{
  "schema_version": 1,
  "name": "优先级任务揭示与联盟重组",
  "state": "completed",
  "created_at": "2026-09-05T16:56:09",
  "finished_at": "2026-09-05T17:53:41",
  "published_at": "2026-09-06T14:35:57+08:00",
  "jobs": {
    "total": 15,
    "completed": 15,
    "failed": 0
  },
  "raw_evaluation_rows": 1440,
  "paired_effect_rows": 960,
  "all_paired_states_verified": true,
  "entries": [
    {
      "training_condition": "hattv20_full",
      "model_role": "architecture_baseline",
      "ablation": "none",
      "contrast": "priority_high_minus_low",
      "training_seeds": 3,
      "eval_scenarios_per_seed": 32,
      "paired_state_check": true,
      "metrics": {
        "mean_event_task_probability": {
          "mean": 0.08811127260330316,
          "training_seed_sd": 0.03435457709955428
        },
        "executed_target_assignments": {
          "mean": 1.3333333333333333,
          "training_seed_sd": 0.7795815517528191
        },
        "executed_switches": {
          "mean": 0.8020833333333334,
          "training_seed_sd": 0.3512440687518201
        },
        "executed_non_target_switches": {
          "mean": -0.53125,
          "training_seed_sd": 0.4341388746703064
        },
        "executed_fragile_departures": {
          "mean": 0.28125,
          "training_seed_sd": 0.11267347735824966
        },
        "event_task_response_delay": {
          "mean": -58.552083333333336,
          "training_seed_sd": 27.505935675701586
        },
        "task_completion_ratio": {
          "mean": -0.0010416666666666664,
          "training_seed_sd": 0.00180421959121758
        },
        "priority_weighted_completion": {
          "mean": -0.00098987318074308,
          "training_seed_sd": 0.0017145106420968254
        },
        "remaining_demand": {
          "mean": 0.3645833333333333,
          "training_seed_sd": 0.6314768569261532
        },
        "team_reward_after_event": {
          "mean": 36.34647961406447,
          "training_seed_sd": 7.794801819588642
        }
      }
    },
    {
      "training_condition": "hattv20_full",
      "model_role": "architecture_baseline",
      "ablation": "none",
      "contrast": "react_minus_keep_current",
      "training_seeds": 3,
      "eval_scenarios_per_seed": 32,
      "paired_state_check": true,
      "metrics": {
        "mean_event_task_probability": {
          "mean": -9.791989441509941e-11,
          "training_seed_sd": 1.6769527927866735e-10
        },
        "executed_target_assignments": {
          "mean": 2.9270833333333335,
          "training_seed_sd": 0.8222571196610786
        },
        "executed_switches": {
          "mean": 5.333333333333333,
          "training_seed_sd": 0.8387195722250276
        },
        "executed_non_target_switches": {
          "mean": 2.40625,
          "training_seed_sd": 0.125
        },
        "executed_fragile_departures": {
          "mean": 1.3020833333333333,
          "training_seed_sd": 0.2303473373697498
        },
        "event_task_response_delay": {
          "mean": -9.25,
          "training_seed_sd": 3.466778719142599
        },
        "task_completion_ratio": {
          "mean": -0.0010416666666666664,
          "training_seed_sd": 0.00180421959121758
        },
        "priority_weighted_completion": {
          "mean": -0.00098987318074308,
          "training_seed_sd": 0.0017145106420968254
        },
        "remaining_demand": {
          "mean": 0.3645833333333333,
          "training_seed_sd": 0.6314768569261532
        },
        "team_reward_after_event": {
          "mean": -8.31980987047732,
          "training_seed_sd": 14.416619636622462
        }
      }
    },
    {
      "training_condition": "mappo64",
      "model_role": "algorithm_baseline",
      "ablation": "none",
      "contrast": "priority_high_minus_low",
      "training_seeds": 3,
      "eval_scenarios_per_seed": 32,
      "paired_state_check": true,
      "metrics": {
        "mean_event_task_probability": {
          "mean": -0.0011910751588099326,
          "training_seed_sd": 0.0012450695897637005
        },
        "executed_target_assignments": {
          "mean": 0.0,
