1{
  2  "bench_report": {
  3    "bench_version": 1,
  4    "track": "dynamics",
  5    "model": "rnn_small",
  6    "metric_direction": "lower is better",
  7    "n_seeds": 8,
  8    "baseline": {
  9      "best_cfg": {
 10        "lr": 0.006
 11      },
 12      "sweep": [
 13        {
 14          "cfg": {
 15            "lr": 0.001
 16          },
 17          "mean": 0.004298779065720737
 18        },
 19        {
 20          "cfg": {
 21            "lr": 0.003
 22          },
 23          "mean": 0.0034615940821822733
 24        },
 25        {
 26          "cfg": {
 27            "lr": 0.006
 28          },
 29          "mean": 0.0018282315868418664
 30        }
 31      ],
 32      "full": {
 33        "mean": 0.002105819425196387,
 34        "std": 0.000389436566226016,
 35        "per_seed": [
 36          0.002004395006224513,
 37          0.001437882543541491,
 38          0.0015748648438602686,
 39          0.002295783953741193,
 40          0.0024020641576498747,
 41          0.0024898042902350426,
 42          0.0020851334556937218,
 43          0.0025566271506249905
 44        ],
 45        "n": 8
 46      }
 47    },
 48    "idea": {
 49      "mean": 0.09830853797029704,
 50      "std": 0.10181642123035443,
 51      "per_seed": [
 52        0.01266887690871954,
 53        0.018287021666765213,
 54        0.3273325264453888,
 55        0.034617792814970016,
 56        0.17559394240379333,
 57        0.02574852854013443,
 58        0.06928706914186478,
 59        0.1229325458407402
 60      ],
 61      "n": 8
 62    },
 63    "comparison": {
 64      "delta_mean": 0.09620271854510065,
 65      "idea_wins": 0,
 66      "n_pairs": 8,
 67      "per_seed_diffs": [
 68        0.010664481902495027,
 69        0.016849139123223722,
 70        0.3257576616015285,
 71        0.032322008861228824,
 72        0.17319187824614346,
 73        0.023258724249899387,
 74        0.06720193568617105,
 75        0.12037591869011521
 76      ],
 77      "p_value": 0.0081,
 78      "mde": 0.09108405536558592,
 79      "mde_rel_pct": 4325.3497558125855,
 80      "verdict": "idea worse (significant)",
 81      "system_worked": false
 82    },
 83    "mechanism_signature": {
 84      "mechanism_signature": {
 85        "trained_baseline_mse": 0.002604018896818161,
 86        "trained_idea_mse": 0.38107627630233765,
 87        "baseline_discrete_exponent": -0.49392220080763893,
 88        "controlled_discrete_exponent": -0.24955386218263162,
 89        "predicted_sign_preserved": true,
 90        "observed_sign_preserved": true,
 91        "confirmed": true
 92      },
 93      "idea_sweep": [
 94        {
 95          "cfg": {
 96            "lr": 0.001
 97          },
 98          "mean": 0.731351688504219
 99        },
100        {
101          "cfg": {
102            "lr": 0.003
103          },
104          "mean": 0.5189185738563538
105        },
106        {
107          "cfg": {
108            "lr": 0.006
109          },
110          "mean": 0.09822655445896089
111        }
112      ],
113      "track_justification": "Controlled pendulum rollout is the built-in stability/control/Lyapunov match.",
114      "protocol": "8 paired seeds; baseline and idea share lr union; standard test MSE."
115    }
116  }
117}