1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.0019458123424556106
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.000998065312160179
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.0007849872272345237
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0007333773064601701,
 33      "std": 0.00023119481143299165,
 34      "per_seed": [
 35        0.0007082645315676928,
 36        0.0011445400305092335,
 37        0.00047928086132742465,
 38        0.0008078634855337441,
 39        0.00035395112354308367,
 40        0.0006696216878481209,
 41        0.0007750760996714234,
 42        0.0009284206316806376
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.0007333773064601701,
 49    "std": 0.00023119481143299165,
 50    "per_seed": [
 51      0.0007082645315676928,
 52      0.0011445400305092335,
 53      0.00047928086132742465,
 54      0.0008078634855337441,
 55      0.00035395112354308367,
 56      0.0006696216878481209,
 57      0.0007750760996714234,
 58      0.0009284206316806376
 59    ],
 60    "n": 8,
 61    "best_cfg": {
 62      "lr": 0.006,
 63      "lambda": 0.01
 64    }
 65  },
 66  "comparison": {
 67    "delta_mean": 0.0,
 68    "idea_wins": 0,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      0.0,
 72      0.0,
 73      0.0,
 74      0.0,
 75      0.0,
 76      0.0,
 77      0.0,
 78      0.0
 79    ],
 80    "p_value": 1.0,
 81    "mde": 0.0,
 82    "mde_rel_pct": 0.0,
 83    "verdict": "no measurable effect",
 84    "system_worked": false
 85  },
 86  "idea_sweep": [
 87    {
 88      "cfg": {
 89        "lr": 0.001,
 90        "lambda": 0.01
 91      },
 92      "mean": 0.0019458123424556106
 93    },
 94    {
 95      "cfg": {
 96        "lr": 0.001,
 97        "lambda": 0.03
 98      },
 99      "mean": 0.0019458123424556106
100    },
101    {
102      "cfg": {
103        "lr": 0.001,
104        "lambda": 0.1
105      },
106      "mean": 0.0019458123424556106
107    },
108    {
109      "cfg": {
110        "lr": 0.003,
111        "lambda": 0.01
112      },
113      "mean": 0.000998065312160179
114    },
115    {
116      "cfg": {
117        "lr": 0.003,
118        "lambda": 0.03
119      },
120      "mean": 0.000998065312160179
121    },
122    {
123      "cfg": {
124        "lr": 0.003,
125        "lambda": 0.1
126      },
127      "mean": 0.000998065312160179
128    },
129    {
130      "cfg": {
131        "lr": 0.006,
132        "lambda": 0.01
133      },
134      "mean": 0.0007849872272345237
135    },
136    {
137      "cfg": {
138        "lr": 0.006,
139        "lambda": 0.03
140      },
141      "mean": 0.0007849872272345237
142    },
143    {
144      "cfg": {
145        "lr": 0.006,
146        "lambda": 0.1
147      },
148      "mean": 0.0007849872272345237
149    }
150  ],
151  "protocol_note": "Baseline and idea use identical rnn_small predictor, data, epochs, batch, Adam, and paired seeds; intervention is the Lyapunov loss.",
152  "mechanism_signature": {
153    "predicted_boundary": 0.0,
154    "observed_mean_residual_eigenvalue": -0.4996017813682556,
155    "observed_violation_fraction": 0.0,
156    "observed_max_residual_eigenvalue": -0.38156095147132874,
157    "alpha": 0.05,
158    "h": 1.0,
159    "confirmed": true,
160    "signature_model_seed_metric": 0.0007082645315676928
161  }
162}