{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "clip": 1.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "clip": 0.5
        },
        "mean": 0.25846289470791817
      },
      {
        "cfg": {
          "lr": 0.001,
          "clip": 1.0
        },
        "mean": 0.2586120665073395
      },
      {
        "cfg": {
          "lr": 0.001,
          "clip": 5.0
        },
        "mean": 0.264630239456892
      },
      {
        "cfg": {
          "lr": 0.003,
          "clip": 0.5
        },
        "mean": 0.010474085924215615
      },
      {
        "cfg": {
          "lr": 0.003,
          "clip": 1.0
        },
        "mean": 0.009937021299265325
      },
      {
        "cfg": {
          "lr": 0.003,
          "clip": 5.0
        },
        "mean": 0.012804438127204776
      },
      {
        "cfg": {
          "lr": 0.01,
          "clip": 0.5
        },
        "mean": 0.011152858496643603
      },
      {
        "cfg": {
          "lr": 0.01,
          "clip": 1.0
        },
        "mean": 0.00783492624759674
      },
      {
        "cfg": {
          "lr": 0.01,
          "clip": 5.0
        },
        "mean": 0.008992245304398239
      }
    ],
    "full": {
      "mean": 0.009677478694356978,
      "std": 0.0029150067189096305,
      "per_seed": [
        0.010251475498080254,
        0.005092725623399019,
        0.009792184457182884,
        0.006203319411724806,
        0.008501410484313965,
        0.010244788601994514,
        0.014377173036336899,
        0.012956752441823483
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.009327424864750355,
    "std": 0.004120870426856388,
    "per_seed": [
      0.01298802811652422,
      0.0055983662605285645,
      0.01386063452810049,
      0.00352195231243968,
      0.00725489342585206,
      0.0069480095990002155,
      0.00851170253008604,
      0.015935812145471573
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": -0.00035005382960662246,
    "idea_wins": 4,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.002736552618443966,
      0.0005056406371295452,
      0.004068450070917606,
      -0.0026813670992851257,
      -0.0012465170584619045,
      -0.003296779002994299,
      -0.005865470506250858,
      0.0029790597036480904
    ],
    "p_value": 0.78655,
    "mde": 0.0029343297181013495,
    "mde_rel_pct": 30.321221164892698,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "prediction": "smallest Fourier recurrent margin predicts dominant hidden-channel Fourier sector",
    "predicted_vs_observed": [
      {
        "seed": 0,
        "metric": 0.01298802811652422,
        "predicted_sector": 5,
        "observed_sector": 29,
        "predicted_margin": 0.9399786254849777
      },
      {
        "seed": 1,
        "metric": 0.0055983662605285645,
        "predicted_sector": 56,
        "observed_sector": 20,
        "predicted_margin": 0.9352336144834422
      },
      {
        "seed": 2,
        "metric": 0.01386063452810049,
        "predicted_sector": 62,
        "observed_sector": 14,
        "predicted_margin": 0.9405557241088474
      },
      {
        "seed": 3,
        "metric": 0.00352195231243968,
        "predicted_sector": 22,
        "observed_sector": 26,
        "predicted_margin": 0.9286565991151029
      },
      {
        "seed": 4,
        "metric": 0.00725489342585206,
        "predicted_sector": 16,
        "observed_sector": 2,
        "predicted_margin": 0.9472155426115653
      },
      {
        "seed": 5,
        "metric": 0.0069480095990002155,
        "predicted_sector": 59,
        "observed_sector": 18,
        "predicted_margin": 0.9560543716567116
      },
      {
        "seed": 6,
        "metric": 0.00851170253008604,
        "predicted_sector": 37,
        "observed_sector": 30,
        "predicted_margin": 0.9501085228512387
      },
      {
        "seed": 7,
        "metric": 0.015935812145471573,
        "predicted_sector": 8,
        "observed_sector": 0,
        "predicted_margin": 0.9243731275838336
      }
    ],
    "agreement_fraction": 0.0,
    "confirmed": false
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001,
        "margin0": 0.2
      },
      "full": {
        "mean": 0.2551296819001436,
        "std": 0.06966527729488384,
        "per_seed": [
          0.1645231992006302,
          0.286499947309494,
          0.23894044756889343,
          0.3685573637485504,
          0.13928723335266113,
          0.2740314304828644,
          0.2623410224914551,
          0.30685681104660034
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.003,
        "margin0": 0.2
      },
      "full": {
        "mean": 0.013181458110921085,
        "std": 0.006411194717789758,
        "per_seed": [
          0.01743449829518795,
          0.00642412668094039,
          0.02289336547255516,
          0.004465762060135603,
          0.010089418850839138,
          0.00904592964798212,
          0.013677984476089478,
          0.02142057940363884
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01,
        "margin0": 0.2
      },
      "full": {
        "mean": 0.009327424864750355,
        "std": 0.004120870426856388,
        "per_seed": [
          0.01298802811652422,
          0.0055983662605285645,
          0.01386063452810049,
          0.00352195231243968,
          0.00725489342585206,
          0.0069480095990002155,
          0.00851170253008604,
          0.015935812145471573
        ],
        "n": 8
      }
    }
  ],
  "track_justification": "Dynamics is structurally matched: the task is an actuated pendulum rollout and the intervention monitors recurrent stability sectors.",
  "protocol": {
    "paired_seeds": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ],
    "baseline_sweep_seeds": [
      0,
      1,
      2,
      3
    ],
    "epochs": 5,
    "batch": 128,
    "baseline_grid": [
      {
        "lr": 0.001,
        "clip": 0.5
      },
      {
        "lr": 0.001,
        "clip": 1.0
      },
      {
        "lr": 0.001,
        "clip": 5.0
      },
      {
        "lr": 0.003,
        "clip": 0.5
      },
      {
        "lr": 0.003,
        "clip": 1.0
      },
      {
        "lr": 0.003,
        "clip": 5.0
      },
      {
        "lr": 0.01,
        "clip": 0.5
      },
      {
        "lr": 0.01,
        "clip": 1.0
      },
      {
        "lr": 0.01,
        "clip": 5.0
      }
    ],
    "idea_grid": [
      {
        "lr": 0.001,
        "margin0": 0.2
      },
      {
        "lr": 0.003,
        "margin0": 0.2
      },
      {
        "lr": 0.01,
        "margin0": 0.2
      }
    ]
  }
}