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      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.0015,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.08054963732138276
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 0.023347282083705068
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.008135722251608968
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.006979767291340977,
 37      "std": 0.0037932412616514086,
 38      "per_seed": [
 39        0.0038006319664418697,
 40        0.0048443530686199665,
 41        0.007569973357021809,
 42        0.016327930614352226,
 43        0.004687489941716194,
 44        0.004746445454657078,
 45        0.008039474487304688,
 46        0.005821839440613985
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "per_seed": [
 53      0.003616119036450982,
 54      0.0048443530686199665,
 55      0.007523232605308294,
 56      0.015394117683172226,
 57      0.0045921714045107365,
 58      0.004602333996444941,
 59      0.00775922043249011,
 60      0.005821839440613985
 61    ],
 62    "mean": 0.006769173458451405
 63  },
 64  "comparison": {
 65    "delta_mean": -0.00021059383288957179,
 66    "idea_wins": 6,
 67    "n_pairs": 8,
 68    "per_seed_diffs": [
 69      -0.00018451292999088764,
 70      0.0,
 71      -4.674075171351433e-05,
 72      -0.0009338129311800003,
 73      -9.531853720545769e-05,
 74      -0.00014411145821213722,
 75      -0.0002802540548145771,
 76      0.0
 77    ],
 78    "p_value": 0.0311,
 79    "mde": 0.00025710573309478267,
 80    "mde_rel_pct": 3.683586033215538,
 81    "verdict": "idea better (significant)",
 82    "system_worked": true
 83  },
 84  "mechanism_signature": {
 85    "window": 8,
 86    "predicted_margin_mean": 0.8796176314353943,
 87    "observed_input_sensitivity_mean": 0.0076210699044167995,
 88    "predicted_vs_observed_correlation": 0.19263989726460348,
 89    "confirmed": false
 90  },
 91  "idea_sweep": [
 92    {
 93      "cfg": {
 94        "lr": 0.0015,
 95        "weight_decay": 0.0
 96      },
 97      "mean": 0.06897093774750829
 98    },
 99    {
100      "cfg": {
101        "lr": 0.003,
102        "weight_decay": 0.0
103      },
104      "mean": 0.02178938896395266
105    },
106    {
107      "cfg": {
108        "lr": 0.006,
109        "weight_decay": 0.0
110      },
111      "mean": 0.007844455598387867
112    }
113  ],
114  "protocol_notes": "8 paired seeds; baseline sweep seeds 0-3 plus full best-config evaluation; idea uses same three learning rates and epochs."
115}