{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.003,
      "epochs": 12
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "epochs": 12
        },
        "mean": 0.03892233967781067
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 12
        },
        "mean": 0.030090928077697754
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 12
        },
        "mean": 0.05556618981063366
      }
    ],
    "full": {
      "mean": 0.030090928077697754,
      "std": 0.004071433001350898,
      "per_seed": [
        0.03666028752923012,
        0.02726038359105587,
        0.030306316912174225,
        0.03209355100989342,
        0.024804387241601944,
        0.033685918897390366,
        0.024255627766251564,
        0.03166095167398453
      ],
      "n": 8
    },
    "searched_grid": [
      {
        "lr": 0.0015,
        "epochs": 12
      },
      {
        "lr": 0.003,
        "epochs": 12
      },
      {
        "lr": 0.006,
        "epochs": 12
      }
    ]
  },
  "idea": {
    "mean": 0.03774771373718977,
    "std": 0.004092688612850228,
    "per_seed": [
      0.044951874762773514,
      0.037032078951597214,
      0.0381438285112381,
      0.04309968277812004,
      0.03429986909031868,
      0.037081118673086166,
      0.03175382688641548,
      0.035619430243968964
    ],
    "n": 8,
    "best_config": {
      "lr": 0.003,
      "epochs": 12
    }
  },
  "comparison": {
    "delta_mean": 0.007656785659492016,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.008291587233543396,
      0.009771695360541344,
      0.007837511599063873,
      0.011006131768226624,
      0.009495481848716736,
      0.0033951997756958008,
      0.0074981991201639175,
      0.003958478569984436
    ],
    "p_value": 0.0081,
    "mde": 0.0022651487884256674,
    "mde_rel_pct": 7.527680045550038,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "predicted": {
      "unsafe_fraction": "elevated near mixed signed threshold",
      "order_effect": "nonzero"
    },
    "observed": {
      "high_drive_fraction": 0.25,
      "mean_output_abs_difference": 0.15396859403699636
    },
    "confirmed": true
  },
  "idea_sweep": {
    "{'lr': 0.0015, 'epochs': 12}": {
      "mean": 0.05525698186829686,
      "std": 0.02605631177795089,
      "per_seed": [
        0.06213795393705368,
        0.06313511729240417,
        0.04352016746997833,
        0.11799561977386475,
        0.03681809827685356,
        0.03047773241996765,
        0.03951030224561691,
        0.048460863530635834
      ],
      "n": 8
    },
    "{'lr': 0.003, 'epochs': 12}": {
      "mean": 0.03774771373718977,
      "std": 0.004092688612850228,
      "per_seed": [
        0.044951874762773514,
        0.037032078951597214,
        0.0381438285112381,
        0.04309968277812004,
        0.03429986909031868,
        0.037081118673086166,
        0.03175382688641548,
        0.035619430243968964
      ],
      "n": 8,
      "best_config": {
        "lr": 0.003,
        "epochs": 12
      }
    },
    "{'lr': 0.006, 'epochs': 12}": {
      "mean": 0.041616966016590595,
      "std": 0.0063707081731607254,
      "per_seed": [
        0.044290054589509964,
        0.04693755879998207,
        0.03771070018410683,
        0.0332266241312027,
        0.036028288304805756,
        0.04784112051129341,
        0.035485830157995224,
        0.05141555145382881
      ],
      "n": 8
    }
  },
  "protocol": {
    "seeds": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ],
    "n_train": 400,
    "n_test": 400,
    "matched_architecture": true,
    "shared_hyperparameter_union": [
      {
        "lr": 0.0015,
        "epochs": 12
      },
      {
        "lr": 0.003,
        "epochs": 12
      },
      {
        "lr": 0.006,
        "epochs": 12
      }
    ]
  }
}