1{
  2  "bench_version": 1,
  3  "track": "bounded_energy_regression",
  4  "model": "energy_potential_mlp",
  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": 0.01
 15        },
 16        "mean": 0.7767523378133774
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.03
 21        },
 22        "mean": 0.8518001586198807
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.08
 27        },
 28        "mean": 1.003566414117813
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.7746003046631813,
 33      "std": 0.02759112122134984,
 34      "per_seed": [
 35        0.7906999588012695,
 36        0.7522919178009033,
 37        0.7769189476966858,
 38        0.7870985269546509,
 39        0.8022679090499878,
 40        0.781724214553833,
 41        0.7115864157676697,
 42        0.7942145466804504
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.7615404203534126,
 49    "std": 0.03034974678225816,
 50    "per_seed": [
 51      0.774912416934967,
 52      0.7445108890533447,
 53      0.7894505262374878,
 54      0.7848820090293884,
 55      0.762522280216217,
 56      0.777182400226593,
 57      0.6890047788619995,
 58      0.7698580622673035
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -0.013059884309768677,
 64    "idea_wins": 7,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.01578754186630249,
 68      -0.007781028747558594,
 69      0.012531578540802002,
 70      -0.002216517925262451,
 71      -0.03974562883377075,
 72      -0.00454181432723999,
 73      -0.022581636905670166,
 74      -0.024356484413146973
 75    ],
 76    "p_value": 0.06495,
 77    "mde": 0.013448348264850439,
 78    "mde_rel_pct": 1.7361661470941674,
 79    "verdict": "no significant win",
 80    "system_worked": false
 81  },
 82  "idea_sweep": [
 83    {
 84      "cfg": {
 85        "lr": 0.01
 86      },
 87      "result": {
 88        "mean": 0.7615404203534126,
 89        "std": 0.03034974678225816,
 90        "per_seed": [
 91          0.774912416934967,
 92          0.7445108890533447,
 93          0.7894505262374878,
 94          0.7848820090293884,
 95          0.762522280216217,
 96          0.777182400226593,
 97          0.6890047788619995,
 98          0.7698580622673035
 99        ],
100        "n": 8
101      }
102    },
103    {
104      "cfg": {
105        "lr": 0.03
106      },
107      "result": {
108        "mean": 0.7638429179787636,
109        "std": 0.032194282961917974,
110        "per_seed": [
111          0.7802979946136475,
112          0.7446085810661316,
113          0.7884130477905273,
114          0.7881810665130615,
115          0.7655128836631775,
116          0.7819029688835144,
117          0.6864311099052429,
118          0.7753956913948059
119        ],
120        "n": 8
121      }
122    },
123    {
124      "cfg": {
125        "lr": 0.08
126      },
127      "result": {
128        "mean": 0.7687475085258484,
129        "std": 0.031868273247181104,
130        "per_seed": [
131          0.7850456833839417,
132          0.7469156980514526,
133          0.7934955954551697,
134          0.7940797805786133,
135          0.771902859210968,
136          0.7864246964454651,
137          0.6932292580604553,
138          0.7788864970207214
139        ],
140        "n": 8
141      }
142    }
143  ],
144  "structural_match": "Registered bounded_energy_regression: 2-D energy surface; latent coordinates are normalized particles conditioned on observed energy.",
145  "mechanism_signature": {
146    "prediction": "trained convex posterior head has covariance Hessian PSD",
147    "observed_gradient_norm": 0.052461687475442886,
148    "observed_covariance_min_eigenvalue": 0.004702107980847359,
149    "confirmed": true
150  }
151}