole upper quartile (18.45 miles) is the version of R2 that could have failed. Length moves from rho = 0.129 to 0.111 and holds at raw p = 0.026, so claim 4's reading -- the result is not driven by extreme values -- stands on the harsher test too. Corrected across the pair as the primary family is, that p would be 0.051 (holm-p75-winsorized-off-reference)."
    },
    {
      "call_id": "winsor-p95-off-reference",
      "tool": "winsorized_tests",
      "args": {
        "winsor_quantile": 0.95
      },
      "expect": {
        "kind": "literal",
        "value": {
          "winsor_cap_miles": 175.0,
          "height_vs_us": {
            "rho": 0.04665298708650824,
            "p": 0.3401992264548424,
            "n": 420
          },
          "length_vs_us": {
            "rho": 0.1290536760069413,
            "p": 0.00941029985963157,
            "n": 404
          }
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "The conventional 5% winsorization caps at 175 miles and is indistinguishable from the published 1% cap: length rho = 0.129, p = 0.0094."
    },
    {
      "call_id": "winsor-both-data-handling-off-reference",
      "tool": "winsorized_tests",
      "args": {
        "include_zero_miles": true,
        "coercion": "bare_only"
      },
      "expect": {
        "kind": "literal",
        "value": {
          "winsor_cap_miles": 618.0499999999995,
          "height_vs_us": {
            "rho": 0.029255541883683105,
            "p": 0.5591274701047753,
            "n": 401
          },
          "length_vs_us": {
            "rho": 0.09860435023354004,
            "p": 0.054156414745496796,
            "n": 382
          }
        }
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "R2 with zero-mile removals counted and bare_only coercion: the cap is recomputed over the bare-number outcome, and length sits at p = 0.054 as in the unwinsorized test -- the winsorization never was what the length result depended on."
    },
    {
      "call_id": "holm-published",
      "tool": "holm",
      "args": {},
      "select": "/1",
      "expect": {
        "kind": "published_pointer",
        "result_path": "/primary/length_vs_us/p_holm"
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true
    },
    {
      "call_id": "holm-extended-family-off-reference",
      "tool": "holm",
      "args": {
        "p_values": [
          0.34044667406584744,
          0.009245880588870069,
          0.5781992116082467,
          0.015268165187900319
        ]
      },
      "expect": {
        "kind": "literal",
        "value": [
          0.6808933481316949,
          0.036983522355480275,
          0.6808933481316949,
          0.045804495563700955
        ]
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "The published family is the two primaries. Counting the two partial correlations in it as well (published p-values, in order: height, length, height|length, length|height), length's primary corrects to 0.037 and its partial to 0.046. Both stay under 0.05, the partial by 0.004."
    },
    {
      "call_id": "holm-p75-winsorized-off-reference",
      "tool": "holm",
      "args": {
        "p_values": [
          0.39277941573945035,
          0.02552396119257339
        ]
      },
      "expect": {
        "kind": "literal",
        "value": [
          0.39277941573945035,
          0.05104792238514678
        ]
      },
      "rel_tolerance": 1e-06,
      "abs_tolerance": 1e-09,
      "enabled": true,
      "notes": "The two upper-quartile-winsorized p-values (winsor-p75-off-reference: height, length) corrected as the primary pair is. Length's becomes 0.051, just over 0.05. Claim 4 reports raw p-values only, so this does not contradict it; it says a Holm-corrected length result would not survive the harshest cap."
    }
  ]
}