{
  "schema_version": 2,
  "manifest_ref": "experiments:agent-toolroute-l2.1-seeds42-44",
  "benchmark": "toolroute-l2.1-v1",
  "seeds": [
    42,
    43,
    44
  ],
  "episodes_per_seed": 60,
  "oracle_fields_exposed": false,
  "guide_endpoint": "absent",
  "results": {
    "long-context": {
      "metrics": {
        "answer_accuracy": 0.35,
        "plan_exact_match": 0.35,
        "plan_step_f1": 0.5212301587301588,
        "joint_success": 0.35,
        "recovery_success_rate": 0,
        "invalid_tool_rate": 0.053211908639440475,
        "irreversible_error_rate": 0.07222222222222222,
        "average_tool_calls": 3.1444444444444444,
        "average_cost": 2.857111111111111
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.35,
          "std": 0.016666666666666663,
          "ci95_radius": 0.01886010879352777
        },
        "plan_exact_match": {
          "mean": 0.35,
          "std": 0.016666666666666663,
          "ci95_radius": 0.01886010879352777
        },
        "plan_step_f1": {
          "mean": 0.5212301587301588,
          "std": 0.012826425681661901,
          "ci95_radius": 0.014514467027294526
        },
        "joint_success": {
          "mean": 0.35,
          "std": 0.016666666666666663,
          "ci95_radius": 0.01886010879352777
        },
        "recovery_success_rate": {
          "mean": 0,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.053211908639440475,
          "std": 0.01516947708034028,
          "ci95_radius": 0.017165879284568628
        },
        "irreversible_error_rate": {
          "mean": 0.07222222222222222,
          "std": 0.00962250448649376,
          "ci95_radius": 0.010888888888888887
        },
        "average_tool_calls": {
          "mean": 3.1444444444444444,
          "std": 0.07876359377087677,
          "ci95_radius": 0.08912939684927776
        },
        "average_cost": {
          "mean": 2.857111111111111,
          "std": 0.0644785350444742,
          "ci95_radius": 0.07296433114716461
        }
      },
      "artifact_path": "docs/agent-research/metrics/toolroute-l2-long-context-seeds42-44.json"
    },
    "react": {
      "metrics": {
        "answer_accuracy": 0.6833333333333333,
        "plan_exact_match": 0.6833333333333333,
        "plan_step_f1": 0.7683076491409825,
        "joint_success": 0.6833333333333333,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.04554420919793445,
        "irreversible_error_rate": 0.07222222222222222,
        "average_tool_calls": 5.366666666666667,
        "average_cost": 4.798444444444445
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.6833333333333333,
          "std": 0.016666666666666663,
          "ci95_radius": 0.01886010879352777
        },
        "plan_exact_match": {
          "mean": 0.6833333333333333,
          "std": 0.016666666666666663,
          "ci95_radius": 0.01886010879352777
        },
        "plan_step_f1": {
          "mean": 0.7683076491409825,
          "std": 0.00644999586686913,
          "ci95_radius": 0.007298857426017377
        },
        "joint_success": {
          "mean": 0.6833333333333333,
          "std": 0.016666666666666663,
          "ci95_radius": 0.01886010879352777
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.04554420919793445,
          "std": 0.0034731968812121063,
          "ci95_radius": 0.0039302922624601016
        },
        "irreversible_error_rate": {
          "mean": 0.07222222222222222,
          "std": 0.00962250448649376,
          "ci95_radius": 0.010888888888888887
        },
        "average_tool_calls": {
          "mean": 5.366666666666667,
          "std": 0.04409585518441019,
          "ci95_radius": 0.04989915756729732
        },
        "average_cost": {
          "mean": 4.798444444444445,
          "std": 0.029622313911669484,
          "ci95_radius": 0.03352080378540707
        }
      },
      "artifact_path": "docs/agent-research/2210.03629-react/metrics/toolroute-l2-seeds42-44.json"
    },
    "reflexion": {
      "metrics": {
        "answer_accuracy": 0.7277777777777777,
        "plan_exact_match": 0.7277777777777777,
        "plan_step_f1": 0.7785640593973927,
        "joint_success": 0.7277777777777777,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.03757984173139372,
        "irreversible_error_rate": 0.07222222222222222,
        "average_tool_calls": 5.322222222222222,
        "average_cost": 4.792000000000001
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.7277777777777777,
          "std": 0.009622504486493728,
          "ci95_radius": 0.010888888888888849
        },
        "plan_exact_match": {
          "mean": 0.7277777777777777,
          "std": 0.009622504486493728,
