{
  "component_effects": {
    "effects": [
      {
        "factor": "T",
        "model": "composer-2.5",
        "off_arms": [
          "neutral_control",
          "delete_first_gate",
          "semantic_net_loc_budget",
          "delete_first_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "task_type_gate",
          "task_type_delete_first",
          "task_type_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 0,
        "raw_net_delta_sample_sd": 0.0,
        "raw_net_deltas": [
          0,
          0,
          0,
          0,
          0
        ],
        "reasoning_effort": "default",
        "task_id": "refactor-dead-compatibility-path"
      },
      {
        "factor": "D",
        "model": "composer-2.5",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "semantic_net_loc_budget",
          "task_type_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "delete_first_gate",
          "task_type_delete_first",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 0,
        "raw_net_delta_sample_sd": 0.0,
        "raw_net_deltas": [
          0,
          0,
          0,
          0,
          0
        ],
        "reasoning_effort": "default",
        "task_id": "refactor-dead-compatibility-path"
      },
      {
        "factor": "B",
        "model": "composer-2.5",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "delete_first_gate",
          "task_type_delete_first"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "semantic_net_loc_budget",
          "task_type_net_loc_budget",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 0,
        "raw_net_delta_sample_sd": 0.0,
        "raw_net_deltas": [
          0,
          0,
          0,
          0,
          0
        ],
        "reasoning_effort": "default",
        "task_id": "refactor-dead-compatibility-path"
      },
      {
        "factor": "T",
        "model": "gpt-5.6-luna",
        "off_arms": [
          "neutral_control",
          "delete_first_gate",
          "semantic_net_loc_budget",
          "delete_first_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "task_type_gate",
          "task_type_delete_first",
          "task_type_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 0.2,
        "raw_net_delta_sample_sd": 0.32596012026013244,
        "raw_net_deltas": [
          0,
          0.25,
          0.0,
          0.75,
          0
        ],
        "reasoning_effort": "maximum",
        "task_id": "refactor-dead-compatibility-path"
      },
      {
        "factor": "D",
        "model": "gpt-5.6-luna",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "semantic_net_loc_budget",
          "task_type_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "delete_first_gate",
          "task_type_delete_first",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 0.3,
        "raw_net_delta_sample_sd": 0.6471089552772392,
        "raw_net_deltas": [
          0,
          -0.25,
          1,
          -0.25,
          1.0
        ],
        "reasoning_effort": "maximum",
        "task_id": "refactor-dead-compatibility-path"
      },
      {
        "factor": "B",
        "model": "gpt-5.6-luna",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "delete_first_gate",
          "task_type_delete_first"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "semantic_net_loc_budget",
          "task_type_net_loc_budget",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 0.3,
        "raw_net_delta_sample_sd": 0.54198708471697,
        "raw_net_deltas": [
          0,
          1.25,
          0.0,
          0.25,
          0
        ],
        "reasoning_effort": "maximum",
        "task_id": "refactor-dead-compatibility-path"
      },
      {
        "factor": "T",
        "model": "grok-4.5",
        "off_arms": [
          "neutral_control",
          "delete_first_gate",
          "semantic_net_loc_budget",
          "delete_first_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "task_type_gate",
          "task_type_delete_first",
          "task_type_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 0.8,
        "raw_net_delta_sample_sd": 1.8148691412881537,
        "raw_net_deltas": [
          3.25,
          -0.5,
          2.25,
          -0.5,
          -0.5
        ],
        "reasoning_effort": "high",
        "task_id": "refactor-dead-compatibility-path"
      },
      {
        "factor": "D",
        "model": "grok-4.5",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "semantic_net_loc_budget",
          "task_type_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "delete_first_gate",
          "task_type_delete_first",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 2.0,
        "raw_net_delta_sample_sd": 1.8285923547909742,
        "raw_net_deltas": [
          4.25,
          -0.5,
          3.25,
          1.5,
          1.5
        ],
        "reasoning_effort": "high",
        "task_id": "refactor-dead-compatibility-path"
      },
      {
        "factor": "B",
        "model": "grok-4.5",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "delete_first_gate",
          "task_type_delete_first"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "semantic_net_loc_budget",
          "task_type_net_loc_budget",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 0.1,
        "raw_net_delta_sample_sd": 2.3358617253596154,
        "raw_net_deltas": [
          -3.25,
          -0.5,
          3.25,
          0.5,
          0.5
        ],
        "reasoning_effort": "high",
        "task_id": "refactor-dead-compatibility-path"
      },
      {
        "factor": "T",
        "model": "composer-2.5",
        "off_arms": [
          "neutral_control",
          "delete_first_gate",
          "semantic_net_loc_budget",
          "delete_first_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "task_type_gate",
          "task_type_delete_first",
          "task_type_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 0.65,
        "raw_net_delta_sample_sd": 1.4534441853748632,
        "raw_net_deltas": [
          0,
          3.25,
          0,
          0,
          0
        ],
        "reasoning_effort": "default",
        "task_id": "refactor-shared-normalizer"
      },
      {
        "factor": "D",
        "model": "composer-2.5",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "semantic_net_loc_budget",
          "task_type_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "delete_first_gate",
          "task_type_delete_first",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": -0.65,
        "raw_net_delta_sample_sd": 1.4534441853748632,
        "raw_net_deltas": [
          0,
          -3.25,
          0,
          0,
          0
        ],
        "reasoning_effort": "default",
        "task_id": "refactor-shared-normalizer"
      },
      {
        "factor": "B",
        "model": "composer-2.5",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "delete_first_gate",
          "task_type_delete_first"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "semantic_net_loc_budget",
          "task_type_net_loc_budget",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": 0.65,
        "raw_net_delta_sample_sd": 1.4534441853748632,
        "raw_net_deltas": [
          0,
          3.25,
          0,
          0,
          0
        ],
        "reasoning_effort": "default",
        "task_id": "refactor-shared-normalizer"
      },
      {
        "factor": "T",
        "model": "gpt-5.6-luna",
        "off_arms": [
          "neutral_control",
          "delete_first_gate",
          "semantic_net_loc_budget",
          "delete_first_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "task_type_gate",
          "task_type_delete_first",
          "task_type_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": -0.5,
        "raw_net_delta_sample_sd": 1.0752906583803283,
        "raw_net_deltas": [
          0.5,
          -0.5,
          -2.25,
          0.25,
          -0.5
        ],
        "reasoning_effort": "maximum",
        "task_id": "refactor-shared-normalizer"
      },
      {
        "factor": "D",
        "model": "gpt-5.6-luna",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "semantic_net_loc_budget",
          "task_type_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "delete_first_gate",
          "task_type_delete_first",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": -3.1,
        "raw_net_delta_sample_sd": 3.1354824190226296,
        "raw_net_deltas": [
          2.0,
          -4.5,
          -2.25,
          -5.75,
          -5.0
        ],
        "reasoning_effort": "maximum",
        "task_id": "refactor-shared-normalizer"
      },
      {
        "factor": "B",
