{
  "as_of_utc": "2026-10-08T12:45:47.706162+00:00",
  "models": {
    "baseline": "gpt-6-sol",
    "subject": "gpt-6.1-sol"
  },
  "complete_primary_pairs": 200,
  "attempted_primary_tasks": 208,
  "full_frozen_primary_tasks": 228,
  "retained_attempts": 536,
  "pending_calls": 64,
  "pending_primary_tasks": 20,
  "pending_repeat_calls": 24,
  "complete_pair_outcomes": {
    "baseline1_subject1": 136,
    "baseline0_subject1": 15,
    "baseline0_subject0": 42,
    "baseline1_subject0": 7
  },
  "quality": {
    "known_pairs": 200,
    "complete_pair_estimate": 0.04246411483253589,
    "full_planned_unknown_bounds": [
      -0.07017543859649122,
      0.15350877192982457
    ],
    "strata": {
      "application_libraries_services": {
        "planned_pairs": 75,
        "known_pairs": 60,
        "complete_pair_estimate": 0.1,
        "full_planned_unknown_bounds": [
          -0.08,
          0.28
        ],
        "weight": 0.32894736842105265
      },
      "developer_tooling": {
        "planned_pairs": 81,
        "known_pairs": 74,
        "complete_pair_estimate": 0,
        "full_planned_unknown_bounds": [
          -0.07407407407407407,
          0.08641975308641975
        ],
        "weight": 0.35526315789473684
      },
      "scientific_computing": {
        "planned_pairs": 72,
        "known_pairs": 66,
        "complete_pair_estimate": 0.030303030303030304,
        "full_planned_unknown_bounds": [
          -0.05555555555555555,
          0.09722222222222222
        ],
        "weight": 0.3157894736842105
      }
    },
    "interpretation": "Descriptive candidate-minus-baseline difference. Bounds vary unknown outcomes over 0/1; they are not confidence intervals. Complete pairs may be selected by outcome-dependent missingness.",
    "cluster_uncertainty": {
      "estimate": 0.04246411483253589,
      "percentile_interval_95": [
        -0.0009976784152482055,
        0.08957760303901226
      ],
      "strata": {
        "developer_tooling": {
          "pairs": 74,
          "repositories": 62
        },
        "scientific_computing": {
          "pairs": 66,
          "repositories": 47
        },
        "application_libraries_services": {
          "pairs": 60,
          "repositories": 43
        }
      },
      "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval."
    }
  },
  "delivery": {
    "known_pairs": 206,
    "complete_pair_estimate": 0.056380272165094586,
    "full_planned_unknown_bounds": [
      -0.04824561403508772,
      0.14473684210526316
    ],
    "strata": {
      "application_libraries_services": {
        "planned_pairs": 75,
        "known_pairs": 63,
        "complete_pair_estimate": 0.14285714285714285,
        "full_planned_unknown_bounds": [
          -0.04,
          0.28
        ],
        "weight": 0.32894736842105265
      },
      "developer_tooling": {
        "planned_pairs": 81,
        "known_pairs": 76,
        "complete_pair_estimate": 0.013157894736842105,
        "full_planned_unknown_bounds": [
          -0.04938271604938271,
          0.07407407407407407
        ],
        "weight": 0.35526315789473684
      },
      "scientific_computing": {
        "planned_pairs": 72,
        "known_pairs": 67,
        "complete_pair_estimate": 0.014925373134328358,
        "full_planned_unknown_bounds": [
          -0.05555555555555555,
          0.08333333333333333
        ],
        "weight": 0.3157894736842105
      }
    },
    "interpretation": "Descriptive candidate-minus-baseline difference. Bounds vary unknown outcomes over 0/1; they are not confidence intervals. Complete pairs may be selected by outcome-dependent missingness.",
    "cluster_uncertainty": {
      "estimate": 0.056380272165094586,
      "percentile_interval_95": [
        0.007082921679988119,
        0.1062818417433145
      ],
      "strata": {
        "developer_tooling": {
          "pairs": 76,
          "repositories": 64
        },
        "scientific_computing": {
          "pairs": 67,
          "repositories": 48
        },
        "application_libraries_services": {
          "pairs": 63,
          "repositories": 43
        }
      },
      "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval."
