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": 0.01
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.001921573159052059
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0011179699067724869
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.0006299250526353717
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0005546553857129766,
 33      "std": 0.00021396768993230646,
 34      "per_seed": [
 35        0.0004971042508259416,
 36        0.0004999673110432923,
 37        0.00046853424282744527,
 38        0.0010540944058448076,
 39        0.0005468997405841947,
 40        0.00023900401720311493,
 41        0.0006055250996723771,
 42        0.0005261140177026391
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.03986519586760551,
 49    "std": 0.034170930267310405,
 50    "per_seed": [
 51      0.028408890590071678,
 52      0.1224212571978569,
 53      0.01613774709403515,
 54      0.053532518446445465,
 55      0.046550970524549484,
 56      0.02247580699622631,
 57      0.018617291003465652,
 58      0.010777085088193417
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.03931054048189253,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.027911786339245737,
 68      0.12192128988681361,
 69      0.015669212851207703,
 70      0.05247842404060066,
 71      0.04600407078396529,
 72      0.022236802979023196,
 73      0.018011765903793275,
 74      0.010250971070490777
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.030512142347341273,
 78    "mde_rel_pct": 5501.099084816372,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "nn_scale_stability": {
 84      "prediction": "known stable relaxation/readout reduces local rollout amplification",
 85      "baseline": {
 86        "mean_jacobian_norm": 0.3944494118914008,
 87        "max_jacobian_norm": 0.4280492067337036,
 88        "perturbation_proxy": 0.06204786151647568
 89      },
 90      "idea": {
 91        "mean_jacobian_norm": 0.9194773212075233,
 92        "max_jacobian_norm": 13.631381034851074,
 93        "perturbation_proxy": 0.8245298266410828
 94      },
 95      "confirmed": false
 96    },
 97    "idea_cfg": {
 98      "lr": 0.01
 99    }
100  },
101  "protocol_note": "8 paired seeds; 400/160 samples; 18 epochs; baseline and idea evaluated on identical lr union."
102}