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      "epochs": 12
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "epochs": 12
 17        },
 18        "mean": 0.004298779065720737
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "epochs": 12
 24        },
 25        "mean": 0.0034615940821822733
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "epochs": 12
 31        },
 32        "mean": 0.0018282315868418664
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.002105819425196387,
 37      "std": 0.000389436566226016,
 38      "per_seed": [
 39        0.002004395006224513,
 40        0.001437882543541491,
 41        0.0015748648438602686,
 42        0.002295783953741193,
 43        0.0024020641576498747,
 44        0.0024898042902350426,
 45        0.0020851334556937218,
 46        0.0025566271506249905
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.0018193616560893133,
 53    "std": 0.00037736014800368497,
 54    "per_seed": [
 55      0.0017986779566854239,
 56      0.0012207203544676304,
 57      0.0014683871995657682,
 58      0.0016677540261298418,
 59      0.0024448649492114782,
 60      0.002230470534414053,
 61      0.0020495401695370674,
 62      0.0016744780587032437
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": -0.00028645776910707355,
 68    "idea_wins": 7,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      -0.0002057170495390892,
 72      -0.00021716218907386065,
 73      -0.00010647764429450035,
 74      -0.000628029927611351,
 75      4.2800791561603546e-05,
 76      -0.0002593337558209896,
 77      -3.559328615665436e-05,
 78      -0.0008821490919217467
 79    ],
 80    "p_value": 0.0224,
 81    "mde": 0.0002620079869585246,
 82    "mde_rel_pct": 12.442091844322785,
 83    "verdict": "idea better (significant)",
 84    "system_worked": true
 85  },
 86  "mechanism_signature": {
 87    "signature": {
 88      "q": 0.03,
 89      "Gamma": 2.0,
 90      "epsilon": 0.13,
 91      "predicted_uniform_radius": 0.17285714285714285,
 92      "observed_proxy_radius_mean": 0.1817476600408554,
 93      "observed_proxy_radius_std": 0.02882220596075058,
 94      "predicted_vs_observed_correlation": NaN,
 95      "controller_refines_when_bound_exceeds_epsilon": true,
 96      "confirmed": false,
 97      "note": "trained benchmark models use adaptive resampling, but the conservative fixed-gap law is not empirically calibrated by this track"
 98    },
 99    "idea_sweep": [
100      {
101        "cfg": {
102          "lr": 0.001,
103          "epochs": 12
104        },
105        "result": {
106          "mean": 0.0052960661123506725,
107          "std": 0.0015572022789016526,
108          "per_seed": [
109            0.008267243392765522,
110            0.005108773708343506,
111            0.0028678099624812603,
112            0.004508262034505606,
113            0.005263541359454393,
114            0.004577877465635538,
115            0.0047037601470947266,
116            0.007071260828524828
117          ],
118          "n": 8
119        }
120      },
121      {
122        "cfg": {
123          "lr": 0.003,
124          "epochs": 12
125        },
126        "result": {
127          "mean": 0.002939960832009092,
128          "std": 0.0017693752721639644,
129          "per_seed": [
130            0.0019986729603260756,
131            0.002653212519362569,
132            0.0012997412122786045,
133            0.007327801547944546,
134            0.0022197901271283627,
135            0.0036344309337437153,
136            0.0022852604743093252,
137            0.002100776880979538
138          ],
139          "n": 8
140        }
141      },
142      {
143        "cfg": {
144          "lr": 0.006,
145          "epochs": 12
146        },
147        "result": {
148          "mean": 0.0018193616560893133,
149          "std": 0.00037736014800368497,
150          "per_seed": [
151            0.0017986779566854239,
152            0.0012207203544676304,
153            0.0014683871995657682,
154            0.0016677540261298418,
155            0.0024448649492114782,
156            0.002230470534414053,
157            0.0020495401695370674,
158            0.0016744780587032437
159          ],
160          "n": 8
161        }
162      }
163    ]
164  },
165  "idea_selected_cfg": {
166    "lr": 0.006,
167    "epochs": 12
168  },
169  "protocol_note": "Baseline and idea share rnn_small, data, seeds, epochs and learning-rate union; adaptive temporal resampling is the sole intervention."
170}