    }
  },
  "both_success_conditional_metrics": {
    "elapsed_seconds": {
      "percentile_interval_95": [
        0.6745782372493282,
        0.7984356727382157
      ],
      "strata": {
        "developer_tooling": {
          "pairs": 49,
          "repositories": 42
        },
        "scientific_computing": {
          "pairs": 44,
          "repositories": 36
        },
        "application_libraries_services": {
          "pairs": 43,
          "repositories": 33
        }
      },
      "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval.",
      "geometric_mean_ratio": 0.7343046964258195,
      "conditional_on_both_success_positive_metrics": true
    },
    "active_work_seconds": {
      "percentile_interval_95": [
        0.5695056087209027,
        0.6758133666962739
      ],
      "strata": {
        "developer_tooling": {
          "pairs": 49,
          "repositories": 42
        },
        "scientific_computing": {
          "pairs": 44,
          "repositories": 36
        },
        "application_libraries_services": {
          "pairs": 43,
          "repositories": 33
        }
      },
      "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval.",
      "geometric_mean_ratio": 0.6203839458521974,
      "conditional_on_both_success_positive_metrics": true
    },
    "test_wait_seconds": {
      "percentile_interval_95": [
        0.9615028317974281,
        1.2294203638178334
      ],
      "strata": {
        "developer_tooling": {
          "pairs": 49,
          "repositories": 42
        },
        "scientific_computing": {
          "pairs": 44,
          "repositories": 36
        },
        "application_libraries_services": {
          "pairs": 43,
          "repositories": 33
        }
      },
      "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval.",
      "geometric_mean_ratio": 1.0876236811680136,
      "conditional_on_both_success_positive_metrics": true
    },
    "observed_total_tokens": {
      "percentile_interval_95": [
        0.3429057639967635,
        0.41764903577177814
      ],
      "strata": {
        "developer_tooling": {
          "pairs": 49,
          "repositories": 42
        },
        "scientific_computing": {
          "pairs": 44,
          "repositories": 36
        },
        "application_libraries_services": {
          "pairs": 43,
          "repositories": 33
        }
      },
      "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval.",
      "geometric_mean_ratio": 0.3785827155156641,
      "conditional_on_both_success_positive_metrics": true
    },
    "repository_tool_calls": {
      "percentile_interval_95": [
        0.6145802228175996,
        0.6915649942757055
      ],
      "strata": {
        "developer_tooling": {
          "pairs": 49,
          "repositories": 42
        },
        "scientific_computing": {
          "pairs": 44,
          "repositories": 36
        },
        "application_libraries_services": {
          "pairs": 43,
          "repositories": 33
        }
      },
      "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval.",
      "geometric_mean_ratio": 0.6517911711433334,
      "conditional_on_both_success_positive_metrics": true
    },
    "repository_test_calls": {
      "percentile_interval_95": [
        1.1299420379215215,
        1.329840548107374
      ],
      "strata": {
        "developer_tooling": {
          "pairs": 49,
          "repositories": 42
        },
        "scientific_computing": {
          "pairs": 44,
          "repositories": 36
        },
        "application_libraries_services": {
          "pairs": 43,
          "repositories": 33
        }
      },
      "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval.",
      "geometric_mean_ratio": 1.2238370099099352,
      "conditional_on_both_success_positive_metrics": true
    }
  },
  "attempted_primary_metric_summaries": {
    "elapsed_seconds": {
      "baseline": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 47917.33472250588,
        "mean_observed": 230.37180155050905,
        "median_observed": 193.0523305205861,
        "sum_incomplete": false
      },
      "subject": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 35507.12245861953,
        "mean_observed": 170.7073195125939,
        "median_observed": 142.8602343959501,
        "sum_incomplete": false
      },
      "paired": {
        "planned_subset_pairs": 208,
        "observed_pairs": 208,
        "missing_metric_task_ids": [],
        "mean_subject_minus_baseline": -59.66448203791515,
        "median_subject_minus_baseline": -30.827048208680935,
        "positive_ratio_pairs": 208,
        "geometric_mean_subject_over_baseline": 0.7478746840790041
      }
    },
    "active_work_seconds": {
      "baseline": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 36809.04316582391,
        "mean_observed": 176.9665536818457,
        "median_observed": 140.19386220851447,
        "sum_incomplete": false
      },