        "model": "gpt-5.6-luna",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "delete_first_gate",
          "task_type_delete_first"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "semantic_net_loc_budget",
          "task_type_net_loc_budget",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": -2.0,
        "raw_net_delta_sample_sd": 3.5044614422190468,
        "raw_net_deltas": [
          0.5,
          -4.0,
          2.75,
          -5.75,
          -3.5
        ],
        "reasoning_effort": "maximum",
        "task_id": "refactor-shared-normalizer"
      },
      {
        "factor": "T",
        "model": "grok-4.5",
        "off_arms": [
          "neutral_control",
          "delete_first_gate",
          "semantic_net_loc_budget",
          "delete_first_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "task_type_gate",
          "task_type_delete_first",
          "task_type_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": -0.45,
        "raw_net_delta_sample_sd": 3.2519225083018197,
        "raw_net_deltas": [
          0.75,
          -5.75,
          -1.25,
          2.25,
          1.75
        ],
        "reasoning_effort": "high",
        "task_id": "refactor-shared-normalizer"
      },
      {
        "factor": "D",
        "model": "grok-4.5",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "semantic_net_loc_budget",
          "task_type_net_loc_budget"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "delete_first_gate",
          "task_type_delete_first",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": -3.25,
        "raw_net_delta_sample_sd": 2.345207879911715,
        "raw_net_deltas": [
          -6.75,
          -1.75,
          -2.75,
          -0.75,
          -4.25
        ],
        "reasoning_effort": "high",
        "task_id": "refactor-shared-normalizer"
      },
      {
        "factor": "B",
        "model": "grok-4.5",
        "off_arms": [
          "neutral_control",
          "task_type_gate",
          "delete_first_gate",
          "task_type_delete_first"
        ],
        "off_behavior_pass_count": 20,
        "on_arms": [
          "semantic_net_loc_budget",
          "task_type_net_loc_budget",
          "delete_first_net_loc_budget",
          "task_type_delete_first_net_loc_budget"
        ],
        "on_behavior_pass_count": 20,
        "raw_net_delta_mean": -3.85,
        "raw_net_delta_sample_sd": 3.4532593299664014,
        "raw_net_deltas": [
          -6.75,
          -2.25,
          -3.25,
          0.75,
          -7.75
        ],
        "reasoning_effort": "high",
        "task_id": "refactor-shared-normalizer"
      }
    ],
    "protocol": "phase-2-component-effects-v1"
  },
  "groups": [
    {
      "arm": "delete_first_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -8,
      "raw_net_mean": -8,
      "raw_net_min": -8,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -8,
        -8,
        -8,
        -8,
        -8
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 358842,
          "sample_sd": 55408.06344748028,
          "values": [
            268537,
            405711,
            345574,
            395703,
            378685
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2840.2,
          "sample_sd": 211.65703390154553,
          "values": [
            2789,
            3143,
            2750,
            2938,
            2581
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 361682.2,
          "sample_sd": 55491.75889445207,
          "values": [
            271326,
            408854,
            348324,
            398641,
            381266
          ]
        }
      }
    },
    {
      "arm": "delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -8,
      "raw_net_mean": -8,
      "raw_net_min": -8,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -8,
        -8,
        -8,
        -8,
        -8
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 463161.2,
          "sample_sd": 94599.86164207642,
          "values": [
            466228,
            602360,
            436335,
            472628,
            338255
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2874.8,
          "sample_sd": 212.79262205255142,
          "values": [
            3074,
            3005,
            2784,
            2965,
            2546
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 466036,
          "sample_sd": 94766.69161683338,
          "values": [
            469302,
            605365,
            439119,
            475593,
            340801
          ]
        }
      }
    },
    {
      "arm": "neutral_control",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -8,
      "raw_net_mean": -8,
      "raw_net_min": -8,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -8,
        -8,
        -8,
        -8,
        -8
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 419679.6,
          "sample_sd": 60117.61439295475,
          "values": [
            497718,
            453402,
            358188,
            427595,
            361495
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2845.2,
          "sample_sd": 263.46384951260393,
          "values": [
            2973,
            2924,
            3005,
            2947,
            2377
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 422524.8,
          "sample_sd": 60246.352750187296,
          "values": [
            500691,
            456326,
            361193,
            430542,
            363872
          ]
        }
      }
    },
    {
      "arm": "semantic_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -8,
      "raw_net_mean": -8,
      "raw_net_min": -8,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -8,
        -8,
        -8,
        -8,
        -8
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 330577.6,
          "sample_sd": 108300.12699577042,
          "values": [
            393880,
            300116,
            174791,
            320800,
            463301
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2637.4,
          "sample_sd": 302.13705499325965,
          "values": [
            3033,
            2818,
            2238,
            2528,
            2570
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 333215,
          "sample_sd": 108470.97292133044,
          "values": [
            396913,
            302934,
            177029,
            323328,
            465871
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -8,
      "raw_net_mean": -8,
      "raw_net_min": -8,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -8,
        -8,
        -8,
        -8,
        -8
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 286126.6,
          "sample_sd": 59216.792122674124,
          "values": [
            345638,
            210639,
            284635,
            246737,
            342984
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2412.2,
          "sample_sd": 452.32256189582233,
          "values": [
            2372,
            1771,
            2795,
            2234,
            2889
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 288538.8,
          "sample_sd": 59553.48130630148,
          "values": [
            348010,
            212410,
            287430,
            248971,
            345873
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -8,
      "raw_net_mean": -8,
      "raw_net_min": -8,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -8,
        -8,
        -8,
        -8,
        -8
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 278080,
          "sample_sd": 129151.56503697507,
          "values": [
            208956,
            286054,
            485799,
            141588,
            268003
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2673.6,
          "sample_sd": 240.16306960063613,
          "values": [
            2801,
            2693,
            2906,
            2694,
            2274
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 280753.6,
          "sample_sd": 129230.86586725325,
          "values": [
            211757,
            288747,
            488705,
            144282,
            270277
          ]
        }
      }
    },
    {
      "arm": "task_type_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -8,
      "raw_net_mean": -8,
      "raw_net_min": -8,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -8,
        -8,
        -8,
        -8,
        -8
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 350913.4,
          "sample_sd": 144789.1858834768,
          "values": [
            411609,
            542970,
            340138,
            146578,
            313272
