1{
  2  "bench_version": 1,
  3  "track": "robust_cbf_pendulum_policy",
  4  "model": "local_mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "eps": 0.0,
 11      "kc": 1.5,
 12      "epochs": 15
 13    },
 14    "sweep": [
 15      {
 16        "cfg": {
 17          "lr": 0.001,
 18          "eps": 0.0,
 19          "kc": 1.5,
 20          "epochs": 15
 21        },
 22        "mean": 0.021794200874865055
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.003,
 27          "eps": 0.0,
 28          "kc": 1.5,
 29          "epochs": 15
 30        },
 31        "mean": 0.002334954886464402
 32      },
 33      {
 34        "cfg": {
 35          "lr": 0.006,
 36          "eps": 0.0,
 37          "kc": 1.5,
 38          "epochs": 15
 39        },
 40        "mean": 0.0018331437022425234
 41      }
 42    ],
 43    "full": {
 44      "mean": 0.0019978795608039945,
 45      "std": 0.0005495084055684182,
 46      "per_seed": [
 47        0.0015628907131031156,
 48        0.0015461408765986562,
 49        0.0021467802580446005,
 50        0.0020767629612237215,
 51        0.0030877823010087013,
 52        0.0011752876453101635,
 53        0.0020312685519456863,
 54        0.0023561231791973114
 55      ],
 56      "n": 8
 57    }
 58  },
 59  "idea": {
 60    "mean": 0.667444035410881,
 61    "std": 0.05083093845847301,
 62    "per_seed": [
 63      0.7455176115036011,
 64      0.6221279501914978,
 65      0.5776458382606506,
 66      0.6847237944602966,
 67      0.7122350931167603,
 68      0.6813809871673584,
 69      0.6877815127372742,
 70      0.6281394958496094
 71    ],
 72    "n": 8,
 73    "selection_sweep": [
 74      {
 75        "cfg": {
 76          "lr": 0.001,
 77          "eps": 0.05,
 78          "kc": 1.5,
 79          "epochs": 15
 80        },
 81        "mean": 0.6575037986040115
 82      },
 83      {
 84        "cfg": {
 85          "lr": 0.003,
 86          "eps": 0.15,
 87          "kc": 1.5,
 88          "epochs": 15
 89        },
 90        "mean": 0.6907831430435181
 91      },
 92      {
 93        "cfg": {
 94          "lr": 0.006,
 95          "eps": 0.3,
 96          "kc": 1.5,
 97          "epochs": 15
 98        },
 99        "mean": 0.7713967263698578
100      }
101    ]
102  },
103  "comparison": {
104    "delta_mean": 0.665446155850077,
105    "idea_wins": 0,
106    "n_pairs": 8,
107    "per_seed_diffs": [
108      0.743954720790498,
109      0.6205818093148991,
110      0.575499058002606,
111      0.6826470314990729,
112      0.7091473108157516,
113      0.6802056995220482,
114      0.6857502441853285,
115      0.6257833726704121
116    ],
117    "p_value": 0.0081,
118    "mde": 0.04544311646025259,
119    "mde_rel_pct": 2274.567363908823,
120    "verdict": "idea worse (significant)",
121    "system_worked": false
122  },
123  "mechanism_signature": {
124    "prediction": {
125      "uncertainty_boundary_shift_per_epsilon": 1.0,
126      "virtual_shift_formula": "Delta/kc"
127    },
128    "observed_from_trained_models": {
129      "shift_per_epsilon_mean": 0.9999994114041328,
130      "mean_intervention": 0.3867955468595028,
131      "feasible_fraction_mean": 0.9265625,
132      "n_models": 8
133    },
134    "confirmed": true
135  },
136  "custom_track": {
137    "name": "robust_cbf_pendulum_policy",
138    "file": "robust_cbf_pendulum_policy.py",
139    "domain": "dynamics"
140  }
141}