[
  {
    "run": "20260830-2132_cfm-marl-mrt2a-delivcoal-adapter-screen",
    "selector": "last",
    "source": "runs/20260830-2132_cfm-marl-mrt2a-delivcoal-adapter-screen/evaluation/final_test/last/cfm_s10/50/summary.json",
    "data": {
      "method": "CFM",
      "environment": "DelivCoal",
      "protocol": {
        "eval_seed": 50000,
        "eval_episodes": 16,
        "paired_seeds": true,
        "greedy": true
      },
      "checkpoints": {
        "50": {
          "success": {
            "mean": 0.0,
            "std": 0.0,
            "ci95": 0.0
          },
          "task_completion_ratio": {
            "mean": 0.03125,
            "std": 0.046351240544347895,
            "ci95": 0.023456964140030283
          },
          "discovered_ratio": {
            "mean": 0.35,
            "std": 0.203100960115899,
            "ci95": 0.10278326712067486
          },
          "sent_percentage": {
            "mean": 0.0274121905428596,
            "std": 0.04238688863855649,
            "ci95": 0.021450725269171213
          },
          "episode_reward": {
            "mean": -233.36031390432117,
            "std": 183.51340168866028,
            "ci95": 92.87059487668431
          },
          "episode_length": {
            "mean": 750.0,
            "std": 0.0,
            "ci95": 0.0
          },
          "coalition_changes": {
            "mean": 329.9375,
            "std": 237.4335351077223,
            "ci95": 120.15794730097271
          },
          "reconfiguration_decisions": {
            "mean": 4136.375,
            "std": 3288.967977857948,
            "ci95": 1664.4474453819
          },
          "seed": {
            "mean": 50007.5,
            "std": 4.6097722286464435,
            "ci95": 2.3328666199906642
          },
          "checkpoint_episode": {
            "mean": 50.0,
            "std": 0.0,
            "ci95": 0.0
          }
        }
      }
    }
  },
  {
    "run": "20260830-2132_cfm-marl-mrt2a-delivcoal-adapter-screen",
    "selector": "last",
    "source": "runs/20260830-2132_cfm-marl-mrt2a-delivcoal-adapter-screen/evaluation/final_test/last/marl_mrt2a_s10/150000/summary.json",
    "data": {
      "method": "MARL-MRT2A",
      "environment": "DelivCoal",
      "protocol": {
        "eval_seed": 50000,
        "eval_episodes": 16,
        "batch_size_run": 8,
        "paired_base_seed": true
      },
      "checkpoints": {
        "150000": {
          "test_return_mean": -679.0538569756703,
          "test_return_std": 61.11521241533723,
          "test_coalition_changes_mean": 437.625,
          "test_episode_limit_mean": 1.0,
          "test_sent_percentage_mean": 0.19111989504709673,
          "test_target_change_mean": 0.04965601420328451,
          "test_task_completion_ratio_mean": 0.037500000000000006,
          "test_target_revision_mean": 0.04524522858411008,
          "test_success_mean": 0.0,
          "test_ep_length_mean": 750.0,
          "loaded_step": 150000,
          "evaluated_episodes": 16
        }
      }
    }
  },
  {
    "run": "20260830-2132_cfm-marl-mrt2a-delivcoal-adapter-screen",
    "selector": "validation-best",
    "source": "runs/20260830-2132_cfm-marl-mrt2a-delivcoal-adapter-screen/evaluation/final_test/validation-best/cfm_s10/39/summary.json",
    "data": {
      "method": "CFM",
      "environment": "DelivCoal",
      "protocol": {
        "eval_seed": 50000,
        "eval_episodes": 16,
        "paired_seeds": true,
        "greedy": true
      },
      "checkpoints": {
        "39": {
          "success": {
            "mean": 0.0,
            "std": 0.0,
            "ci95": 0.0
          },
          "task_completion_ratio": {
            "mean": 0.025,
            "std": 0.05590169943749475,
            "ci95": 0.028290163190291664
          },
          "discovered_ratio": {
            "mean": 0.33125000000000004,
            "std": 0.17577951388031543,
            "ci95": 0.08895670763541855
          },
          "sent_percentage": {
            "mean": 0.02346828088219275,
            "std": 0.03624505025439637,
            "ci95": 0.018342526199649798
          },
          "episode_reward": {
            "mean": -56.25576876008287,
            "std": 6.6159039242724464,
            "ci95": 3.3481093339252226
          },
          "episode_length": {
            "mean": 750.0,
            "std": 0.0,
            "ci95": 0.0
          },
          "coalition_changes": {
            "mean": 582.875,
            "std": 53.0305041933414,
            "ci95": 26.837137918687727
          },
          "reconfiguration_decisions": {
            "mean": 674.4375,
            "std": 99.0271230206654,
            "ci95": 50.114638708831045
          },
          "seed": {
            "mean": 50007.5,
            "std": 4.6097722286464435,