      "subject": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 23816.733225028962,
        "mean_observed": 114.50352512033155,
        "median_observed": 78.720375269535,
        "sum_incomplete": false
      },
      "paired": {
        "planned_subset_pairs": 208,
        "observed_pairs": 208,
        "missing_metric_task_ids": [],
        "mean_subject_minus_baseline": -62.46302856151418,
        "median_subject_minus_baseline": -43.74870629212819,
        "positive_ratio_pairs": 208,
        "geometric_mean_subject_over_baseline": 0.6302032836105758
      }
    },
    "test_wait_seconds": {
      "baseline": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 11108.291556681972,
        "mean_observed": 53.405247868663324,
        "median_observed": 37.85946054197848,
        "sum_incomplete": false
      },
      "subject": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 11690.38923359057,
        "mean_observed": 56.20379439226235,
        "median_observed": 44.929926082957536,
        "sum_incomplete": false
      },
      "paired": {
        "planned_subset_pairs": 208,
        "observed_pairs": 208,
        "missing_metric_task_ids": [],
        "mean_subject_minus_baseline": 2.798546523599026,
        "median_subject_minus_baseline": 1.5205699579091743,
        "positive_ratio_pairs": 203,
        "geometric_mean_subject_over_baseline": 1.0843204289259822
      }
    },
    "observed_total_tokens": {
      "baseline": {
        "attempts": 208,
        "observed": 207,
        "missing_ids": [
          "uniform001--r0--baseline--jupyter__qtconsole-611"
        ],
        "sum_observed": 139579976,
        "mean_observed": 674299.4009661836,
        "median_observed": 425863,
        "sum_incomplete": true
      },
      "subject": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 46047725,
        "mean_observed": 221383.29326923078,
        "median_observed": 144098.0,
        "sum_incomplete": false
      },
      "paired": {
        "planned_subset_pairs": 208,
        "observed_pairs": 207,
        "missing_metric_task_ids": [
          "jupyter__qtconsole-611"
        ],
        "mean_subject_minus_baseline": -452082.1256038647,
        "median_subject_minus_baseline": -245255,
        "positive_ratio_pairs": 207,
        "geometric_mean_subject_over_baseline": 0.3746107196998558
      }
    },
    "repository_tool_calls": {
      "baseline": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 6205,
        "mean_observed": 29.83173076923077,
        "median_observed": 25.5,
        "sum_incomplete": false
      },
      "subject": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 3997,
        "mean_observed": 19.216346153846153,
        "median_observed": 16.0,
        "sum_incomplete": false
      },
      "paired": {
        "planned_subset_pairs": 208,
        "observed_pairs": 208,
        "missing_metric_task_ids": [],
        "mean_subject_minus_baseline": -10.615384615384615,
        "median_subject_minus_baseline": -7.5,
        "positive_ratio_pairs": 207,
        "geometric_mean_subject_over_baseline": 0.6506425096995748
      }
    },
    "repository_test_calls": {
      "baseline": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 324,
        "mean_observed": 1.5576923076923077,
        "median_observed": 1.0,
        "sum_incomplete": false
      },
      "subject": {
        "attempts": 208,
        "observed": 208,
        "missing_ids": [],
        "sum_observed": 387,
        "mean_observed": 1.8605769230769231,
        "median_observed": 2.0,
        "sum_incomplete": false
      },
      "paired": {
        "planned_subset_pairs": 208,
        "observed_pairs": 208,
        "missing_metric_task_ids": [],
        "mean_subject_minus_baseline": 0.30288461538461536,
        "median_subject_minus_baseline": 0.0,
        "positive_ratio_pairs": 203,
        "geometric_mean_subject_over_baseline": 1.225553697829186
      }
    }
  },
  "attempted_quality_bounds_by_domain": {
    "developer_tooling": {
      "attempted_pairs": 78,
      "unknown_quality_difference_bounds": [
        -0.038461538461538464,
        0.05128205128205128
      ]
    },
    "scientific_computing": {
      "attempted_pairs": 67,
      "unknown_quality_difference_bounds": [
        0.014925373134328358,
        0.029850746268656716
      ]
    },
    "application_libraries_services": {
      "attempted_pairs": 63,
      "unknown_quality_difference_bounds": [
        0.09523809523809523,
        0.14285714285714285
      ]
    }
  },
  "attempted_weighted_quality_bounds": [
    0.022377628917298983,
    0.0746376561380489
  ],
  "known_individual_grades": 525,
  "unknown_individual_grades": 11,
  "observed_usage_records": 535,
  "primary_arm_counts": {
    "baseline": {
      "invoked": 208,
      "known_grades": 201,
      "unknown_grades": 7,
      "passed_known": 144,
      "complete_pair_passes": 143,
      "last_observed_usage_records": 207
    },
    "subject": {
      "invoked": 208,
      "known_grades": 204,
