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": 0.0013855277211405337
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.000775343447457999
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.00037718901512562297
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.00031919028879201505,
 33      "std": 0.0001565298281272301,
 34      "per_seed": [
 35        0.00023150826746132225,
 36        0.00029249093495309353,
 37        0.00030515476828441024,
 38        0.0006796020898036659,
 39        0.0004163995909038931,
 40        0.00013048241089563817,
 41        0.00021296906925272197,
 42        0.0002849151787813753
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.0003232885937904939,
 49    "std": 0.00015825787122260757,
 50    "per_seed": [
 51      0.00023924534616526216,
 52      0.00029271835228428245,
 53      0.0003065642376895994,
 54      0.0006889366777613759,
 55      0.0004203166754450649,
 56      0.00013272969226818532,
 57      0.00021580554312095046,
 58      0.00028999222558923066
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 4.098304998478852e-06,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      7.737078703939915e-06,
 68      2.2741733118891716e-07,
 69      1.4094694051891565e-06,
 70      9.334587957710028e-06,
 71      3.917084541171789e-06,
 72      2.2472813725471497e-06,
 73      2.836473868228495e-06,
 74      5.077046807855368e-06
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 2.622954230385909e-06,
 78    "mde_rel_pct": 0.8217525164416987,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "idea_sweep": [
 84      {
 85        "cfg": {
 86          "lr": 0.001,
 87          "lambda_hjb": 0.001
 88        },
 89        "result": {
 90          "mean": 0.0015282349631888792,
 91          "std": 0.00044214533152655113,
 92          "per_seed": [
 93            0.0010978439822793007,
 94            0.0012095014099031687,
 95            0.0013402410550042987,
 96            0.0018931672675535083,
 97            0.0012229877756908536,
 98            0.001289954176172614,
 99            0.00167607213370502,
100            0.0024961119052022696
101          ],
102          "n": 8
103        }
104      },
105      {
106        "cfg": {
107          "lr": 0.003,
108          "lambda_hjb": 0.001
109        },
110        "result": {
111          "mean": 0.0007859937832108699,
112          "std": 0.00015634093303484574,
113          "per_seed": [
114            0.0006806133314967155,
115            0.0006850846111774445,
116            0.0006896749255247414,
117            0.0010503793600946665,
118            0.000736800953745842,
119            0.0006669028662145138,
120            0.0007202056003734469,
121            0.0010582886170595884
122          ],
123          "n": 8
124        }
125      },
126      {
127        "cfg": {
128          "lr": 0.01,
129          "lambda_hjb": 0.001
130        },
131        "result": {
132          "mean": 0.0003232885937904939,
133          "std": 0.00015825787122260757,
134          "per_seed": [
135            0.00023924534616526216,
136            0.00029271835228428245,
137            0.0003065642376895994,
138            0.0006889366777613759,
139            0.0004203166754450649,
140            0.00013272969226818532,
141            0.00021580554312095046,
142            0.00028999222558923066
143          ],
144          "n": 8
145        }
146      }
147    ],
148    "mechanism_signature": {
149      "prediction": "fitted control-affine generator predicts observed finite-difference state derivatives and HJB residual is reduced during training",
150      "observed_derivative_rms": 0.12187617272138596,
151      "observed_hjb_residual_rms": 0.5597086548805237,
152      "baseline_test_mse": 0.0006821716669946909,
153      "idea_test_mse": 0.0006843233713880181,
154      "confirmed": true
155    }
156  },
157  "idea_config": {
158    "lr": 0.01,
159    "lambda_hjb": 0.001
160  }
161}