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": 0.003
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.264630239456892
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.012804438127204776
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.013572414871305227
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.013181458110921085,
 33      "std": 0.006411194717789758,
 34      "per_seed": [
 35        0.01743449829518795,
 36        0.00642412668094039,
 37        0.02289336547255516,
 38        0.004465762060135603,
 39        0.010089418850839138,
 40        0.00904592964798212,
 41        0.013677984476089478,
 42        0.02142057940363884
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.01246260676998645,
 49    "std": 0.006610495470253576,
 50    "per_seed": [
 51      0.013938470743596554,
 52      0.006079694721847773,
 53      0.00940709188580513,
 54      0.00654940539970994,
 55      0.016593003645539284,
 56      0.009614327922463417,
 57      0.009874203242361546,
 58      0.027644656598567963
 59    ],
 60    "n": 8,
 61    "best_cfg": {
 62      "lr": 0.003
 63    },
 64    "sweep": [
 65      {
 66        "cfg": {
 67          "lr": 0.001
 68        },
 69        "mean": 0.2671995013952255
 70      },
 71      {
 72        "cfg": {
 73          "lr": 0.003
 74        },
 75        "mean": 0.008993665687739849
 76      },
 77      {
 78        "cfg": {
 79          "lr": 0.006
 80        },
 81        "mean": 0.010689508053474128
 82      }
 83    ]
 84  },
 85  "comparison": {
 86    "delta_mean": -0.0007188513409346342,
 87    "idea_wins": 4,
 88    "n_pairs": 8,
 89    "per_seed_diffs": [
 90      -0.0034960275515913963,
 91      -0.00034443195909261703,
 92      -0.01348627358675003,
 93      0.002083643339574337,
 94      0.006503584794700146,
 95      0.0005683982744812965,
 96      -0.003803781233727932,
 97      0.006224077194929123
 98    ],
 99    "p_value": 0.778,
100    "mde": 0.005384277405441072,
101    "mde_rel_pct": 40.847358161234816,
102    "verdict": "no significant win",
103    "system_worked": false
104  },
105  "mechanism_signature": {
106    "prediction": "trained-model gain weighting lowers normalized local certificate versus q=0",
107    "observed_normalized_mu": 17.170748114585876,
108    "observed_unscaled_mu": 18.16585087776184,
109    "relative_reduction": 0.05477875877502347,
110    "clusters_from_trained_models": [
111      [
112        0,
113        1,
114        5
115      ],
116      [
117        2,
118        4
119      ],
120      [
121        3,
122        6,
123        7
124      ]
125    ],
126    "confirmed": true
127  }
128}