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2711,
          "sample_sd": 479.8421615489827,
          "values": [
            3003,
            3176,
            2523,
            1967,
            2886
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 353624.4,
          "sample_sd": 145228.73458203787,
          "values": [
            414612,
            546146,
            342661,
            148545,
            316158
          ]
        }
      }
    },
    {
      "arm": "task_type_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -8,
      "raw_net_mean": -8,
      "raw_net_min": -8,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -8,
        -8,
        -8,
        -8,
        -8
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 299322.6,
          "sample_sd": 152940.4719484022,
          "values": [
            565781,
            261308,
            234486,
            259534,
            175504
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2297.8,
          "sample_sd": 506.96765182800374,
          "values": [
            3073,
            1936,
            2436,
            2272,
            1772
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 301620.4,
          "sample_sd": 153394.21784017805,
          "values": [
            568854,
            263244,
            236922,
            261806,
            177276
          ]
        }
      }
    },
    {
      "arm": "delete_first_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -8,
      "raw_net_mean": -9.2,
      "raw_net_min": -10,
      "raw_net_sample_sd": 1.0954451150103321,
      "raw_net_values": [
        -10,
        -10,
        -8,
        -10,
        -8
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 199162.2,
          "sample_sd": 29805.462615097924,
          "values": [
            172135,
            214095,
            174851,
            191420,
            243310
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2016.8,
          "sample_sd": 239.54373295914047,
          "values": [
            1836,
            2336,
            2062,
            1730,
            2120
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 201179,
          "sample_sd": 29934.734699008106,
          "values": [
            173971,
            216431,
            176913,
            193150,
            245430
          ]
        }
      }
    },
    {
      "arm": "delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -8,
      "raw_net_mean": -9.2,
      "raw_net_min": -10,
      "raw_net_sample_sd": 1.0954451150103321,
      "raw_net_values": [
        -10,
        -8,
        -10,
        -10,
        -8
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 247379,
          "sample_sd": 68798.05829382106,
          "values": [
            247310,
            228368,
            173676,
            360234,
            227307
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1950.4,
          "sample_sd": 284.30054519821095,
          "values": [
            2137,
            1853,
            1590,
            2323,
            1849
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 249329.4,
          "sample_sd": 69061.54405528448,
          "values": [
            249447,
            230221,
            175266,
            362557,
            229156
          ]
        }
      }
    },
    {
      "arm": "neutral_control",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -10,
      "raw_net_mean": -10,
      "raw_net_min": -10,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -10,
        -10,
        -10,
        -10,
        -10
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 171812.8,
          "sample_sd": 42378.76915626503,
          "values": [
            185924,
            98852,
            206312,
            174640,
            193336
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1583.8,
          "sample_sd": 345.4022292921689,
          "values": [
            1256,
            1163,
            1820,
            1893,
            1787
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 173396.6,
          "sample_sd": 42612.38907055083,
          "values": [
            187180,
            100015,
            208132,
            176533,
            195123
          ]
        }
      }
    },
    {
      "arm": "semantic_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -8,
      "raw_net_mean": -9.4,
      "raw_net_min": -10,
      "raw_net_sample_sd": 0.8944271909999159,
      "raw_net_values": [
        -10,
        -8,
        -10,
        -9,
        -10
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 206934.6,
          "sample_sd": 92334.1248742847,
          "values": [
            136002,
            364089,
            212157,
            152126,
            170299
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1933.4,
          "sample_sd": 562.5818162720868,
          "values": [
            1665,
            2720,
            2329,
            1501,
            1452
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 208868,
          "sample_sd": 92838.22431520328,
          "values": [
            137667,
            366809,
            214486,
            153627,
            171751
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -8,
      "raw_net_mean": -9.6,
      "raw_net_min": -10,
      "raw_net_sample_sd": 0.8944271909999159,
      "raw_net_values": [
        -10,
        -10,
        -10,
        -8,
        -10
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 146441.6,
          "sample_sd": 52209.737562067865,
          "values": [
            225474,
            118526,
            97694,
            117881,
            172633
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1623.8,
          "sample_sd": 325.5398593106534,
          "values": [
            1971,
            1500,
            1146,
            1634,
            1868
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 148065.4,
          "sample_sd": 52498.6550808685,
          "values": [
            227445,
            120026,
            98840,
            119515,
            174501
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -8,
      "raw_net_mean": -8.8,
      "raw_net_min": -10,
      "raw_net_sample_sd": 1.0954451150103321,
      "raw_net_values": [
        -10,
        -8,
        -8,
        -10,
        -8
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 212200,
          "sample_sd": 21898.185004698447,
          "values": [
            216325,
            212288,
            190320,
            246196,
            195871
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1955.8,
          "sample_sd": 177.0838784305336,
          "values": [
            2137,
            2088,
            1695,
            1876,
            1983
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 214155.8,
          "sample_sd": 21936.63703031985,
          "values": [
            218462,
            214376,
            192015,
            248072,
            197854
          ]
        }
      }
    },
    {
      "arm": "task_type_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -8,
      "raw_net_mean": -9.2,
      "raw_net_min": -10,
      "raw_net_sample_sd": 1.0954451150103321,
      "raw_net_values": [
        -10,
        -8,
        -10,
        -10,
        -8
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 152536,
          "sample_sd": 30404.66734730048,
          "values": [
            156349,
            133188,
            204028,
            136904,
            132211
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1469.6,
          "sample_sd": 273.1882135085626,
          "values": [
            1735,
            1291,
            1701,
            1525,
            1096
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 154005.6,
          "sample_sd": 30595.318560198062,
          "values": [
            158084,
            134479,
            205729,
            138429,
            133307
          ]
        }
      }
    },
    {
      "arm": "task_type_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -8,
      "raw_net_mean": -9.4,
      "raw_net_min": -10,
      "raw_net_sample_sd": 0.8944271909999159,
      "raw_net_values": [
        -10,
        -9,
        -10,
        -8,
        -10
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 160689.2,
          "sample_sd": 49976.14713740946,
          "values": [
            116383,
            153219,
            245642,
            154208,
            133994
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1490.4,
          "sample_sd": 288.53128079984674,
          "values": [
            1317,
            1517,