          "training_seed_sd": 0.0
        },
        "executed_switches": {
          "mean": 0.010416666666666666,
          "training_seed_sd": 0.018042195912175808
        },
        "executed_non_target_switches": {
          "mean": 0.010416666666666666,
          "training_seed_sd": 0.018042195912175808
        },
        "executed_fragile_departures": {
          "mean": 0.010416666666666666,
          "training_seed_sd": 0.018042195912175808
        },
        "event_task_response_delay": {
          "mean": 0.03125,
          "training_seed_sd": 0.05412658773652741
        },
        "task_completion_ratio": {
          "mean": 0.0010416666666666664,
          "training_seed_sd": 0.00180421959121758
        },
        "priority_weighted_completion": {
          "mean": 0.0011419578276609052,
          "training_seed_sd": 0.001977928977609672
        },
        "remaining_demand": {
          "mean": -0.03125,
          "training_seed_sd": 0.05412658773652741
        },
        "team_reward_after_event": {
          "mean": 33.204025388529935,
          "training_seed_sd": 4.8533940789691075
        }
      }
    },
    {
      "training_condition": "mappo64",
      "model_role": "algorithm_baseline",
      "ablation": "none",
      "contrast": "react_minus_keep_current",
      "training_seeds": 3,
      "eval_scenarios_per_seed": 32,
      "paired_state_check": true,
      "metrics": {
        "mean_event_task_probability": {
          "mean": 0.0,
          "training_seed_sd": 0.0
        },
        "executed_target_assignments": {
          "mean": 5.177083333333333,
          "training_seed_sd": 4.484819591782186
        },
        "executed_switches": {
          "mean": 5.729166666666667,
          "training_seed_sd": 3.8795755915864474
        },
        "executed_non_target_switches": {
          "mean": 0.5520833333333334,
          "training_seed_sd": 0.6062285219563109
        },
        "executed_fragile_departures": {
          "mean": 0.8541666666666666,
          "training_seed_sd": 0.4532147899543144
        },
        "event_task_response_delay": {
          "mean": -5.770833333333333,
          "training_seed_sd": 3.2604000265816055
        },
        "task_completion_ratio": {
          "mean": 0.0,
          "training_seed_sd": 0.0
        },
        "priority_weighted_completion": {
          "mean": 0.0,
          "training_seed_sd": 0.0
        },
        "remaining_demand": {
          "mean": 0.0,
          "training_seed_sd": 0.0
        },
        "team_reward_after_event": {
          "mean": -8.650994882308163,
          "training_seed_sd": 19.66828750080269
        }
      }
    },
    {
      "training_condition": "v33aa_learned",
      "model_role": "inference_ablation",
      "ablation": "zero_AA_message_at_inference",
      "contrast": "priority_high_minus_low",
      "training_seeds": 3,
      "eval_scenarios_per_seed": 32,
      "paired_state_check": true,
      "metrics": {
        "mean_event_task_probability": {
          "mean": 0.06333472383260859,
          "training_seed_sd": 0.01647916034700583
        },
        "executed_target_assignments": {
          "mean": 0.6666666666666666,
          "training_seed_sd": 0.19846415503393386
        },
        "executed_switches": {
          "mean": 0.17708333333333334,
          "training_seed_sd": 0.036084391824351615
        },
        "executed_non_target_switches": {
          "mean": -0.4895833333333333,
          "training_seed_sd": 0.18310828990882236
        },
        "executed_fragile_departures": {
          "mean": 0.09375,
          "training_seed_sd": 0.05412658773652741
        },
        "event_task_response_delay": {
          "mean": -22.416666666666668,
          "training_seed_sd": 6.911008918445507
        },
        "task_completion_ratio": {
          "mean": 0.0,
          "training_seed_sd": 0.0
        },
        "priority_weighted_completion": {
          "mean": 0.0004014828335411405,
          "training_seed_sd": 0.000377223974983969
        },
        "remaining_demand": {
          "mean": 2.3854166666666665,
          "training_seed_sd": 2.3906420206365766