            "ci95": 2.3328666199906642
          },
          "checkpoint_episode": {
            "mean": 39.0,
            "std": 0.0,
            "ci95": 0.0
          }
        }
      }
    }
  },
  {
    "run": "20260830-2132_cfm-marl-mrt2a-delivcoal-adapter-screen",
    "selector": "validation-best",
    "source": "runs/20260830-2132_cfm-marl-mrt2a-delivcoal-adapter-screen/evaluation/final_test/validation-best/marl_mrt2a_s10/39000/summary.json",
    "data": {
      "method": "MARL-MRT2A",
      "environment": "DelivCoal",
      "protocol": {
        "eval_seed": 50000,
        "eval_episodes": 16,
        "batch_size_run": 8,
        "paired_base_seed": true
      },
      "checkpoints": {
        "39000": {
          "test_return_mean": -637.9795021748234,
          "test_return_std": 89.91651420590784,
          "test_target_revision_mean": 0.03750554815801154,
          "test_target_change_mean": 0.040903794940079896,
          "test_episode_limit_mean": 1.0,
          "test_success_mean": 0.0,
          "test_sent_percentage_mean": 0.2887909284475581,
          "test_coalition_changes_mean": 362.25,
          "test_task_completion_ratio_mean": 0.13749999999999998,
          "test_ep_length_mean": 750.0,
          "loaded_step": 39000,
          "evaluated_episodes": 16
        }
      }
    }
  },
  {
    "run": "20260831-1333_cfm-long-s10-sampled",
    "selector": "last",
    "source": "runs/20260831-1333_cfm-long-s10-sampled/evaluation/final_test/last/cfm_s10/375/summary.json",
    "data": {
      "method": "CFM",
      "environment": "DelivCoal",
      "protocol": {
        "eval_seed": 50000,
        "eval_episodes": 128,
        "paired_seeds": true,
        "greedy": false,
        "policy_seed_offset": 1000000
      },
      "checkpoints": {
        "375": {
          "success": {
            "mean": 0.0078125,
            "std": 0.08804240366863003,
            "ci95": 0.015312499999999998
          },
          "task_completion_ratio": {
            "mean": 0.534375,
            "std": 0.16697562509240682,
            "ci95": 0.029040713936556065
          },
          "discovered_ratio": {
            "mean": 0.6421874999999999,
            "std": 0.17923856963206888,
            "ci95": 0.031173508254284138
          },
          "sent_percentage": {
            "mean": 0.585735858320225,
            "std": 0.15587825513968862,
            "ci95": 0.027110638537427185
          },
          "episode_reward": {
            "mean": -21.976773745272975,
            "std": 12.47443153353308,
            "ci95": 2.1695765324190575
          },
          "episode_length": {
            "mean": 748.875,
            "std": 12.678106128282726,
            "ci95": 2.2049999999999996
          },
          "coalition_changes": {
            "mean": 363.515625,
            "std": 46.177474279776014,
            "ci95": 8.031272948548665
          },
          "reconfiguration_decisions": {
            "mean": 410.953125,
            "std": 54.00099237731076,
            "ci95": 9.391953891783581
          },
          "seed": {
            "mean": 50063.5,
            "std": 36.94928957368463,
            "ci95": 6.42628975381596
          },
          "checkpoint_episode": {
            "mean": 375.0,
            "std": 0.0,
            "ci95": 0.0
          }
        }
      }
    }
  },
  {
    "run": "20260831-1333_cfm-long-s10-sampled",
    "selector": "validation-best",
    "source": "runs/20260831-1333_cfm-long-s10-sampled/evaluation/final_test/validation-best/cfm_s10/375/summary.json",
    "data": {
      "method": "CFM",
      "environment": "DelivCoal",
      "protocol": {
        "eval_seed": 50000,
        "eval_episodes": 128,
        "paired_seeds": true,
        "greedy": false,
        "policy_seed_offset": 1000000
      },
      "checkpoints": {
        "375": {
          "success": {
            "mean": 0.0078125,
            "std": 0.08804240366863003,
            "ci95": 0.015312499999999998
          },
          "task_completion_ratio": {
            "mean": 0.534375,
            "std": 0.16697562509240682,
            "ci95": 0.029040713936556065
          },
          "discovered_ratio": {
            "mean": 0.6421874999999999,
            "std": 0.17923856963206888,
            "ci95": 0.031173508254284138
          },
          "sent_percentage": {
            "mean": 0.585735858320225,
            "std": 0.15587825513968862,
            "ci95": 0.027110638537427185
          },
          "episode_reward": {
            "mean": -21.976773745272975,
            "std": 12.47443153353308,
            "ci95": 2.1695765324190575
          },
          "episode_length": {
            "mean": 748.875,
            "std": 12.678106128282726,
            "ci95": 2.2049999999999996
          },
          "coalition_changes": {