            1928,
            1530,
            1160
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 162179.6,
          "sample_sd": 50242.51107180054,
          "values": [
            117700,
            154736,
            247570,
            155738,
            135154
          ]
        }
      }
    },
    {
      "arm": "delete_first_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -8,
      "raw_net_mean": -8.4,
      "raw_net_min": -10,
      "raw_net_sample_sd": 0.8944271909999159,
      "raw_net_values": [
        -8,
        -10,
        -8,
        -8,
        -8
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 166910.4,
          "sample_sd": 107667.30414243686,
          "values": [
            156013,
            355408,
            102023,
            101710,
            119398
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2716.8,
          "sample_sd": 1158.811546369814,
          "values": [
            2876,
            4631,
            1959,
            1711,
            2407
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 169627.2,
          "sample_sd": 108801.97439936464,
          "values": [
            158889,
            360039,
            103982,
            103421,
            121805
          ]
        }
      }
    },
    {
      "arm": "delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -8,
      "raw_net_mean": -8,
      "raw_net_min": -8,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -8,
        -8,
        -8,
        -8,
        -8
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5,
        "failed": 1
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 120532.4,
          "sample_sd": 31840.56728923026,
          "values": [
            102458,
            176205,
            102720,
            118260,
            103019
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2142.4,
          "sample_sd": 258.2117348224127,
          "values": [
            1886,
            2556,
            2035,
            2025,
            2210
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 122674.8,
          "sample_sd": 32065.816584019813,
          "values": [
            104344,
            178761,
            104755,
            120285,
            105229
          ]
        }
      }
    },
    {
      "arm": "neutral_control",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -8,
      "raw_net_mean": -11.8,
      "raw_net_min": -21,
      "raw_net_sample_sd": 5.215361924162119,
      "raw_net_values": [
        -10,
        -8,
        -21,
        -10,
        -10
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 173266,
          "sample_sd": 130999.92070799127,
          "values": [
            102184,
            135803,
            406283,
            102538,
            119522
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2580.2,
          "sample_sd": 968.9706909912188,
          "values": [
            2032,
            2220,
            4309,
            2164,
            2176
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 175846.2,
          "sample_sd": 131966.4686433641,
          "values": [
            104216,
            138023,
            410592,
            104702,
            121698
          ]
        }
      }
    },
    {
      "arm": "semantic_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -8,
      "raw_net_mean": -11,
      "raw_net_min": -21,
      "raw_net_sample_sd": 5.656854249492381,
      "raw_net_values": [
        -21,
        -8,
        -8,
        -8,
        -10
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 143625.8,
          "sample_sd": 51064.12132505562,
          "values": [
            214278,
            105465,
            135975,
            173721,
            88690
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2810,
          "sample_sd": 539.122898790248,
          "values": [
            3404,
            3041,
            2544,
            3045,
            2016
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 146435.8,
          "sample_sd": 51485.67301881175,
          "values": [
            217682,
            108506,
            138519,
            176766,
            90706
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -8,
      "raw_net_mean": -8.4,
      "raw_net_min": -10,
      "raw_net_sample_sd": 0.8944271909999159,
      "raw_net_values": [
        -8,
        -8,
        -8,
        -8,
        -10
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 122498.4,
          "sample_sd": 34524.929359811875,
          "values": [
            103022,
            102795,
            120460,
            103442,
            182773
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2556,
          "sample_sd": 586.2273449780383,
          "values": [
            2238,
            2111,
            2682,
            2220,
            3529
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 125054.4,
          "sample_sd": 35102.58983038146,
          "values": [
            105260,
            104906,
            123142,
            105662,
            186302
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -8,
      "raw_net_mean": -8.8,
      "raw_net_min": -10,
      "raw_net_sample_sd": 1.0954451150103321,
      "raw_net_values": [
        -8,
        -10,
        -10,
        -8,
        -8
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5,
        "failed": 1
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 122882.6,
          "sample_sd": 24172.790163735754,
          "values": [
            163919,
            107511,
            120947,
            119134,
            102902
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2506,
          "sample_sd": 390.5553737948052,
          "values": [
            2784,
            2911,
            2655,
            2106,
            2074
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 125388.6,
          "sample_sd": 24330.823163222405,
          "values": [
            166703,
            110422,
            123602,
            121240,
            104976
          ]
        }
      }
    },
    {
      "arm": "task_type_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -8,
      "raw_net_mean": -9.2,
      "raw_net_min": -10,
      "raw_net_sample_sd": 1.0954451150103321,
      "raw_net_values": [
        -8,
        -8,
        -10,
        -10,
        -10
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 145251.6,
          "sample_sd": 66555.11774687203,
          "values": [
            138617,
            103396,
            117996,
            104668,
            261581
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2847.2,
          "sample_sd": 641.9265534311538,
          "values": [
            3096,
            2380,
            2340,
            2559,
            3861
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 148098.8,
          "sample_sd": 67161.56178723065,
          "values": [
            141713,
            105776,
            120336,
            107227,
            265442
          ]
        }
      }
    },
    {
      "arm": "task_type_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -8,
      "raw_net_mean": -9.6,
      "raw_net_min": -10,
      "raw_net_sample_sd": 0.8944271909999159,
      "raw_net_values": [
        -10,
        -10,
        -8,
        -10,
        -10
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5,
        "failed": 2
      },
      "task_id": "refactor-dead-compatibility-path",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 139102,
          "sample_sd": 73151.57254974086,
          "values": [
            118920,
            103089,
            102627,
            269330,
            101544
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2314.4,
          "sample_sd": 662.6600184106477,
          "values": [
            2365,
            2090,
            1949,
            3428,
            1740
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 141416.4,
          "sample_sd": 73790.80789163377,
          "values": [
            121285,
            105179,
            104576,
            272758,
            103284
          ]
        }
      }
    },
    {
      "arm": "delete_first_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -3,
      "raw_net_mean": -3,
      "raw_net_min": -3,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -3,
        -3,
        -3,
        -3,
        -3
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 443944,
          "sample_sd": 164910.18928647193,
          "values": [
            723138,
            375187,
            458389,
            348968,
            314038
