{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.003,
      "weight_decay": 0.0001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "weight_decay": 0.0
        },
        "mean": 0.01389938301872462
      },
      {
        "cfg": {
          "lr": 0.0015,
          "weight_decay": 0.0001
        },
        "mean": 0.01393047539750114
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 0.006244247197173536
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001
        },
        "mean": 0.00622874335385859
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0
        },
        "mean": 0.007728174212388694
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0001
        },
        "mean": 0.007723002112470567
      }
    ],
    "full": {
      "mean": 0.006740233220625669,
      "std": 0.003932537515930856,
      "per_seed": [
        0.005818837322294712,
        0.00311187538318336,
        0.012360582128167152,
        0.003623678581789136,
        0.0028440626338124275,
        0.004673089366406202,
        0.007840032689273357,
        0.013649707660079002
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.006728245556587353,
    "std": 0.003915532713508233,
    "per_seed": [
      0.005818843841552734,
      0.003111880971118808,
      0.012264670804142952,
      0.003623704193159938,
      0.0028439818415790796,
      0.004673039074987173,
      0.00784000102430582,
      0.013649842701852322
    ],
    "n": 8,
    "best_cfg": {
      "lr": 0.003,
      "weight_decay": 0.0001,
      "reg": 0.1
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001,
          "reg": 0.01
        },
        "mean": 0.006728254316840321,
        "std": 0.003915514027606794,
        "per_seed": [
          0.005818838719278574,
          0.003111876081675291,
          0.012264743447303772,
          0.0036236809100955725,
          0.0028440479654818773,
          0.004673085641115904,
          0.007840028963983059,
          0.013649732805788517
        ],
        "n": 8
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001,
          "reg": 0.03
        },
        "mean": 0.006728248088620603,
        "std": 0.003915514311451397,
        "per_seed": [
          0.005818837787955999,
          0.003111874917522073,
          0.012264708988368511,
          0.0036236881278455257,
          0.002844026545062661,
          0.0046730744652450085,
          0.007840024307370186,
          0.01364974956959486
        ],
        "n": 8
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001,
          "reg": 0.1
        },
        "mean": 0.006728245556587353,
        "std": 0.003915532713508233,
        "per_seed": [
          0.005818843841552734,
          0.003111880971118808,
          0.012264670804142952,
          0.003623704193159938,
          0.0028439818415790796,
          0.004673039074987173,
          0.00784000102430582,
          0.013649842701852322
        ],
        "n": 8
      }
    ]
  },
  "comparison": {
    "delta_mean": -1.1987664038315415e-05,
    "idea_wins": 4,
    "n_pairs": 8,
    "per_seed_diffs": [
      6.51925802230835e-09,
      5.587935447692871e-09,
      -9.591132402420044e-05,
      2.561137080192566e-08,
      -8.079223334789276e-08,
      -5.029141902923584e-08,
      -3.166496753692627e-08,
      1.3504177331924438e-07
    ],
    "p_value": 0.5437,
    "mde": 2.8354259077042443e-05,
    "mde_rel_pct": 0.42067178017336365,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "mechanism_signature": {
    "baseline_normalized_observability": -1.568709631527554e-06,
    "idea_normalized_observability": -1.7481720213652352e-06,
    "baseline_normalized_reachability": -1.48555855794541e-06,
    "idea_normalized_reachability": -1.1099894585697712e-06,
    "predicted": "regularization increases both normalized minimum Gramian eigenvalues",
    "confirmed": false
  },
  "protocol_notes": {
    "epochs": 8,
    "paired_seeds": 8,
    "structural_match": "control/dynamics latent-state track",
    "baseline_grid": [
      {
        "lr": 0.0015,
        "weight_decay": 0.0
      },
      {
        "lr": 0.0015,
        "weight_decay": 0.0001
      },
      {
        "lr": 0.003,
        "weight_decay": 0.0
      },
      {
        "lr": 0.003,
        "weight_decay": 0.0001
      },
      {
        "lr": 0.006,
        "weight_decay": 0.0
      },
      {
        "lr": 0.006,
        "weight_decay": 0.0001
      }
    ],
    "idea_grid": [
      {
        "lr": 0.003,
        "weight_decay": 0.0001,
        "reg": 0.01
      },
      {
        "lr": 0.003,
        "weight_decay": 0.0001,
        "reg": 0.03
      },
      {
        "lr": 0.003,
        "weight_decay": 0.0001,
        "reg": 0.1
      }
    ]
  }
}