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": 0.00430093941395171
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0033809992601163685
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.002017002771026455
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0022443737179855816,
 33      "std": 0.0006826101486098914,
 34      "per_seed": [
 35        0.0008340506465174258,
 36        0.0019196165958419442,
 37        0.002481805393472314,
 38        0.0028325384482741356,
 39        0.0020983400754630566,
 40        0.0018523626495152712,
 41        0.003032150212675333,
 42        0.0029041257221251726
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.00194204316358082,
 49    "std": 0.0005946512584653817,
 50    "per_seed": [
 51      0.0025444745551794767,
 52      0.0021839358378201723,
 53      0.0014850844163447618,
 54      0.0013838530285283923,
 55      0.002973762573674321,
 56      0.002029680646955967,
 57      0.0018950217636302114,
 58      0.001040532486513257
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -0.00030233055440476164,
 64    "idea_wins": 4,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.001710423908662051,
 68      0.0002643192419782281,
 69      -0.000996720977127552,
 70      -0.0014486854197457433,
 71      0.0008754224982112646,
 72      0.00017731799744069576,
 73      -0.0011371284490451217,
 74      -0.0018635932356119156
 75    ],
 76    "p_value": 0.51515,
 77    "mde": 0.0010443640298036328,
 78    "mde_rel_pct": 46.532536958283075,
 79    "verdict": "no significant win",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "math_check": {
 84      "checked": true,
 85      "L": 0.7,
 86      "delta": 0.767539284296568,
 87      "sample_min": 0.36195731104790524,
 88      "cloud_min": 0.3268021435301365,
 89      "bound": -0.17532018795969234
 90    },
 91    "mechanism_signature": {
 92      "prediction": "adaptive reachable-state selection should reduce empirical coverage radius at equal simulator budget",
 93      "observed": {
 94        "baseline_delta": 0.7605222459149111,
 95        "idea_delta": 0.3445205023070951,
 96        "baseline_pool_mse": 1.55548095703125,
 97        "idea_pool_mse": 1.5370749235153198,
 98        "baseline_min_sample_residual": -0.5907788276672363,
 99        "idea_min_sample_residual": -0.6713848114013672
100      },
101      "relative_delta_reduction_pct": 54.699483919417034,
102      "confirmed": true
103    },
104    "idea_grid_results": [
105      [
106        0.012647395371459424,
107        {
108          "lr": 0.001
109        }
110      ],
111      [
112        0.003680584253743291,
113        {
114          "lr": 0.003
115        }
116      ],
117      [
118        0.0018993369594682008,
119        {
120          "lr": 0.006
121        }
122      ]
123    ],
124    "selected_idea_cfg": {
125      "lr": 0.006
126    },
127    "budget": {
128      "epochs": 12,
129      "train_samples": 400,
130      "pool_samples": 1600
131    }
132  }
133}