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 3538.2,
          "sample_sd": 330.67158934507813,
          "values": [
            3826,
            3843,
            3601,
            3357,
            3064
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 447482.2,
          "sample_sd": 165121.27796017085,
          "values": [
            726964,
            379030,
            461990,
            352325,
            317102
          ]
        }
      }
    },
    {
      "arm": "delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -3,
      "raw_net_mean": -3,
      "raw_net_min": -3,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -3,
        -3,
        -3,
        -3,
        -3
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 434864.2,
          "sample_sd": 269406.1326245192,
          "values": [
            392797,
            906885,
            334773,
            295648,
            244218
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 3925,
          "sample_sd": 1051.5835677681541,
          "values": [
            3458,
            5628,
            3784,
            2792,
            3963
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 438789.2,
          "sample_sd": 270323.54439670994,
          "values": [
            396255,
            912513,
            338557,
            298440,
            248181
          ]
        }
      }
    },
    {
      "arm": "neutral_control",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -3,
      "raw_net_mean": -3,
      "raw_net_min": -3,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -3,
        -3,
        -3,
        -3,
        -3
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 372906.8,
          "sample_sd": 72805.67988282233,
          "values": [
            316641,
            389032,
            477227,
            291651,
            389983
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2572.4,
          "sample_sd": 306.1899737091337,
          "values": [
            2283,
            2710,
            3023,
            2312,
            2534
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 375479.2,
          "sample_sd": 73101.07272605512,
          "values": [
            318924,
            391742,
            480250,
            293963,
            392517
          ]
        }
      }
    },
    {
      "arm": "semantic_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -3,
      "raw_net_mean": -3,
      "raw_net_min": -3,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -3,
        -3,
        -3,
        -3,
        -3
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 341412.2,
          "sample_sd": 84770.21613573957,
          "values": [
            254315,
            381070,
            248011,
            390113,
            433552
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2848.2,
          "sample_sd": 712.3452112564526,
          "values": [
            2357,
            2275,
            2460,
            3218,
            3931
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 344260.4,
          "sample_sd": 85285.05475638741,
          "values": [
            256672,
            383345,
            250471,
            393331,
            437483
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -3,
      "raw_net_mean": -3,
      "raw_net_min": -3,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -3,
        -3,
        -3,
        -3,
        -3
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 403348.2,
          "sample_sd": 137845.44570895334,
          "values": [
            443667,
            392086,
            329982,
            609743,
            241263
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 3234.4,
          "sample_sd": 704.7842222978604,
          "values": [
            2937,
            3537,
            2899,
            4305,
            2494
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 406582.6,
          "sample_sd": 138480.88256615063,
          "values": [
            446604,
            395623,
            332881,
            614048,
            243757
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -3,
      "raw_net_mean": -3,
      "raw_net_min": -3,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -3,
        -3,
        -3,
        -3,
        -3
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 409341.8,
          "sample_sd": 102922.67643090127,
          "values": [
            346525,
            377102,
            592283,
            370767,
            360032
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 3101.4,
          "sample_sd": 513.2984511957931,
          "values": [
            3540,
            3109,
            3445,
            3171,
            2242
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 412443.2,
          "sample_sd": 103105.3727004563,
          "values": [
            350065,
            380211,
            595728,
            373938,
            362274
          ]
        }
      }
    },
    {
      "arm": "task_type_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": -3,
      "raw_net_mean": -3,
      "raw_net_min": -3,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -3,
        -3,
        -3,
        -3,
        -3
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 339044.6,
          "sample_sd": 177267.1179076932,
          "values": [
            234062,
            655001,
            270609,
            268944,
            266607
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2560.2,
          "sample_sd": 476.8644461479593,
          "values": [
            2375,
            3353,
            2450,
            2547,
            2076
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 341604.8,
          "sample_sd": 177709.3008404456,
          "values": [
            236437,
            658354,
            273059,
            271491,
            268683
          ]
        }
      }
    },
    {
      "arm": "task_type_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "composer-2.5",
      "raw_net_max": 10,
      "raw_net_mean": -0.4,
      "raw_net_min": -3,
      "raw_net_sample_sd": 5.813776741499453,
      "raw_net_values": [
        -3,
        10,
        -3,
        -3,
        -3
      ],
      "reasoning_effort": "default",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 313346,
          "sample_sd": 215360.25957450925,
          "values": [
            664911,
            100303,
            230323,
            221949,
            349244
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2691.8,
          "sample_sd": 450.4982796859495,
          "values": [
            3309,
            2105,
            2885,
            2470,
            2690
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 316037.8,
          "sample_sd": 215762.2017991103,
          "values": [
            668220,
            102408,
            233208,
            224419,
            351934
          ]
        }
      }
    },
    {
      "arm": "delete_first_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -5,
      "raw_net_mean": -5.8,
      "raw_net_min": -6,
      "raw_net_sample_sd": 0.4472135954999579,
      "raw_net_values": [
        -5,
        -6,
        -6,
        -6,
        -6
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 174946.6,
          "sample_sd": 40350.25979222439,
          "values": [
            207303,
            150792,
            209249,
            116263,
            191126
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1846.6,
          "sample_sd": 328.72222316113647,
          "values": [
            2079,
            1559,
            2026,
            1427,
            2142
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 176793.2,
          "sample_sd": 40660.339530062956,
          "values": [
            209382,
            152351,
            211275,
            117690,
            193268
          ]
        }
      }
    },
    {
      "arm": "delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -3,
      "raw_net_mean": -4.2,
      "raw_net_min": -5,
      "raw_net_sample_sd": 1.0954451150103321,
      "raw_net_values": [
        -3,
        -5,
        -5,
        -5,
        -3
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5,
        "failed": 1
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 211069,
          "sample_sd": 71627.32192746006,
          "values": [
            153894,
            116273,
            245749,
            260088,
            279341
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1871.6,
          "sample_sd": 376.7569508317,
          "values": [
            1547,
            1382,