        },
        "team_reward_after_event": {
          "mean": 27.82864982871384,
          "training_seed_sd": 14.262045736882436
        }
      }
    },
    {
      "training_condition": "v33aa_learned",
      "model_role": "inference_ablation",
      "ablation": "zero_AA_message_at_inference",
      "contrast": "react_minus_keep_current",
      "training_seeds": 3,
      "eval_scenarios_per_seed": 32,
      "paired_state_check": true,
      "metrics": {
        "mean_event_task_probability": {
          "mean": 0.0,
          "training_seed_sd": 0.0
        },
        "executed_target_assignments": {
          "mean": 4.072916666666667,
          "training_seed_sd": 0.15728821740147397
        },
        "executed_switches": {
          "mean": 6.697916666666667,
          "training_seed_sd": 0.09547032697824667
        },
        "executed_non_target_switches": {
          "mean": 2.625,
          "training_seed_sd": 0.16535945694153692
        },
        "executed_fragile_departures": {
          "mean": 2.5104166666666665,
          "training_seed_sd": 0.29536347664078805
        },
        "event_task_response_delay": {
          "mean": -7.3125,
          "training_seed_sd": 3.7205877811845807
        },
        "task_completion_ratio": {
          "mean": -0.0010416666666666664,
          "training_seed_sd": 0.00180421959121758
        },
        "priority_weighted_completion": {
          "mean": -0.0013338018442399155,
          "training_seed_sd": 0.001998520590157703
        },
        "remaining_demand": {
          "mean": 2.4583333333333335,
          "training_seed_sd": 4.753425300068713
        },
        "team_reward_after_event": {
          "mean": 0.37748839867205203,
          "training_seed_sd": 15.504722482313662
        }
      }
    },
    {
      "training_condition": "v33aa_learned",
      "model_role": "main_method",
      "ablation": "none",
      "contrast": "priority_high_minus_low",
      "training_seeds": 3,
      "eval_scenarios_per_seed": 32,
      "paired_state_check": true,
      "metrics": {
        "mean_event_task_probability": {
          "mean": 0.04185526311166029,
          "training_seed_sd": 0.018935434844976333
        },
        "executed_target_assignments": {
          "mean": 0.4895833333333333,
          "training_seed_sd": 0.06505206248331664
        },
        "executed_switches": {
          "mean": 0.1875,
          "training_seed_sd": 0.03125
        },
        "executed_non_target_switches": {
          "mean": -0.3020833333333333,
          "training_seed_sd": 0.09547032697824666
        },
        "executed_fragile_departures": {
          "mean": 0.052083333333333336,
          "training_seed_sd": 0.036084391824351615
        },
        "event_task_response_delay": {
          "mean": -25.927083333333332,
          "training_seed_sd": 11.816481361675875
        },
        "task_completion_ratio": {
          "mean": -0.002083333333333333,
          "training_seed_sd": 0.00180421959121758
        },
        "priority_weighted_completion": {
          "mean": -0.0008939695503464631,
          "training_seed_sd": 0.0009218149215776153
        },
        "remaining_demand": {
          "mean": -0.40625,
          "training_seed_sd": 4.653942381196828
        },
        "team_reward_after_event": {
          "mean": 18.34946580944403,
          "training_seed_sd": 6.331544714847401
        }
      }
    },
    {
      "training_condition": "v33aa_learned",
      "model_role": "main_method",
      "ablation": "none",
      "contrast": "react_minus_keep_current",
      "training_seeds": 3,
      "eval_scenarios_per_seed": 32,
      "paired_state_check": true,
      "metrics": {
        "mean_event_task_probability": {
          "mean": 0.0,
          "training_seed_sd": 0.0
        },
        "executed_target_assignments": {
          "mean": 2.25,
          "training_seed_sd": 0.5685439187081329
        },
        "executed_switches": {
          "mean": 5.197916666666667,
          "training_seed_sd": 0.39072915278148024
        },
        "executed_non_target_switches": {
          "mean": 2.9479166666666665,
