{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "epochs": 12
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 12
        },
        "mean": 0.004103701037820429
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 12
        },
        "mean": 0.0033861721749417484
      },
      {
        "cfg": {
          "lr": 0.01,
          "epochs": 12
        },
        "mean": 0.0014241317112464458
      }
    ],
    "full": {
      "mean": 0.001394080012687482,
      "std": 0.0003941947646392458,
      "per_seed": [
        0.0008339877240359783,
        0.0018332540057599545,
        0.001035665743984282,
        0.0019936193712055683,
        0.0009847625624388456,
        0.0013270708732306957,
        0.0016533465823158622,
        0.0014909332385286689
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.0013935517708887346,
    "std": 0.0003435936282481101,
    "per_seed": [
      0.000962425721809268,
      0.0018949415534734726,
      0.0011497323866933584,
      0.0018263080855831504,
      0.0009521312895230949,
      0.0014383804518729448,
      0.0016268157633021474,
      0.0012976789148524404
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": -5.282417987473309e-07,
    "idea_wins": 4,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.00012843799777328968,
      6.168754771351814e-05,
      0.0001140666427090764,
      -0.00016731128562241793,
      -3.263127291575074e-05,
      0.00011130957864224911,
      -2.653081901371479e-05,
      -0.00019325432367622852
    ],
    "p_value": 0.98515,
    "mde": 0.0001061475231632145,
    "mde_rel_pct": 7.614162902930173,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "mechanism_signature": {
    "prediction": "accepted Petri successors remain admissible",
    "closure_violations": 0,
    "predicted_unsafe_accepted_rate": 0.0,
    "observed_unsafe_prediction_rate_trained_idea": 0.07,
    "observed_rejection_rate_trained_task_inputs": 0.05347656179219484,
    "trained_model_rows": [
      {
        "seed": 0,
        "baseline_metric": 0.0008339877240359783,
        "idea_metric": 0.000962425721809268,
        "baseline_unsafe_prediction_rate": 0.045,
        "idea_unsafe_prediction_rate": 0.0425,
        "shield_rejection_rate": 0.03531249985098839
      },
      {
        "seed": 1,
        "baseline_metric": 0.0018332540057599545,
        "idea_metric": 0.0018949415534734726,
        "baseline_unsafe_prediction_rate": 0.095,
        "idea_unsafe_prediction_rate": 0.0975,
        "shield_rejection_rate": 0.059687498956918716
      },
      {
        "seed": 2,
        "baseline_metric": 0.001035665743984282,
        "idea_metric": 0.0011497323866933584,
        "baseline_unsafe_prediction_rate": 0.055,
        "idea_unsafe_prediction_rate": 0.055,
        "shield_rejection_rate": 0.051874998956918716
      },
      {
        "seed": 3,
        "baseline_metric": 0.0019936193712055683,
        "idea_metric": 0.0018263080855831504,
        "baseline_unsafe_prediction_rate": 0.1225,
        "idea_unsafe_prediction_rate": 0.1225,
        "shield_rejection_rate": 0.0793749988079071
      },
      {
        "seed": 4,
        "baseline_metric": 0.0009847625624388456,
        "idea_metric": 0.0009521312895230949,
        "baseline_unsafe_prediction_rate": 0.0725,
        "idea_unsafe_prediction_rate": 0.07,
        "shield_rejection_rate": 0.054999999701976776
      },
      {
        "seed": 5,
        "baseline_metric": 0.0013270708732306957,
        "idea_metric": 0.0014383804518729448,
        "baseline_unsafe_prediction_rate": 0.0325,
        "idea_unsafe_prediction_rate": 0.0325,
        "shield_rejection_rate": 0.03312499821186066
      },
      {
        "seed": 6,
        "baseline_metric": 0.0016533465823158622,
        "idea_metric": 0.0016268157633021474,
        "baseline_unsafe_prediction_rate": 0.055,
        "idea_unsafe_prediction_rate": 0.0575,
        "shield_rejection_rate": 0.05843750014901161
      },
      {
        "seed": 7,
        "baseline_metric": 0.0014909332385286689,
        "idea_metric": 0.0012976789148524404,
        "baseline_unsafe_prediction_rate": 0.0825,
        "idea_unsafe_prediction_rate": 0.0825,
        "shield_rejection_rate": 0.054999999701976776
      }
    ],
    "confirmed": true
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001,
        "epochs": 12
      },
      "result": {
        "mean": 0.004844218114158139,
        "std": 0.0021060335026899866,
        "per_seed": [
          0.0076133571565151215,
          0.0035649114288389683,
          0.001954183680936694,
          0.002968552988022566,
          0.00587603310123086,
          0.004371007438749075,
          0.00404327642172575,
          0.008362422697246075
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.003,
        "epochs": 12
      },
      "result": {
        "mean": 0.002841877590981312,
        "std": 0.0016170159627390552,
        "per_seed": [
          0.0020566112361848354,
          0.0030648154206573963,
          0.0009700310183688998,
          0.006774549372494221,
          0.0028293621726334095,
          0.002948844339698553,
          0.002073001815006137,
          0.002017805352807045
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01,
        "epochs": 12
      },
      "result": {
        "mean": 0.0013935517708887346,
        "std": 0.0003435936282481101,
        "per_seed": [
          0.000962425721809268,
          0.0018949415534734726,
          0.0011497323866933584,
          0.0018263080855831504,
          0.0009521312895230949,
          0.0014383804518729448,
          0.0016268157633021474,
          0.0012976789148524404
        ],
        "n": 8
      }
    }
  ],
  "selected_idea_cfg": {
    "lr": 0.01,
    "epochs": 12
  },
  "track_justification": "dynamics is the matched control/stability benchmark: actuated pendulum rollouts."
}