          "ci95_radius": 0.010888888888888849
        },
        "plan_step_f1": {
          "mean": 0.7785640593973927,
          "std": 0.006752434179036209,
          "ci95_radius": 0.007641098594265499
        },
        "joint_success": {
          "mean": 0.7277777777777777,
          "std": 0.009622504486493728,
          "ci95_radius": 0.010888888888888849
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.03757984173139372,
          "std": 0.0002957135766726038,
          "ci95_radius": 0.0003346314136661115
        },
        "irreversible_error_rate": {
          "mean": 0.07222222222222222,
          "std": 0.00962250448649376,
          "ci95_radius": 0.010888888888888887
        },
        "average_tool_calls": {
          "mean": 5.322222222222222,
          "std": 0.04194352464039291,
          "ci95_radius": 0.0474635662741094
        },
        "average_cost": {
          "mean": 4.792000000000001,
          "std": 0.029310597249307582,
          "ci95_radius": 0.03316806317551302
        }
      },
      "artifact_path": "docs/agent-research/2303.11366-reflexion/metrics/toolroute-l2-seeds42-44.json"
    },
    "agent-g2": {
      "metrics": {
        "answer_accuracy": 0.7777777777777778,
        "plan_exact_match": 0.7777777777777778,
        "plan_step_f1": 0.8276025209358543,
        "joint_success": 0.7777777777777778,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.03944502980868784,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 5.633333333333334,
        "average_cost": 5.066222222222223
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.7777777777777778,
          "std": 0.025458753860865796,
          "ci95_radius": 0.028809292053814453
        },
        "plan_exact_match": {
          "mean": 0.7777777777777778,
          "std": 0.025458753860865796,
          "ci95_radius": 0.028809292053814453
        },
        "plan_step_f1": {
          "mean": 0.8276025209358543,
          "std": 0.006782471749829366,
          "ci95_radius": 0.007675089305448633
        },
        "joint_success": {
          "mean": 0.7777777777777778,
          "std": 0.025458753860865796,
          "ci95_radius": 0.028809292053814453
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.03944502980868784,
          "std": 0.004486914687208469,
          "ci95_radius": 0.005077421948881763
        },
        "irreversible_error_rate": {
          "mean": 0.0,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "average_tool_calls": {
          "mean": 5.633333333333334,
          "std": 0.01666666666666705,
          "ci95_radius": 0.018860108793528212
        },
        "average_cost": {
          "mean": 5.066222222222223,
          "std": 0.012098362301169978,
          "ci95_radius": 0.01369058575341485
        }
      },
      "artifact_path": "docs/agent-research/2608.23318-agent-g2/metrics/toolroute-l2-seeds42-44.json"
    },
    "ahead": {
      "metrics": {
        "answer_accuracy": 0.7555555555555555,
        "plan_exact_match": 0.7555555555555555,
        "plan_step_f1": 0.7886946386946387,
        "joint_success": 0.7555555555555555,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.033423200650924474,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 5.011111111111111,
        "average_cost": 4.557111111111111
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "plan_exact_match": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "plan_step_f1": {
          "mean": 0.7886946386946387,
          "std": 0.014839329090731549,
          "ci95_radius": 0.016792281664449515
        },
        "joint_success": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.033423200650924474,
          "std": 0.009765182300528781,
          "ci95_radius": 0.011050344034596279
        },
        "irreversible_error_rate": {
          "mean": 0.0,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "average_tool_calls": {
          "mean": 5.011111111111111,
          "std": 0.12619796324000615,
          "ci95_radius": 0.1428064389736881
        },
        "average_cost": {
          "mean": 4.557111111111111,
          "std": 0.10424401572663461,
          "ci95_radius": 0.11796320866071296
        }
      },
      "artifact_path": "docs/agent-research/2608.24114-ahead/metrics/toolroute-l2-seeds42-44.json"
    },
    "auso": {
      "metrics": {
        "answer_accuracy": 0.7555555555555555,
        "plan_exact_match": 0.7555555555555555,
        "plan_step_f1": 0.7886946386946387,
        "joint_success": 0.7555555555555555,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.033423200650924474,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 5.011111111111111,
