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      "penalty": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "penalty": 0.0
 17        },
 18        "mean": 0.11797566118184477
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "penalty": 0.0
 24        },
 25        "mean": 0.007828571717254817
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "penalty": 0.0
 31        },
 32        "mean": 0.0025760137650649995
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.002240322035504505,
 37      "std": 0.0007111673246843795,
 38      "per_seed": [
 39        0.0027267802506685257,
 40        0.003143619978800416,
 41        0.0015068432549014688,
 42        0.0029268115758895874,
 43        0.0025003517512232065,
 44        0.001474660588428378,
 45        0.002521744929254055,
 46        0.0011217639548704028
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.0023943046835483983,
 53    "std": 0.0006552643263981685,
 54    "per_seed": [
 55      0.001565085374750197,
 56      0.001931023900397122,
 57      0.003373584244400263,
 58      0.002406168496236205,
 59      0.0027771489694714546,
 60      0.0014064890565350652,
 61      0.002705090679228306,
 62      0.002989846747368574
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": 0.00015398264804389328,
 68    "idea_wins": 4,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      -0.0011616948759183288,
 72      -0.001212596078403294,
 73      0.001866740989498794,
 74      -0.0005206430796533823,
 75      0.0002767972182482481,
 76      -6.817153189331293e-05,
 77      0.0001833457499742508,
 78      0.0018680827924981713
 79    ],
 80    "p_value": 0.69925,
 81    "mde": 0.0009989272382128146,
 82    "mde_rel_pct": 44.58855568002585,
 83    "verdict": "no significant win",
 84    "system_worked": false
 85  },
 86  "math_sanity": {
 87    "a": 0.8,
 88    "sigma": [
 89      [
 90        0.001565085374750197,
 91        1.0062328577041626,
 92        1.0018754711727258
 93      ],
 94      [
 95        0.001931023900397122,
 96        1.0306692123413086,
 97        1.001999129195911
 98      ],
 99      [
100        0.003373584244400263,
101        1.0256351232528687,
102        1.0008405618216547
103      ],
104      [
105        0.002406168496236205,
106        1.0819001197814941,
107        1.000112452055684
108      ],
109      [
110        0.0027771489694714546,
111        0.9877385497093201,
112        1.001914544541687
113      ],
114      [
115        0.0014064890565350652,
116        1.1168031692504883,
117        1.0022314286996865
118      ],
119      [
120        0.002705090679228306,
121        1.0017019510269165,
122        0.9949402147998869
123      ],
124      [
125        0.002989846747368574,
126        0.9434056282043457,
127        0.9980177719281845
128      ]
129    ],
130    "K_scalar": 1.0
131  },
132  "idea_sweep": {
133    "configs": [
134      {
135        "lr": 0.01,
136        "penalty": 1.0
137      },
138      {
139        "lr": 0.001,
140        "penalty": 3.0
141      },
142      {
143        "lr": 0.003,
144        "penalty": 10.0
145      }
146    ],
147    "best_cfg": {
148      "lr": 0.01,
149      "penalty": 1.0
150    }
151  },
152  "mechanism_signature": {
153    "noise_sigma": 0.2,
154    "predicted_lifted_rho_mean": 1.024260826408863,
155    "observed_hidden_second_moment_ratio_mean": 1.0002414467769276,
156    "relative_error": 0.02345045227996197,
157    "confirmed": true,
158    "source": "trained idea GRU models on dynamics test rollout"
159  }
160}