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": 0.001154391880845651
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0006730457389494404
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.0004064472232130356
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0003541622754710261,
 33      "std": 9.525887076242907e-05,
 34      "per_seed": [
 35        0.0004017692699562758,
 36        0.00043160535278730094,
 37        0.00029359731706790626,
 38        0.0004988169530406594,
 39        0.0003143931389786303,
 40        0.00029283223557285964,
 41        0.00018064776668325067,
 42        0.00041963616968132555
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.7773510366678238,
 49    "std": 0.04344097334599117,
 50    "per_seed": [
 51      0.7313887476921082,
 52      0.8112629652023315,
 53      0.8064488172531128,
 54      0.8520755171775818,
 55      0.7981084585189819,
 56      0.7174731492996216,
 57      0.7519327402114868,
 58      0.7501178979873657
 59    ],
 60    "n": 8,
 61    "cfg": {
 62      "lr": 0.006,
 63      "beta": 0.01,
 64      "gamma": 0.02
 65    },
 66    "sweep": [
 67      {
 68        "cfg": {
 69          "lr": 0.001,
 70          "beta": 0.01,
 71          "gamma": 0.02
 72        },
 73        "mean": 0.8052886575460434
 74      },
 75      {
 76        "cfg": {
 77          "lr": 0.003,
 78          "beta": 0.01,
 79          "gamma": 0.02
 80        },
 81        "mean": 0.8011865019798279
 82      },
 83      {
 84        "cfg": {
 85          "lr": 0.006,
 86          "beta": 0.01,
 87          "gamma": 0.02
 88        },
 89        "mean": 0.8002940118312836
 90      }
 91    ],
 92    "signature_per_seed": [
 93      {
 94        "info_nats": -0.11848548799753189,
 95        "noise_sigma": 0.24387867748737335,
 96        "reverse_mse": 0.8134815096855164
 97      },
 98      {
 99        "info_nats": -0.114313505589962,
100        "noise_sigma": 0.25354620814323425,
101        "reverse_mse": 0.8361767530441284
102      },
103      {
104        "info_nats": -0.022014688700437546,
105        "noise_sigma": 0.23950694501399994,
106        "reverse_mse": 0.8750950694084167
107      },
108      {
109        "info_nats": -0.10838189721107483,
110        "noise_sigma": 0.26771512627601624,
111        "reverse_mse": 0.8871616125106812
112      },
113      {
114        "info_nats": -0.061512164771556854,
115        "noise_sigma": 0.250619113445282,
116        "reverse_mse": 0.8385130167007446
117      },
118      {
119        "info_nats": -0.08916227519512177,
120        "noise_sigma": 0.24852284789085388,
121        "reverse_mse": 0.7741023898124695
122      },
123      {
124        "info_nats": -0.05122513696551323,
125        "noise_sigma": 0.25014737248420715,
126        "reverse_mse": 0.8042869567871094
127      },
128      {
129        "info_nats": -0.12155700474977493,
130        "noise_sigma": 0.23886628448963165,
131        "reverse_mse": 0.8002150058746338
132      }
133    ]
134  },
135  "comparison": {
136    "delta_mean": 0.7769968743923528,
137    "idea_wins": 0,
138    "n_pairs": 8,
139    "per_seed_diffs": [
140      0.7309869784221519,
141      0.8108313598495442,
142      0.8061552199360449,
143      0.8515767002245411,
144      0.7977940653800033,
145      0.7171803170640487,
146      0.7517520924448036,
147      0.7496982618176844
148    ],
149    "p_value": 0.0081,
150    "mde": 0.03879287423905792,
151    "mde_rel_pct": 10953.417945901907,
152    "verdict": "idea worse (significant)",
153    "system_worked": false
154  },
155  "mechanism_signature": {
156    "track_choice": "dynamics: actuated pendulum rollout is structurally matched to control/stability.",
157    "prediction": "The intended prediction was lower information with a useful reverse prior; the trained signature did not validate this quantitatively because the estimated information was negative and no baseline information measurement was recorded.",
158    "predicted_info_nats": -0.08583152014762163,
159    "observed_reverse_mse": 0.8286290392279625,
160    "observed_noise_sigma": 0.2491003219038248,
161    "confirmed": false
162  }
163}