      "unknown_grades": 4,
      "passed_known": 155,
      "complete_pair_passes": 151,
      "last_observed_usage_records": 208
    }
  },
  "analysis": {
    "seed": "967a16d298f78dc9a2f6580a8989ce4ff04d297c876985cba040b54f62714df8-approved200-twoissue-uniform001",
    "resamples": 10000,
    "method": "Repository-within-domain clustered bootstrap with unchanged domain weights; pointwise exploratory intervals"
  },
  "unexecuted_partition": {
    "primary": 36,
    "startup_held": 4,
    "repeat": 24
  },
  "coverage": {
    "raw_trace_attempts": 536,
    "recorded_tools": 13083,
    "raw_requested_tools": 13085,
    "extra_request_execution_unknown": 2,
    "recorded_test_requests_joined": 919,
    "internal_complete_grades": 916,
    "internal_incomplete_grades": 3,
    "retained_lifecycle_partitions": 36,
    "lifecycle_attempts": 536
  },
  "repeat_supplement": {
    "scope": "Conditional30-task subset, three attempts/model, not original36panel completion",
    "arms": {
      "baseline": {
        "tasks": 30,
        "attempts": 90,
        "known_grades": 90,
        "passes": 60,
        "quality_unknowns": 0,
        "stable_tasks": 26,
        "all_three_pass": 18,
        "all_three_fail": 8,
        "mixed_tasks": 4,
        "round_passes": {
          "0": 21,
          "1": 18,
          "2": 21
        },
        "delivery_success": 60,
        "metric_variation": {
          "elapsed_seconds": {
            "complete_positive_series": 30,
            "median_coefficient_of_variation": 0.3790805125992988,
            "median_max_over_min": 2.1005510379889385
          },
          "active_work_seconds": {
            "complete_positive_series": 30,
            "median_coefficient_of_variation": 0.34431498050553977,
            "median_max_over_min": 1.8936305740426644
          },
          "test_wait_seconds": {
            "complete_positive_series": 29,
            "median_coefficient_of_variation": 0.5345296947213513,
            "median_max_over_min": 3.1568131125718644
          },
          "observed_total_tokens": {
            "complete_positive_series": 30,
            "median_coefficient_of_variation": 0.2995276098147054,
            "median_max_over_min": 1.7392695610752584
          },
          "repository_tool_calls": {
            "complete_positive_series": 30,
            "median_coefficient_of_variation": 0.2051336464415457,
            "median_max_over_min": 1.4673913043478262
          },
          "repository_test_calls": {
            "complete_positive_series": 29,
            "median_coefficient_of_variation": 0.3333333333333333,
            "median_max_over_min": 2.0
          }
        }
      },
      "subject": {
        "tasks": 30,
        "attempts": 90,
        "known_grades": 90,
        "passes": 68,
        "quality_unknowns": 0,
        "stable_tasks": 27,
        "all_three_pass": 21,
        "all_three_fail": 6,
        "mixed_tasks": 3,
        "round_passes": {
          "0": 22,
          "1": 23,
          "2": 23
        },
        "delivery_success": 68,
        "metric_variation": {
          "elapsed_seconds": {
            "complete_positive_series": 30,
            "median_coefficient_of_variation": 0.29421080324012044,
            "median_max_over_min": 1.755437296212611
          },
          "active_work_seconds": {
            "complete_positive_series": 30,
            "median_coefficient_of_variation": 0.2926355397620106,
            "median_max_over_min": 1.8312819034475485
          },
          "test_wait_seconds": {
            "complete_positive_series": 30,
            "median_coefficient_of_variation": 0.4139239478524227,
            "median_max_over_min": 2.048372545623665
          },
          "observed_total_tokens": {
            "complete_positive_series": 30,
            "median_coefficient_of_variation": 0.2741394115842789,
            "median_max_over_min": 1.734927470924348
          },
          "repository_tool_calls": {
            "complete_positive_series": 30,
            "median_coefficient_of_variation": 0.1257700495776139,
            "median_max_over_min": 1.287593984962406
          },
          "repository_test_calls": {
            "complete_positive_series": 30,
            "median_coefficient_of_variation": 0.24743582965269673,
            "median_max_over_min": 1.5
          }
        }
      }
    },
    "independent_replay_targets": {
      "quality_mean": {
        "estimate": 0.0912280701754386,
        "percentile_interval_95": [
          0.0,
          0.19473684210526318
        ],
        "strata": {
          "developer_tooling": {
            "pairs": 10,
            "repositories": 10
          },
          "scientific_computing": {
            "pairs": 10,
            "repositories": 10
          },
          "application_libraries_services": {
            "pairs": 10,
            "repositories": 10
          }
        },
        "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval."