        "average_cost": 4.519222222222223
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "plan_exact_match": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "plan_step_f1": {
          "mean": 0.7886946386946387,
          "std": 0.014839329090731549,
          "ci95_radius": 0.016792281664449515
        },
        "joint_success": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.033423200650924474,
          "std": 0.009765182300528781,
          "ci95_radius": 0.011050344034596279
        },
        "irreversible_error_rate": {
          "mean": 0.0,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "average_tool_calls": {
          "mean": 5.011111111111111,
          "std": 0.12619796324000615,
          "ci95_radius": 0.1428064389736881
        },
        "average_cost": {
          "mean": 4.519222222222223,
          "std": 0.10003184678076471,
          "ci95_radius": 0.11319669078616341
        }
      },
      "artifact_path": "docs/agent-research/2608.21292-auso/metrics/toolroute-l2-seeds42-44.json"
    },
    "jit-agent": {
      "metrics": {
        "answer_accuracy": 0.7555555555555555,
        "plan_exact_match": 0.7555555555555555,
        "plan_step_f1": 0.7886946386946387,
        "joint_success": 0.7555555555555555,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.033423200650924474,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 5.011111111111111,
        "average_cost": 4.4912222222222224
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "plan_exact_match": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "plan_step_f1": {
          "mean": 0.7886946386946387,
          "std": 0.014839329090731549,
          "ci95_radius": 0.016792281664449515
        },
        "joint_success": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.033423200650924474,
          "std": 0.009765182300528781,
          "ci95_radius": 0.011050344034596279
        },
        "irreversible_error_rate": {
          "mean": 0.0,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "average_tool_calls": {
          "mean": 5.011111111111111,
          "std": 0.12619796324000615,
          "ci95_radius": 0.1428064389736881
        },
        "average_cost": {
          "mean": 4.4912222222222224,
          "std": 0.10118044021193778,
          "ci95_radius": 0.11449644661045075
        }
      },
      "artifact_path": "docs/agent-research/2608.25593-jit-agent/metrics/toolroute-l2-seeds42-44.json"
    },
    "traceml": {
      "metrics": {
        "answer_accuracy": 0.7277777777777777,
        "plan_exact_match": 0.7277777777777777,
        "plan_step_f1": 0.7785640593973927,
        "joint_success": 0.7277777777777777,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.03757984173139372,
        "irreversible_error_rate": 0.07222222222222222,
        "average_tool_calls": 5.322222222222222,
        "average_cost": 4.792000000000001
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.7277777777777777,
          "std": 0.009622504486493728,
          "ci95_radius": 0.010888888888888849
        },
        "plan_exact_match": {
          "mean": 0.7277777777777777,
          "std": 0.009622504486493728,
          "ci95_radius": 0.010888888888888849
        },
        "plan_step_f1": {
          "mean": 0.7785640593973927,
          "std": 0.006752434179036209,
          "ci95_radius": 0.007641098594265499
        },
        "joint_success": {
          "mean": 0.7277777777777777,
          "std": 0.009622504486493728,
          "ci95_radius": 0.010888888888888849
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.03757984173139372,
          "std": 0.0002957135766726038,
          "ci95_radius": 0.0003346314136661115
        },
        "irreversible_error_rate": {
          "mean": 0.07222222222222222,
          "std": 0.00962250448649376,
          "ci95_radius": 0.010888888888888887
        },
        "average_tool_calls": {
          "mean": 5.322222222222222,
          "std": 0.04194352464039291,
          "ci95_radius": 0.0474635662741094
        },
        "average_cost": {
          "mean": 4.792000000000001,
          "std": 0.029310597249307582,
          "ci95_radius": 0.03316806317551302
        }
      },
      "artifact_path": "docs/agent-research/2608.26086-traceml/metrics/toolroute-l2-seeds42-44.json"
    },
    "adavdr": {
      "metrics": {
        "answer_accuracy": 0.8166666666666667,
        "plan_exact_match": 0.8166666666666667,
        "plan_step_f1": 0.8365768799102132,
        "joint_success": 0.8166666666666667,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.03277105884317278,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 5.594444444444444,