            "mean": 363.515625,
            "std": 46.177474279776014,
            "ci95": 8.031272948548665
          },
          "reconfiguration_decisions": {
            "mean": 410.953125,
            "std": 54.00099237731076,
            "ci95": 9.391953891783581
          },
          "seed": {
            "mean": 50063.5,
            "std": 36.94928957368463,
            "ci95": 6.42628975381596
          },
          "checkpoint_episode": {
            "mean": 375.0,
            "std": 0.0,
            "ci95": 0.0
          }
        }
      }
    }
  },
  {
    "run": "20260831-1333_mrt2a-cap-lr-s10",
    "selector": "last",
    "source": "runs/20260831-1333_mrt2a-cap-lr-s10/evaluation/final_test/last/marl_mrt2a_s10/150000/summary.json",
    "data": {
      "method": "MARL-MRT2A",
      "environment": "DelivCoal",
      "protocol": {
        "eval_seed": 50000,
        "eval_episodes": 128,
        "batch_size_run": 8,
        "paired_base_seed": true
      },
      "checkpoints": {
        "150000": {
          "test_return_mean": -652.6897697448583,
          "test_return_std": 73.56459732594973,
          "test_task_completion_ratio_mean": 0.0875,
          "test_target_change_mean": 0.044537838437638706,
          "test_sent_percentage_mean": 0.23317799615255555,
          "test_coalition_changes_mean": 391.5,
          "test_target_revision_mean": 0.04082057256990679,
          "test_episode_limit_mean": 1.0,
          "test_success_mean": 0.0,
          "test_ep_length_mean": 750.0,
          "loaded_step": 150000,
          "evaluated_episodes": 128
        }
      }
    }
  },
  {
    "run": "20260831-1333_mrt2a-cap-lr-s10",
    "selector": "validation-best",
    "source": "runs/20260831-1333_mrt2a-cap-lr-s10/evaluation/final_test/validation-best/marl_mrt2a_s10/39000/summary.json",
    "data": {
      "method": "MARL-MRT2A",
      "environment": "DelivCoal",
      "protocol": {
        "eval_seed": 50000,
        "eval_episodes": 128,
        "batch_size_run": 8,
        "paired_base_seed": true
      },
      "checkpoints": {
        "39000": {
          "test_return_mean": -532.286486472206,
          "test_return_std": 83.23239092860989,
          "test_success_mean": 0.0,
          "test_target_change_mean": 0.04256824234354195,
          "test_coalition_changes_mean": 375.75,
          "test_episode_limit_mean": 1.0,
          "test_sent_percentage_mean": 0.44684911251922244,
          "test_target_revision_mean": 0.03968320017754105,
          "test_task_completion_ratio_mean": 0.2125,
          "test_ep_length_mean": 750.0,
          "loaded_step": 39000,
          "evaluated_episodes": 128
        }
      }
    }
  },
  {
    "run": "20260831-1333_mrt2a-cap-entropy-s10",
    "selector": "last",
    "source": "runs/20260831-1333_mrt2a-cap-entropy-s10/evaluation/final_test/last/marl_mrt2a_s10/150000/summary.json",
    "data": {
      "method": "MARL-MRT2A",
      "environment": "DelivCoal",
      "protocol": {
        "eval_seed": 50000,
        "eval_episodes": 128,
        "batch_size_run": 8,
        "paired_base_seed": true
      },
      "checkpoints": {
        "150000": {
          "test_return_mean": -698.310959663275,
          "test_return_std": 58.947891005763545,
          "test_coalition_changes_mean": 396.625,
          "test_target_revision_mean": 0.04166666666666667,
          "test_target_change_mean": 0.04487072791833111,
          "test_success_mean": 0.0,
          "test_episode_limit_mean": 1.0,
          "test_task_completion_ratio_mean": 0.125,
          "test_sent_percentage_mean": 0.2632063496132372,
          "test_ep_length_mean": 750.0,
          "loaded_step": 150000,
          "evaluated_episodes": 128
        }
      }
    }
  },
  {
    "run": "20260831-1333_mrt2a-cap-entropy-s10",
    "selector": "validation-best",
    "source": "runs/20260831-1333_mrt2a-cap-entropy-s10/evaluation/final_test/validation-best/marl_mrt2a_s10/39000/summary.json",
    "data": {
      "method": "MARL-MRT2A",
      "environment": "DelivCoal",
      "protocol": {
        "eval_seed": 50000,
        "eval_episodes": 128,
        "batch_size_run": 8,
        "paired_base_seed": true
      },
      "checkpoints": {
        "39000": {
          "test_return_mean": -535.4128171423029,
          "test_return_std": 126.43777825146992,
          "test_task_completion_ratio_mean": 0.2125,
          "test_success_mean": 0.0,
          "test_target_change_mean": 0.042401797603195734,
          "test_sent_percentage_mean": 0.45694256096051644,
          "test_episode_limit_mean": 1.0,
          "test_coalition_changes_mean": 375.25,
          "test_target_revision_mean": 0.039378051486906346,
          "test_ep_length_mean": 750.0,
          "loaded_step": 39000,
          "evaluated_episodes": 128
        }
      }
    }
  }
]