      },
      "quality_stability": {
        "estimate": 0.031578947368421054,
        "percentile_interval_95": [
          -0.10657894736842105,
          0.1724013157894732
        ],
        "strata": {
          "developer_tooling": {
            "pairs": 10,
            "repositories": 10
          },
          "scientific_computing": {
            "pairs": 10,
            "repositories": 10
          },
          "application_libraries_services": {
            "pairs": 10,
            "repositories": 10
          }
        },
        "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval."
      },
      "delivery_mean": {
        "estimate": 0.0912280701754386,
        "percentile_interval_95": [
          1.7529837230923525e-18,
          0.19430921052631564
        ],
        "strata": {
          "developer_tooling": {
            "pairs": 10,
            "repositories": 10
          },
          "scientific_computing": {
            "pairs": 10,
            "repositories": 10
          },
          "application_libraries_services": {
            "pairs": 10,
            "repositories": 10
          }
        },
        "interpretation": "Domain-stratified repository bootstrap; all eligible task pairs in a drawn repository travel together. Exploratory pointwise uncertainty conditional on observed pairs; missing outcomes require the separate full-cohort bounds. Fewer than two observed repositories in any domain suppresses the interval."
      }
    },
    "domain_counts": {
      "developer_tooling": 10,
      "scientific_computing": 10,
      "application_libraries_services": 10
    },
    "all_180_panel_attempts_preserved": true,
    "repeats_add_primary_pairs": 0,
    "resamples": 10000,
    "seed": "967a16d298f78dc9a2f6580a8989ce4ff04d297c876985cba040b54f62714df8-approved200-twoissue-uniform001",
    "primary_checkpoint_unchanged": true
  },
  "expanded_metric_summaries": {
    "all_retained": {
      "baseline": {
        "attempts": 268,
        "known_grades": 261,
        "unknown_grades": 7,
        "passes": 183,
        "delivery_success": 183,
        "delivery_unknown": 2,
        "terminal_observed": 266,
        "cleanup_confirmed": 268,
        "metrics": {
          "elapsed_seconds": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 61524.745344043244,
            "mean_observed": 229.5699453135942,
            "median_observed": 186.57783937489148,
            "min_observed": 1.8388292910531163,
            "max_observed": 831.0675201250706
          },
          "active_work_seconds": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 46536.23313449137,
            "mean_observed": 173.64266094959467,
            "median_observed": 129.85476220806595,
            "min_observed": 1.8388292910531163,
            "max_observed": 622.2555198760238
          },
          "test_wait_seconds": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 14988.512209551875,
            "mean_observed": 55.92728436399953,
            "median_observed": 38.75918372999877,
            "min_observed": 0.0,
            "max_observed": 327.01584812486544
          },
          "observed_total_tokens": {
            "observed": 267,
            "unknown": 1,
            "sum_observed": 177362625,
            "mean_observed": 664279.4943820224,
            "median_observed": 428507,
            "min_observed": 18139,
            "max_observed": 8904847
          },
          "repository_tool_calls": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 7953,
            "mean_observed": 29.675373134328357,
            "median_observed": 25.0,
            "min_observed": 0,
            "max_observed": 100
          },
          "repository_test_calls": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 416,
            "mean_observed": 1.5522388059701493,
            "median_observed": 1.0,
            "min_observed": 0,
            "max_observed": 6
          }
        }
      },
      "subject": {
        "attempts": 268,
        "known_grades": 264,
        "unknown_grades": 4,
        "passes": 201,
        "delivery_success": 201,
        "delivery_unknown": 2,
        "terminal_observed": 266,
        "cleanup_confirmed": 267,
        "metrics": {
          "elapsed_seconds": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 46012.59096162231,
            "mean_observed": 171.68877224485936,
            "median_observed": 142.66973000008147,
            "min_observed": 24.043436917010695,
            "max_observed": 681.0334505001083
          },
          "active_work_seconds": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 30202.22287170356,
            "mean_observed": 112.69486146158046,
            "median_observed": 75.78570591635071,