            2159,
            2161,
            2109
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 212940.6,
          "sample_sd": 71995.0082109864,
          "values": [
            155441,
            117655,
            247908,
            262249,
            281450
          ]
        }
      }
    },
    {
      "arm": "neutral_control",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": 6,
      "raw_net_mean": 0.6,
      "raw_net_min": -6,
      "raw_net_sample_sd": 6.06630035524124,
      "raw_net_values": [
        -6,
        4,
        -6,
        5,
        6
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 132463.4,
          "sample_sd": 20572.108601210522,
          "values": [
            131863,
            98853,
            150448,
            147916,
            133237
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1301.4,
          "sample_sd": 114.60279228709918,
          "values": [
            1238,
            1176,
            1480,
            1327,
            1286
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 133764.8,
          "sample_sd": 20667.58457585211,
          "values": [
            133101,
            100029,
            151928,
            149243,
            134523
          ]
        }
      }
    },
    {
      "arm": "semantic_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": 6,
      "raw_net_mean": -2.8,
      "raw_net_min": -5,
      "raw_net_sample_sd": 4.919349550499537,
      "raw_net_values": [
        -5,
        -5,
        6,
        -5,
        -5
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 159827.4,
          "sample_sd": 16916.30919556627,
          "values": [
            174735,
            132966,
            153784,
            167730,
            169922
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1645.2,
          "sample_sd": 253.43184488141975,
          "values": [
            2028,
            1484,
            1615,
            1369,
            1730
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 161472.6,
          "sample_sd": 17049.10570381919,
          "values": [
            176763,
            134450,
            155399,
            169099,
            171652
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -3,
      "raw_net_mean": -4.4,
      "raw_net_min": -6,
      "raw_net_sample_sd": 1.3416407864998738,
      "raw_net_values": [
        -3,
        -6,
        -3,
        -5,
        -5
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 207750.2,
          "sample_sd": 47748.881077989674,
          "values": [
            169779,
            174340,
            270022,
            175956,
            248654
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2132.8,
          "sample_sd": 395.53722960045116,
          "values": [
            1610,
            1942,
            2683,
            2213,
            2216
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 209883,
          "sample_sd": 48070.60142644359,
          "values": [
            171389,
            176282,
            272705,
            178169,
            250870
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -3,
      "raw_net_mean": -5,
      "raw_net_min": -6,
      "raw_net_sample_sd": 1.224744871391589,
      "raw_net_values": [
        -3,
        -5,
        -6,
        -6,
        -5
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 201137.6,
          "sample_sd": 62997.758379167746,
          "values": [
            115077,
            283013,
            170133,
            228335,
            209130
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1886.6,
          "sample_sd": 553.1331666063788,
          "values": [
            931,
            2296,
            1899,
            2168,
            2139
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 203024.2,
          "sample_sd": 63494.45844087498,
          "values": [
            116008,
            285309,
            172032,
            230503,
            211269
          ]
        }
      }
    },
    {
      "arm": "task_type_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": 7,
      "raw_net_mean": 0.4,
      "raw_net_min": -6,
      "raw_net_sample_sd": 5.639148871948674,
      "raw_net_values": [
        -5,
        3,
        -6,
        7,
        3
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 190557,
          "sample_sd": 6477.20510714305,
          "values": [
            189026,
            192631,
            184480,
            185944,
            200704
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1792.2,
          "sample_sd": 380.8118958225964,
          "values": [
            1967,
            2365,
            1466,
            1696,
            1467
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 192349.2,
          "sample_sd": 6490.187185282101,
          "values": [
            190993,
            194996,
            185946,
            187640,
            202171
          ]
        }
      }
    },
    {
      "arm": "task_type_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "gpt-5.6-luna",
      "raw_net_max": -3,
      "raw_net_mean": -5.2,
      "raw_net_min": -6,
      "raw_net_sample_sd": 1.3038404810405297,
      "raw_net_values": [
        -6,
        -6,
        -5,
        -6,
        -3
      ],
      "reasoning_effort": "maximum",
      "status_counts": {
        "completed": 5,
        "failed": 1
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 205105.4,
          "sample_sd": 42064.43891578729,
          "values": [
            211411,
            204802,
            267871,
            151567,
            189876
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 1945.8,
          "sample_sd": 436.05297843266703,
          "values": [
            2030,
            1756,
            2649,
            1498,
            1796
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 207051.2,
          "sample_sd": 42486.44179853144,
          "values": [
            213441,
            206558,
            270520,
            153065,
            191672
          ]
        }
      }
    },
    {
      "arm": "delete_first_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": 1,
      "raw_net_mean": -4,
      "raw_net_min": -6,
      "raw_net_sample_sd": 3.082207001484488,
      "raw_net_values": [
        -6,
        1,
        -6,
        -6,
        -3
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 131027.2,
          "sample_sd": 28405.343629324394,
          "values": [
            138829,
            117554,
            119380,
            176425,
            102948
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2795.6,
          "sample_sd": 574.2571723539899,
          "values": [
            3456,
            2277,
            2455,
            3384,
            2406
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 133822.8,
          "sample_sd": 28880.312536397523,
          "values": [
            142285,
            119831,
            121835,
            179809,
            105354
          ]
        }
      }
    },
    {
      "arm": "delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": 1,
      "raw_net_mean": -3.4,
      "raw_net_min": -6,
      "raw_net_sample_sd": 2.8809720581775866,
      "raw_net_values": [
        -6,
        1,
        -3,
        -6,
        -3
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 127655.2,
          "sample_sd": 29479.755523070406,
          "values": [
            153610,
            104379,
            88378,
            141037,
            150872
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2690.8,
          "sample_sd": 501.4047267427781,
          "values": [
            2527,
            2787,
            1910,
            3116,
            3114
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 130346,
          "sample_sd": 29818.67826547649,
          "values": [
            156137,
            107166,
            90288,
            144153,
            153986
          ]
        }
      }
    },
    {
      "arm": "neutral_control",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": 9,
      "raw_net_mean": 6,
      "raw_net_min": -6,
      "raw_net_sample_sd": 6.708203932499369,
      "raw_net_values": [
        9,
        9,
        9,
        -6,
        9
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 121470.8,
          "sample_sd": 27568.555769209237,
          "values": [
            101341,
            102403,
            150801,
            152504,
            100305
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2357.6,
          "sample_sd": 543.1669172547238,
          "values": [