        "average_cost": 5.057666666666667
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.8166666666666667,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "plan_exact_match": {
          "mean": 0.8166666666666667,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "plan_step_f1": {
          "mean": 0.8365768799102132,
          "std": 0.0009395476107850423,
          "ci95_radius": 0.0010631982093662993
        },
        "joint_success": {
          "mean": 0.8166666666666667,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.03277105884317278,
          "std": 0.00014934489395767876,
          "ci95_radius": 0.00016899965686798142
        },
        "irreversible_error_rate": {
          "mean": 0.0,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "average_tool_calls": {
          "mean": 5.594444444444444,
          "std": 0.02545875386086569,
          "ci95_radius": 0.02880929205381433
        },
        "average_cost": {
          "mean": 5.057666666666667,
          "std": 0.02554299208089073,
          "ci95_radius": 0.028904616573469055
        }
      },
      "artifact_path": "docs/agent-research/2608.25559-adavdr/metrics/toolroute-l2-seeds42-44.json"
    },
    "topas": {
      "metrics": {
        "answer_accuracy": 0.7555555555555555,
        "plan_exact_match": 0.7555555555555555,
        "plan_step_f1": 0.7886946386946387,
        "joint_success": 0.7555555555555555,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.033423200650924474,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 5.011111111111111,
        "average_cost": 4.557111111111111
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "plan_exact_match": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "plan_step_f1": {
          "mean": 0.7886946386946387,
          "std": 0.014839329090731549,
          "ci95_radius": 0.016792281664449515
        },
        "joint_success": {
          "mean": 0.7555555555555555,
          "std": 0.019245008972987584,
          "ci95_radius": 0.021777777777777844
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.033423200650924474,
          "std": 0.009765182300528781,
          "ci95_radius": 0.011050344034596279
        },
        "irreversible_error_rate": {
          "mean": 0.0,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "average_tool_calls": {
          "mean": 5.011111111111111,
          "std": 0.12619796324000615,
          "ci95_radius": 0.1428064389736881
        },
        "average_cost": {
          "mean": 4.557111111111111,
          "std": 0.10424401572663461,
          "ci95_radius": 0.11796320866071296
        }
      },
      "artifact_path": "docs/agent-research/2608.25523-topas/metrics/toolroute-l2-seeds42-44.json"
    },
    "caskg": {
      "metrics": {
        "answer_accuracy": 0.8,
        "plan_exact_match": 0.8,
        "plan_step_f1": 0.8327307260640594,
        "joint_success": 0.8,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.035644054235870325,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 5.611111111111111,
        "average_cost": 5.011555555555556
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.8,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "plan_exact_match": {
          "mean": 0.8,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "plan_step_f1": {
          "mean": 0.8327307260640594,
          "std": 0.0009395476107850075,
          "ci95_radius": 0.00106319820936626
        },
        "joint_success": {
          "mean": 0.8,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.035644054235870325,
          "std": 0.00016195458446435212,
          "ci95_radius": 0.00018326886495649588
        },
        "irreversible_error_rate": {
          "mean": 0.0,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "average_tool_calls": {
          "mean": 5.611111111111111,
          "std": 0.025458753860866074,
          "ci95_radius": 0.028809292053814765
        },
        "average_cost": {
          "mean": 5.011555555555556,
          "std": 0.021922422142468423,
          "ci95_radius": 0.024807555997475802
        }
      },
      "artifact_path": "docs/agent-research/2608.25500-caskg/metrics/toolroute-l2-seeds42-44.json"
    },
    "progrouter": {
      "metrics": {
        "answer_accuracy": 0.7777777777777778,
        "plan_exact_match": 0.7777777777777778,
        "plan_step_f1": 0.8036637744971079,
        "joint_success": 0.7777777777777778,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.016382715153875006,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 5.161111111111111,
        "average_cost": 4.632444444444444
      },
      "aggregate_metrics": {
        "answer_accuracy": {