            "min_observed": 12.060376292094588,
            "max_observed": 600.1159225828014
          },
          "test_wait_seconds": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 15810.368089918746,
            "mean_observed": 58.9939107832789,
            "median_observed": 46.38071381300688,
            "min_observed": 0.0,
            "max_observed": 268.8993578739464
          },
          "observed_total_tokens": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 59648094,
            "mean_observed": 222567.51492537314,
            "median_observed": 150567.0,
            "min_observed": 22549,
            "max_observed": 1559522
          },
          "repository_tool_calls": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 5130,
            "mean_observed": 19.14179104477612,
            "median_observed": 16.0,
            "min_observed": 4,
            "max_observed": 96
          },
          "repository_test_calls": {
            "observed": 268,
            "unknown": 0,
            "sum_observed": 503,
            "mean_observed": 1.876865671641791,
            "median_observed": 2.0,
            "min_observed": 0,
            "max_observed": 5
          }
        }
      }
    },
    "primary": {
      "baseline": {
        "attempts": 208,
        "known_grades": 201,
        "unknown_grades": 7,
        "passes": 144,
        "delivery_success": 144,
        "delivery_unknown": 2,
        "terminal_observed": 207,
        "cleanup_confirmed": 208,
        "metrics": {
          "elapsed_seconds": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 47917.33472250588,
            "mean_observed": 230.37180155050905,
            "median_observed": 193.0523305205861,
            "min_observed": 1.8388292910531163,
            "max_observed": 831.0675201250706
          },
          "active_work_seconds": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 36809.04316582391,
            "mean_observed": 176.9665536818457,
            "median_observed": 140.19386220851447,
            "min_observed": 1.8388292910531163,
            "max_observed": 600.0943160830066
          },
          "test_wait_seconds": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 11108.291556681972,
            "mean_observed": 53.405247868663324,
            "median_observed": 37.85946054197848,
            "min_observed": 0.0,
            "max_observed": 327.01584812486544
          },
          "observed_total_tokens": {
            "observed": 207,
            "unknown": 1,
            "sum_observed": 139579976,
            "mean_observed": 674299.4009661836,
            "median_observed": 425863,
            "min_observed": 18139,
            "max_observed": 8904847
          },
          "repository_tool_calls": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 6205,
            "mean_observed": 29.83173076923077,
            "median_observed": 25.5,
            "min_observed": 0,
            "max_observed": 100
          },
          "repository_test_calls": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 324,
            "mean_observed": 1.5576923076923077,
            "median_observed": 1.0,
            "min_observed": 0,
            "max_observed": 6
          }
        }
      },
      "subject": {
        "attempts": 208,
        "known_grades": 204,
        "unknown_grades": 4,
        "passes": 155,
        "delivery_success": 155,
        "delivery_unknown": 2,
        "terminal_observed": 206,
        "cleanup_confirmed": 207,
        "metrics": {
          "elapsed_seconds": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 35507.12245861953,
            "mean_observed": 170.7073195125939,
            "median_observed": 142.8602343959501,
            "min_observed": 24.043436917010695,
            "max_observed": 652.5338487080298
          },
          "active_work_seconds": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 23816.733225028962,
            "mean_observed": 114.50352512033155,
            "median_observed": 78.720375269535,
            "min_observed": 12.060376292094588,
            "max_observed": 600.1159225828014
          },
          "test_wait_seconds": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 11690.38923359057,
            "mean_observed": 56.20379439226235,
            "median_observed": 44.929926082957536,
            "min_observed": 0.0,
            "max_observed": 268.8993578739464
          },
          "observed_total_tokens": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 46047725,
            "mean_observed": 221383.29326923078,
            "median_observed": 144098.0,
            "min_observed": 22549,
            "max_observed": 1559522
          },
          "repository_tool_calls": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 3997,