          "training_seed_sd": 0.5388548815157318
        },
        "executed_fragile_departures": {
          "mean": 1.6666666666666667,
          "training_seed_sd": 0.6422109434082647
        },
        "event_task_response_delay": {
          "mean": -8.135416666666666,
          "training_seed_sd": 8.235387943705708
        },
        "task_completion_ratio": {
          "mean": 0.0,
          "training_seed_sd": 0.0031249999999999993
        },
        "priority_weighted_completion": {
          "mean": 0.0007363279973373582,
          "training_seed_sd": 0.0032393345814927133
        },
        "remaining_demand": {
          "mean": 1.2916666666666667,
          "training_seed_sd": 2.738850504013195
        },
        "team_reward_after_event": {
          "mean": -1.9130222253296847,
          "training_seed_sd": 13.734194459704604
        }
      }
    },
    {
      "training_condition": "v33aa_uniform",
      "model_role": "training_control",
      "ablation": "uniform_AA_gate_trained",
      "contrast": "priority_high_minus_low",
      "training_seeds": 3,
      "eval_scenarios_per_seed": 32,
      "paired_state_check": true,
      "metrics": {
        "mean_event_task_probability": {
          "mean": 0.026003780436100483,
          "training_seed_sd": 0.019844328857352267
        },
        "executed_target_assignments": {
          "mean": 0.3958333333333333,
          "training_seed_sd": 0.2886751345948129
        },
        "executed_switches": {
          "mean": 0.25,
          "training_seed_sd": 0.21650635094610965
        },
        "executed_non_target_switches": {
          "mean": -0.14583333333333334,
          "training_seed_sd": 0.07216878364870322
        },
        "executed_fragile_departures": {
          "mean": 0.03125,
          "training_seed_sd": 0.03125
        },
        "event_task_response_delay": {
          "mean": -16.229166666666668,
          "training_seed_sd": 4.511306686630086
        },
        "task_completion_ratio": {
          "mean": 0.0,
          "training_seed_sd": 0.00541265877365274
        },
        "priority_weighted_completion": {
          "mean": 0.0010699401777154334,
          "training_seed_sd": 0.005766652487716197
        },
        "remaining_demand": {
          "mean": -3.5208333333333335,
          "training_seed_sd": 6.866566797412906
        },
        "team_reward_after_event": {
          "mean": 18.675202423764738,
          "training_seed_sd": 29.496108222831154
        }
      }
    },
    {
      "training_condition": "v33aa_uniform",
      "model_role": "training_control",
      "ablation": "uniform_AA_gate_trained",
      "contrast": "react_minus_keep_current",
      "training_seeds": 3,
      "eval_scenarios_per_seed": 32,
      "paired_state_check": true,
      "metrics": {
        "mean_event_task_probability": {
          "mean": 0.0,
          "training_seed_sd": 0.0
        },
        "executed_target_assignments": {
          "mean": 2.3020833333333335,
          "training_seed_sd": 0.07864410870073697
        },
        "executed_switches": {
          "mean": 5.458333333333333,
          "training_seed_sd": 0.8281741337625399
        },
        "executed_non_target_switches": {
          "mean": 3.15625,
          "training_seed_sd": 0.8800079900205452
        },
        "executed_fragile_departures": {
          "mean": 1.6666666666666667,
          "training_seed_sd": 0.18042195912175807
        },
        "event_task_response_delay": {
          "mean": -7.666666666666667,
          "training_seed_sd": 5.296052972104162
        },
        "task_completion_ratio": {
          "mean": 0.0,
          "training_seed_sd": 0.0031249999999999993
        },
        "priority_weighted_completion": {
          "mean": -0.00015419577857192038,
          "training_seed_sd": 0.003040010164025634
        },
        "remaining_demand": {
          "mean": -3.4791666666666665,
          "training_seed_sd": 4.389966958398814
        },
        "team_reward_after_event": {
          "mean": 9.523871113688443,
          "training_seed_sd": 24.205765445908828
        }
      }
    }
  ]
}