          "mean": 0.7777777777777778,
          "std": 0.025458753860865796,
          "ci95_radius": 0.028809292053814453
        },
        "plan_exact_match": {
          "mean": 0.7777777777777778,
          "std": 0.025458753860865796,
          "ci95_radius": 0.028809292053814453
        },
        "plan_step_f1": {
          "mean": 0.8036637744971079,
          "std": 0.019300096976820252,
          "ci95_radius": 0.02184011572250799
        },
        "joint_success": {
          "mean": 0.7777777777777778,
          "std": 0.025458753860865796,
          "ci95_radius": 0.028809292053814453
        },
        "recovery_success_rate": {
          "mean": 0.6666666666666666,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "invalid_tool_rate": {
          "mean": 0.016382715153875006,
          "std": 0.01187984526243933,
          "ci95_radius": 0.013443310445992878
        },
        "irreversible_error_rate": {
          "mean": 0.0,
          "std": 0.0,
          "ci95_radius": 0.0
        },
        "average_tool_calls": {
          "mean": 5.161111111111111,
          "std": 0.15485955405295945,
          "ci95_radius": 0.17524008222936063
        },
        "average_cost": {
          "mean": 4.632444444444444,
          "std": 0.12171110299636827,
          "ci95_radius": 0.1377290786335062
        }
      },
      "artifact_path": "docs/agent-research/2608.25992-progrouter/metrics/toolroute-l2-seeds42-44.json"
    }
  },
  "ablations": {
    "react-no-retry": {
      "metrics": {
        "answer_accuracy": 0.5166666666666667,
        "plan_exact_match": 0.5166666666666667,
        "plan_step_f1": 0.6608220791554125,
        "joint_success": 0.5166666666666667,
        "recovery_success_rate": 0.3333333333333333,
        "invalid_tool_rate": 0.052805653041113364,
        "irreversible_error_rate": 0.07222222222222222,
        "average_tool_calls": 4.627777777777778,
        "average_cost": 4.147333333333333
      },
      "validation_metrics": {
        "answer_accuracy": 0.5111111111111111,
        "plan_exact_match": 0.5111111111111111,
        "plan_step_f1": 0.6578122494789161,
        "joint_success": 0.5111111111111111,
        "recovery_success_rate": 0.3333333333333333,
        "invalid_tool_rate": 0.05915516460463897,
        "irreversible_error_rate": 0.08888888888888889,
        "average_tool_calls": 3.9444444444444446,
        "average_cost": 3.612444444444445
      }
    },
    "reflexion-no-reflection": {
      "metrics": {
        "answer_accuracy": 0.6833333333333333,
        "plan_exact_match": 0.6833333333333333,
        "plan_step_f1": 0.7683076491409825,
        "joint_success": 0.6833333333333333,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.04554420919793445,
        "irreversible_error_rate": 0.07222222222222222,
        "average_tool_calls": 5.366666666666667,
        "average_cost": 4.798444444444445
      },
      "validation_metrics": {
        "answer_accuracy": 0.6777777777777778,
        "plan_exact_match": 0.6777777777777778,
        "plan_step_f1": 0.7571869488536155,
        "joint_success": 0.6777777777777778,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.05136523384698567,
        "irreversible_error_rate": 0.08888888888888889,
        "average_tool_calls": 4.544444444444444,
        "average_cost": 4.172888888888889
      }
    },
    "agent-g2-no-verifier": {
      "metrics": {
        "answer_accuracy": 0.6833333333333333,
        "plan_exact_match": 0.6833333333333333,
        "plan_step_f1": 0.7683076491409825,
        "joint_success": 0.6833333333333333,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.04554420919793445,
        "irreversible_error_rate": 0.07222222222222222,
        "average_tool_calls": 5.366666666666667,
        "average_cost": 4.798444444444445
      },
      "validation_metrics": {
        "answer_accuracy": 0.6777777777777778,
        "plan_exact_match": 0.6777777777777778,
        "plan_step_f1": 0.7571869488536155,
        "joint_success": 0.6777777777777778,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.05136523384698567,
        "irreversible_error_rate": 0.08888888888888889,
        "average_tool_calls": 4.544444444444444,
        "average_cost": 4.172888888888889
      }
    },
    "ahead-no-compression": {
      "metrics": {
        "answer_accuracy": 0.7555555555555555,
        "plan_exact_match": 0.7555555555555555,
        "plan_step_f1": 0.8224743158076492,
        "joint_success": 0.7555555555555555,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.04322401526982474,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 5.655555555555556,
        "average_cost": 5.089333333333333