            "mean_observed": 19.216346153846153,
            "median_observed": 16.0,
            "min_observed": 4,
            "max_observed": 96
          },
          "repository_test_calls": {
            "observed": 208,
            "unknown": 0,
            "sum_observed": 387,
            "mean_observed": 1.8605769230769231,
            "median_observed": 2.0,
            "min_observed": 0,
            "max_observed": 5
          }
        }
      }
    },
    "additional_repeats": {
      "baseline": {
        "attempts": 60,
        "known_grades": 60,
        "unknown_grades": 0,
        "passes": 39,
        "delivery_success": 39,
        "delivery_unknown": 0,
        "terminal_observed": 59,
        "cleanup_confirmed": 60,
        "metrics": {
          "elapsed_seconds": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 13607.410621537361,
            "mean_observed": 226.79017702562268,
            "median_observed": 166.70373052114155,
            "min_observed": 36.30796404113062,
            "max_observed": 742.9976760419086
          },
          "active_work_seconds": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 9727.189968667459,
            "mean_observed": 162.11983281112433,
            "median_observed": 115.49364606256131,
            "min_observed": 26.703777666203678,
            "max_observed": 622.2555198760238
          },
          "test_wait_seconds": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 3880.2206528699026,
            "mean_observed": 64.67034421449837,
            "median_observed": 38.99910675047431,
            "min_observed": 7.9358601660933346,
            "max_observed": 302.6027099159546
          },
          "observed_total_tokens": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 37782649,
            "mean_observed": 629710.8166666667,
            "median_observed": 428537.0,
            "min_observed": 67849,
            "max_observed": 2921700
          },
          "repository_tool_calls": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 1748,
            "mean_observed": 29.133333333333333,
            "median_observed": 23.5,
            "min_observed": 8,
            "max_observed": 97
          },
          "repository_test_calls": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 92,
            "mean_observed": 1.5333333333333334,
            "median_observed": 1.0,
            "min_observed": 1,
            "max_observed": 4
          }
        }
      },
      "subject": {
        "attempts": 60,
        "known_grades": 60,
        "unknown_grades": 0,
        "passes": 46,
        "delivery_success": 46,
        "delivery_unknown": 0,
        "terminal_observed": 60,
        "cleanup_confirmed": 60,
        "metrics": {
          "elapsed_seconds": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 10505.468503002776,
            "mean_observed": 175.09114171671294,
            "median_observed": 142.53209808364045,
            "min_observed": 36.04535583406687,
            "max_observed": 681.0334505001083
          },
          "active_work_seconds": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 6385.4896466745995,
            "mean_observed": 106.42482744457666,
            "median_observed": 68.9170082489727,
            "min_observed": 18.912620207993314,
            "max_observed": 456.18938966607675
          },
          "test_wait_seconds": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 4119.978856328176,
            "mean_observed": 68.66631427213628,
            "median_observed": 53.71185251988936,
            "min_observed": 7.913252500118688,
            "max_observed": 239.24911429197527
          },
          "observed_total_tokens": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 13600369,
            "mean_observed": 226672.81666666668,
            "median_observed": 157752.5,
            "min_observed": 31163,
            "max_observed": 1125043
          },
          "repository_tool_calls": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 1133,
            "mean_observed": 18.883333333333333,
            "median_observed": 14.0,
            "min_observed": 5,
            "max_observed": 68
          },
          "repository_test_calls": {
            "observed": 60,
            "unknown": 0,
            "sum_observed": 116,
            "mean_observed": 1.9333333333333333,
            "median_observed": 2.0,
            "min_observed": 1,
            "max_observed": 5
          }
        }
      }
    }
  },
  "publication_note": "Unsigned curated aggregate export. Source hashes and omitted fields are documented in provenance.json and the report. Null values remain unknown; repeats are dependent observations."
}