            2149,
            2309,
            3042,
            2680,
            1608
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 123828.4,
          "sample_sd": 28033.113835248485,
          "values": [
            103490,
            104712,
            153843,
            155184,
            101913
          ]
        }
      }
    },
    {
      "arm": "semantic_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -3,
      "raw_net_mean": -3.6,
      "raw_net_min": -6,
      "raw_net_sample_sd": 1.3416407864998738,
      "raw_net_values": [
        -3,
        -3,
        -3,
        -3,
        -6
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 124306.8,
          "sample_sd": 33674.78033632885,
          "values": [
            136759,
            102087,
            102048,
            102380,
            178260
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2674.4,
          "sample_sd": 632.2078772049585,
          "values": [
            3382,
            2027,
            1983,
            3019,
            2961
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 126981.2,
          "sample_sd": 34024.0654272825,
          "values": [
            140141,
            104114,
            104031,
            105399,
            181221
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": 1,
      "raw_net_mean": -2,
      "raw_net_min": -6,
      "raw_net_sample_sd": 3.0,
      "raw_net_values": [
        -3,
        -6,
        1,
        -3,
        1
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 115261.8,
          "sample_sd": 8014.609110867479,
          "values": [
            119024,
            119631,
            118856,
            100972,
            117826
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2300.8,
          "sample_sd": 396.27351665232436,
          "values": [
            2658,
            2624,
            2428,
            1733,
            2061
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 117562.6,
          "sample_sd": 8351.467014842363,
          "values": [
            121682,
            122255,
            121284,
            102705,
            119887
          ]
        }
      }
    },
    {
      "arm": "task_type_delete_first_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -3,
      "raw_net_mean": -3,
      "raw_net_min": -3,
      "raw_net_sample_sd": 0.0,
      "raw_net_values": [
        -3,
        -3,
        -3,
        -3,
        -3
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 134682.8,
          "sample_sd": 46341.8705567654,
          "values": [
            103485,
            102390,
            205466,
            103312,
            158761
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2621.4,
          "sample_sd": 433.0303684500661,
          "values": [
            2304,
            2231,
            2870,
            2445,
            3257
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 137304.2,
          "sample_sd": 46671.49489463563,
          "values": [
            105789,
            104621,
            208336,
            105757,
            162018
          ]
        }
      }
    },
    {
      "arm": "task_type_gate",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": 9,
      "raw_net_mean": 1.8,
      "raw_net_min": -3,
      "raw_net_sample_sd": 6.572670690061994,
      "raw_net_values": [
        9,
        -3,
        -3,
        -3,
        9
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 102091.6,
          "sample_sd": 1483.1821533446255,
          "values": [
            104107,
            100152,
            102773,
            101378,
            102048
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2073,
          "sample_sd": 478.1359639265802,
          "values": [
            2707,
            1418,
            2288,
            1880,
            2072
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 104164.6,
          "sample_sd": 1960.7852508625213,
          "values": [
            106814,
            101570,
            105061,
            103258,
            104120
          ]
        }
      }
    },
    {
      "arm": "task_type_net_loc_budget",
      "behavior_failure_count": 0,
      "behavior_pass_count": 5,
      "completed_count": 5,
      "completed_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "model": "grok-4.5",
      "raw_net_max": -3,
      "raw_net_mean": -3.6,
      "raw_net_min": -6,
      "raw_net_sample_sd": 1.3416407864998738,
      "raw_net_values": [
        -6,
        -3,
        -3,
        -3,
        -3
      ],
      "reasoning_effort": "high",
      "status_counts": {
        "completed": 5
      },
      "task_id": "refactor-shared-normalizer",
      "token_stats": {
        "input_tokens": {
          "count": 5,
          "mean": 115151.4,
          "sample_sd": 31261.271548035278,
          "values": [
            152505,
            88183,
            144817,
            87901,
            102351
          ]
        },
        "output_tokens": {
          "count": 5,
          "mean": 2325,
          "sample_sd": 504.73408840695515,
          "values": [
            2425,
            2034,
            3157,
            1889,
            2120
          ]
        },
        "total_tokens": {
          "count": 5,
          "mean": 117476.4,
          "sample_sd": 31668.706246072004,
          "values": [
            154930,
            90217,
            147974,
            89790,
            104471
          ]
        }
      }
    }
  ],
  "paired_comparisons": [
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "delete_first_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "semantic_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_delete_first",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0.8,
      "raw_net_delta_sample_sd": 1.0954451150103321,
      "raw_net_deltas": [
        0,
        0,
        2,
        0,
        2
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "delete_first_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0.8,
      "raw_net_delta_sample_sd": 1.0954451150103321,
      "raw_net_deltas": [
        0,
        2,
        0,
        0,
        2
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0.6,
      "raw_net_delta_sample_sd": 0.8944271909999159,
      "raw_net_deltas": [
        0,
        2,
        0,
        1,
        0
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 1,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "semantic_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0.4,
      "raw_net_delta_sample_sd": 0.8944271909999159,
      "raw_net_deltas": [
        0,
        0,
        0,
        2,
        0
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_delete_first",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 1.2,
      "raw_net_delta_sample_sd": 1.0954451150103321,
      "raw_net_deltas": [
        0,
        2,
        2,
        0,
        2
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0.8,
      "raw_net_delta_sample_sd": 1.0954451150103321,
      "raw_net_deltas": [
        0,
        2,
        0,
        0,
        2
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0.6,
      "raw_net_delta_sample_sd": 0.8944271909999159,
      "raw_net_deltas": [
        0,
        1,
        0,
        2,
        0
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 1,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 3.4,
      "raw_net_delta_sample_sd": 5.639148871948674,
      "raw_net_deltas": [
        2,
        -2,
        13,
        2,
        2
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -2,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 13,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "delete_first_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 3.8,
      "raw_net_delta_sample_sd": 5.215361924162119,
      "raw_net_deltas": [
        2,
        0,
        13,
        2,
        2
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 13,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0.8,
      "raw_net_delta_sample_sd": 8.526429498916883,
      "raw_net_deltas": [
        -11,
        0,
        13,
        2,
        0
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -11,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 13,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "semantic_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 3.4,
      "raw_net_delta_sample_sd": 5.458937625582473,
      "raw_net_deltas": [
        2,
        0,
        13,
        2,
        0
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 13,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_delete_first",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 3,
