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": 0.0014015047054272145
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.000840683322167024
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.0006119587633293122
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0005385368749557529,
 33      "std": 0.00013641301474292053,
 34      "per_seed": [
 35        0.0004973900504410267,
 36        0.0007723030867055058,
 37        0.0004330792580731213,
 38        0.000745062658097595,
 39        0.0003953906998503953,
 40        0.0005369232967495918,
 41        0.00040419393917545676,
 42        0.0005239520105533302
 43      ],
 44      "n": 8
 45    },
 46    "idea_union_sweep": [
 47      {
 48        "cfg": {
 49          "lr": 0.001
 50        },
 51        "mean": 0.0004685235799115617
 52      },
 53      {
 54        "cfg": {
 55          "lr": 0.003
 56        },
 57        "mean": 0.14334793618763797
 58      },
 59      {
 60        "cfg": {
 61          "lr": 0.006
 62        },
 63        "mean": 0.0005863967526238412
 64      }
 65    ]
 66  },
 67  "idea": {
 68    "mean": 0.00040997614269144833,
 69    "std": 0.000259649820853824,
 70    "per_seed": [
 71      0.000214326151763089,
 72      0.0007017290918156505,
 73      0.0006978304008953273,
 74      0.00026020867517217994,
 75      0.00015163708303589374,
 76      0.00015855272067710757,
 77      0.0002806181728374213,
 78      0.0008149068453349173
 79    ],
 80    "n": 8
 81  },
 82  "comparison": {
 83    "delta_mean": -0.00012856073226430453,
 84    "idea_wins": 6,
 85    "n_pairs": 8,
 86    "per_seed_diffs": [
 87      -0.0002830638986779377,
 88      -7.057399488985538e-05,
 89      0.000264751142822206,
 90      -0.00048485398292541504,
 91      -0.00024375361681450158,
 92      -0.00037837057607248425,
 93      -0.00012357576633803546,
 94      0.0002909548347815871
 95    ],
 96    "p_value": 0.2357,
 97    "mde": 0.00023661934353507038,
 98    "mde_rel_pct": 43.93744505508772,
 99    "verdict": "no significant win",
100    "system_worked": false
101  },
102  "idea_best_cfg": {
103    "lr": 0.001
104  },
105  "epochs": 10,
106  "n_train": 1000,
107  "n_test": 250,
108  "mechanism_signature": {
109    "prediction": "positive-real resolvent is nonexpansive: max ||R v||/||v|| <= 1",
110    "observed_max_ratio_mean": 3.9459029138088226,
111    "observed_max_ratio_max": 5.293459892272949,
112    "failed_linear_solves_total": 0,
113    "confirmed": false
114  }
115}