{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "radius": "constant"
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "radius": "constant"
        },
        "mean": 0.14777550008147955
      },
      {
        "cfg": {
          "lr": 0.003,
          "radius": "constant"
        },
        "mean": 0.02788990270346403
      },
      {
        "cfg": {
          "lr": 0.01,
          "radius": "constant"
        },
        "mean": 0.010466416599228978
      }
    ],
    "full": {
      "mean": 0.009070940839592367,
      "std": 0.0043108428117023265,
      "per_seed": [
        0.01107354648411274,
        0.007685416843742132,
        0.006489071529358625,
        0.016617631539702415,
        0.014132681302726269,
        0.008140470832586288,
        0.002562329638749361,
        0.005866378545761108
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.009070940839592367,
    "std": 0.0043108428117023265,
    "per_seed": [
      0.01107354648411274,
      0.007685416843742132,
      0.006489071529358625,
      0.016617631539702415,
      0.014132681302726269,
      0.008140470832586288,
      0.002562329638749361,
      0.005866378545761108
    ],
    "n": 8,
    "selected_per_seed": [
      {
        "seed": 0,
        "best_lr": 0.01,
        "metric": 0.01107354648411274
      },
      {
        "seed": 1,
        "best_lr": 0.01,
        "metric": 0.007685416843742132
      },
      {
        "seed": 2,
        "best_lr": 0.01,
        "metric": 0.006489071529358625
      },
      {
        "seed": 3,
        "best_lr": 0.01,
        "metric": 0.016617631539702415
      },
      {
        "seed": 4,
        "best_lr": 0.01,
        "metric": 0.014132681302726269
      },
      {
        "seed": 5,
        "best_lr": 0.01,
        "metric": 0.008140470832586288
      },
      {
        "seed": 6,
        "best_lr": 0.01,
        "metric": 0.002562329638749361
      },
      {
        "seed": 7,
        "best_lr": 0.01,
        "metric": 0.005866378545761108
      }
    ]
  },
  "comparison": {
    "delta_mean": 0.0,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0
    ],
    "p_value": 1.0,
    "mde": 0.0,
    "mde_rel_pct": 0.0,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "mechanism_signature": {
    "prediction": "dense tube radius is affine in nearest-observation distance with slope Gamma",
    "radius_slope_predicted_mean": 0.23295989949256185,
    "radius_slope_observed_mean": 0.2329598994925619,
    "radius_slope_abs_error_mean": 8.326672684688674e-17,
    "tube_coverage_mean": 0.9995028409090909,
    "constant_radius_coverage_mean": 0.9989109848484848,
    "median_radius_mean": 0.6557790094055236,
    "confirmed": true,
    "trained_model_measurements": true,
    "note": "The fixed dynamics bench predicts a single endpoint, so dense coverage is a diagnostic interpolation signature; endpoint MSE is the independent primary metric."
  },
  "protocol_notes": {
    "baseline_grid": [
      {
        "lr": 0.001,
        "radius": "constant"
      },
      {
        "lr": 0.003,
        "radius": "constant"
      },
      {
        "lr": 0.01,
        "radius": "constant"
      }
    ],
    "idea_grid": [
      {
        "lr": 0.001,
        "tube": "q+Gamma*delta"
      },
      {
        "lr": 0.003,
        "tube": "q+Gamma*delta"
      },
      {
        "lr": 0.01,
        "tube": "q+Gamma*delta"
      }
    ],
    "paired_seeds": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ],
    "epochs": 6,
    "n_train": 500,
    "n_test": 160,
    "structural_match": "continuous-time trajectory uncertainty and stability -> dynamics"
  }
}