      },
      "validation_metrics": {
        "answer_accuracy": 0.7666666666666667,
        "plan_exact_match": 0.7666666666666667,
        "plan_step_f1": 0.820679012345679,
        "joint_success": 0.7666666666666667,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.04850297962322184,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 4.811111111111111,
        "average_cost": 4.453777777777778
      }
    },
    "auso-no-memory": {
      "metrics": {
        "answer_accuracy": 0.7555555555555555,
        "plan_exact_match": 0.7555555555555555,
        "plan_step_f1": 0.7886946386946387,
        "joint_success": 0.7555555555555555,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.033423200650924474,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 5.011111111111111,
        "average_cost": 4.557111111111111
      },
      "validation_metrics": {
        "answer_accuracy": 0.7666666666666667,
        "plan_exact_match": 0.7666666666666667,
        "plan_step_f1": 0.7996632996632996,
        "joint_success": 0.7666666666666667,
        "recovery_success_rate": 0.6666666666666666,
        "invalid_tool_rate": 0.036436151817406885,
        "irreversible_error_rate": 0.0,
        "average_tool_calls": 4.288888888888889,
        "average_cost": 4.014444444444445
      }
    }
  },
  "evolve_artifact": "docs/agent-research/metrics/toolroute-l21-evolve-seeds42-44.json",
  "evaluation_protocol": {
    "tier": "l2_capability",
    "seeds": [
      42,
      43,
      44
    ],
    "formal_comparison": true,
    "claim_policy": "held-out no-guide test; compare only within toolroute-l2.1-v1"
  },
  "provenance": {
    "artifact_path": "docs/experiments/agent-toolroute-l2-seeds42-44.json",
    "dataset_fingerprint": "25d906c4d33d7476f423f672765ab420349dbc390492744612d987d6c614ec60",
    "original_code_commit": "generated from repository source"
  },
  "evolve_champion": {
    "metrics": {
      "answer_accuracy": 0.8000000000000002,
      "answer_accuracy_std": 1.3597399555105182e-16,
      "plan_exact_match": 0.8000000000000002,
      "plan_exact_match_std": 1.3597399555105182e-16,
      "plan_step_f1": 0.8327307260640594,
      "plan_step_f1_std": 0.0009395476107850075,
      "joint_success": 0.8000000000000002,
      "joint_success_std": 1.3597399555105182e-16,
      "recovery_success_rate": 0.6666666666666666,
      "recovery_success_rate_std": 0.0,
      "invalid_tool_rate": 0.035644054235870325,
      "invalid_tool_rate_std": 0.00016195458446435212,
      "irreversible_error_rate": 0.0,
      "irreversible_error_rate_std": 0.0,
      "average_tool_calls": 5.611111111111111,
      "average_tool_calls_std": 0.025458753860866077,
      "average_cost": 4.966222222222222,
      "average_cost_std": 0.021922422142468423,
      "reuse_rate": 4.566666666666666,
      "reuse_rate_std": 0.028867513459481187,
      "recovery_rate": 0.6666666666666666,
      "recovery_rate_std": 0.0,
      "verification_rate": 0.0,
      "verification_rate_std": 0.0,
      "compression_rate": 0.0,
      "compression_rate_std": 0.0,
      "primary": 0.8137175560717079
    },
    "champion_id": "g2-t2",
    "champion_genome": {
      "architecture": "composable-agent",
      "dimensions": 64,
      "layers": 2,
      "learning_rate": 0.0003,
      "optimizer": "adamw",
      "batch_size": 48,
      "experts": 4,
      "interval_residual": 2,
      "auxiliary_weight": 0.15,
      "heads": 4,
      "kv_heads": 4,
      "sequence_length": 128,
      "expansion": 4,
      "data_recipe": "wikitext",
      "data_mix_ratio": 0.2,
      "post_training": "none",
      "neftune_alpha": 0.0,
      "post_steps": 0,
      "group_size": 4,
      "post_data_recipe": "base",
      "post_teacher": "auto",
      "post_rollout": "on-policy",
      "gradient_accumulation": 1,
      "mixed_precision": "no",
      "agent_memory": "evoharness-rl",
      "agent_planner": "pearl",
      "agent_tool_policy": "direct",
      "agent_critic": "none",
      "agent_policy": "heuristic",
      "agent_failure_recovery": "retry",
      "agent_reflection": "none",
      "agent_verifier": "none",
      "agent_context_compression": "full",
      "memory_size": 48,
      "genrec_context": "recent",
      "genrec_head": "id-catalog",
      "genrec_reward": "uniform",
      "genrec_distillation": "none",
      "multimodal_objective": "cross_entropy",
      "checkpoint_prompt_style": "direct",
      "checkpoint_use_hint": true,
      "checkpoint_image_size": 0,
      "checkpoint_max_new_tokens": 16,
      "reasoning_samples": 1,
      "reasoning_max_new_tokens": 96,
      "reasoning_stop_consensus": 1.0,
      "reasoning_verifier": "self-consistency"
    }
  }
}
