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": 0.021498297923244536
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0067647010437212884
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.00675402011256665
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.00675402011256665,
 33      "std": 0.0028601230170336904,
 34      "per_seed": [
 35        0.006943476852029562,
 36        0.005382216069847345,
 37        0.012206831015646458,
 38        0.006380172912031412,
 39        0.0068011293187737465,
 40        0.009766676463186741,
 41        0.0037104773800820112,
 42        0.0028411808889359236
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.007599645119626075,
 49    "std": 0.0019136134219505131,
 50    "per_seed": [
 51      0.009002679027616978,
 52      0.010305776260793209,
 53      0.008358007296919823,
 54      0.008188792504370213,
 55      0.00483678886666894,
 56      0.009082159027457237,
 57      0.006025956943631172,
 58      0.004997001029551029
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.0008456250070594251,
 64    "idea_wins": 3,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.0020592021755874157,
 68      0.004923560190945864,
 69      -0.003848823718726635,
 70      0.0018086195923388004,
 71      -0.001964340452104807,
 72      -0.0006845174357295036,
 73      0.002315479563549161,
 74      0.0021558201406151056
 75    ],
 76    "p_value": 0.4058,
 77    "mde": 0.002347660486662723,
 78    "mde_rel_pct": 34.75945359260367,
 79    "verdict": "no significant win",
 80    "system_worked": false
 81  },
 82  "protocol": {
 83    "epochs": 8,
 84    "n_train": 400,
 85    "n_test": 400,
 86    "grid": [
 87      {
 88        "lr": 0.001
 89      },
 90      {
 91        "lr": 0.003
 92      },
 93      {
 94        "lr": 0.006
 95      }
 96    ],
 97    "seeds": [
 98      0,
 99      1,
100      2,
101      3,
102      4,
103      5,
104      6,
105      7
106    ]
107  },
108  "mechanism_signature": {
109    "math_check_max_moment_residual": 1.430511474609375e-06,
110    "trained_model": {
111      "raw_norm": 39.96965408325195,
112      "compressed_relative_error": 0.40474483370780945,
113      "moment_residual": 1.3113021850585938e-06,
114      "target": "hidden moments",
115      "predicted_residual": 0.0,
116      "observed_residual": 1.3113021850585938e-06,
117      "confirmed": true
118    },
119    "predicted": "four hidden-state moments preserved after rank-2 compression",
120    "observed": "test-time hidden-state moment residual measured from trained model",
121    "confirmed": true
122  },
123  "math_verification": {
124    "max_moment_residual": 1.430511474609375e-06,
125    "passed": true
126  }
127}