      "raw_net_delta_sample_sd": 4.795831523312719,
      "raw_net_deltas": [
        2,
        -2,
        11,
        2,
        2
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -2,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 11,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 2.6,
      "raw_net_delta_sample_sd": 4.774934554525329,
      "raw_net_deltas": [
        2,
        0,
        11,
        0,
        0
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 11,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 2.2,
      "raw_net_delta_sample_sd": 6.099180272790763,
      "raw_net_deltas": [
        0,
        -2,
        13,
        0,
        0
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -2,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 13,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-dead-compatibility-path",
      "treatment_arm": "task_type_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "delete_first_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "semantic_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_delete_first",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 0,
      "raw_net_delta_sample_sd": 0.0,
      "raw_net_deltas": [
        0,
        0,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "composer-2.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": 2.6,
      "raw_net_delta_sample_sd": 5.813776741499453,
      "raw_net_deltas": [
        0,
        13,
        0,
        0,
        0
      ],
      "reasoning_effort": "default",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 13,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -6.4,
      "raw_net_delta_sample_sd": 6.348228099241552,
      "raw_net_deltas": [
        1,
        -10,
        0,
        -11,
        -12
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 1,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -10,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -11,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "delete_first_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -4.8,
      "raw_net_delta_sample_sd": 6.260990336999411,
      "raw_net_deltas": [
        3,
        -9,
        1,
        -10,
        -9
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 3,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -9,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 1,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -10,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -9,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -3.4,
      "raw_net_delta_sample_sd": 9.864076236526156,
      "raw_net_deltas": [
        1,
        -9,
        12,
        -10,
        -11
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 1,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -9,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 12,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -10,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -11,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "semantic_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -5,
      "raw_net_delta_sample_sd": 7.314369419163897,
      "raw_net_deltas": [
        3,
        -10,
        3,
        -10,
        -11
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 3,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -10,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 3,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -10,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -11,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_delete_first",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -5.6,
      "raw_net_delta_sample_sd": 6.618156843109719,
      "raw_net_deltas": [
        3,
        -9,
        0,
        -11,
        -11
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 3,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -9,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -11,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -11,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -0.2,
      "raw_net_delta_sample_sd": 1.9235384061671346,
      "raw_net_deltas": [
        1,
        -1,
        0,
        2,
        -3
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 1,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -1,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 2,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -3,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "gpt-5.6-luna",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -5.8,
      "raw_net_delta_sample_sd": 5.805170109479997,
      "raw_net_deltas": [
        0,
        -10,
        1,
        -11,
        -9
      ],
      "reasoning_effort": "maximum",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -10,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 1,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -11,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -9,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -10,
      "raw_net_delta_sample_sd": 6.284902544988268,
      "raw_net_deltas": [
        -15,
        -8,
        -15,
        0,
        -12
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -15,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -8,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -15,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "delete_first_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -9.4,
      "raw_net_delta_sample_sd": 5.813776741499453,
      "raw_net_deltas": [
        -15,
        -8,
        -12,
        0,
        -12
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -15,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -8,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -9.6,
      "raw_net_delta_sample_sd": 7.162401831787993,
      "raw_net_deltas": [
        -12,
        -12,
        -12,
        3,
        -15
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 3,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -15,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "semantic_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -8,
      "raw_net_delta_sample_sd": 6.819090848492928,
      "raw_net_deltas": [
        -12,
        -15,
        -8,
        3,
        -8
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -15,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -8,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 3,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -8,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_delete_first",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -9,
      "raw_net_delta_sample_sd": 6.708203932499369,
      "raw_net_deltas": [
        -12,
        -12,
        -12,
        3,
        -12
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 3,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_delete_first_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -4.2,
      "raw_net_delta_sample_sd": 7.224956747275377,
      "raw_net_deltas": [
        0,
        -12,
        -12,
        3,
        0
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 3,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 0,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_gate",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    },
    {
      "both_behavior_pass_count": 5,
      "model": "grok-4.5",
      "neutral_arm": "neutral_control",
      "neutral_only_behavior_pass_count": 0,
      "neutral_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ],
      "raw_net_delta_mean": -9.6,
      "raw_net_delta_sample_sd": 7.162401831787993,
      "raw_net_deltas": [
        -15,
        -12,
        -12,
        3,
        -12
      ],
      "reasoning_effort": "high",
      "repetitions": [
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -15,
          "repetition": 1,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 2,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 3,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": 3,
          "repetition": 4,
          "treatment_behavior_pass": true
        },
        {
          "neutral_behavior_pass": true,
          "raw_net_delta": -12,
          "repetition": 5,
          "treatment_behavior_pass": true
        }
      ],
      "status": "matched",
      "task_id": "refactor-shared-normalizer",
      "treatment_arm": "task_type_net_loc_budget",
      "treatment_only_behavior_pass_count": 0,
      "treatment_repetition_ids": [
        1,
        2,
        3,
        4,
        5
      ]
    }
  ],
  "protocol": "phase-2-controlled-ablation-variance-v1",
  "status_counts": {
    "completed": 240,
    "failed": 6
  }
}