1{
  2  "bench_report": {
  3    "baseline": {
  4      "best_cfg": {
  5        "lr": 0.003
  6      },
  7      "full": {
  8        "mean": 0.4984375163912773,
  9        "n": 8,
 10        "per_seed": [
 11          0.48750001192092896,
 12          0.5062500238418579,
 13          0.48125001788139343,
 14          0.4937500059604645,
 15          0.5562500357627869,
 16          0.4625000059604645,
 17          0.5062500238418579,
 18          0.4937500059604645
 19        ],
 20        "std": 0.025531468853458108
 21      },
 22      "sweep": [
 23        {
 24          "cfg": {
 25            "lr": 0.001
 26          },
 27          "mean": 0.5187500193715096
 28        },
 29        {
 30          "cfg": {
 31            "lr": 0.003
 32          },
 33          "mean": 0.4921875149011612
 34        },
 35        {
 36          "cfg": {
 37            "lr": 0.01
 38          },
 39          "mean": 0.4984375163912773
 40        }
 41      ]
 42    },
 43    "bench_version": 1,
 44    "comparison": {
 45      "delta_mean": -0.4984375163912773,
 46      "idea_wins": 8,
 47      "mde": 0.0228222220228749,
 48      "mde_rel_pct": 4.578752857150359,
 49      "n_pairs": 8,
 50      "p_value": 0.0081,
 51      "per_seed_diffs": [
 52        -0.48750001192092896,
 53        -0.5062500238418579,
 54        -0.48125001788139343,
 55        -0.4937500059604645,
 56        -0.5562500357627869,
 57        -0.4625000059604645,
 58        -0.5062500238418579,
 59        -0.4937500059604645
 60      ],
 61      "system_worked": true,
 62      "verdict": "idea better (significant)"
 63    },
 64    "idea": {
 65      "best_cfg": {
 66        "lr": 0.001
 67      },
 68      "mean": 0.0,
 69      "n": 8,
 70      "per_seed": [
 71        0.0,
 72        0.0,
 73        0.0,
 74        0.0,
 75        0.0,
 76        0.0,
 77        0.0,
 78        0.0
 79      ],
 80      "std": 0.0,
 81      "sweep": [
 82        {
 83          "cfg": {
 84            "lr": 0.001
 85          },
 86          "mean": 0.0
 87        },
 88        {
 89          "cfg": {
 90            "lr": 0.003
 91          },
 92          "mean": 0.0
 93        },
 94        {
 95          "cfg": {
 96            "lr": 0.01
 97          },
 98          "mean": 0.0
 99        }
100      ]
101    },
102    "mechanism_signature": {
103      "baseline_max_output_change": 0.4521815776824951,
104      "baseline_mean_output_change": 0.02940310724079609,
105      "confirmed": true,
106      "idea_max_output_change": 1.7881393432617188e-07,
107      "idea_mean_output_change": 3.8091179099808414e-09,
108      "observed_feature_change": 0.0,
109      "predicted_invariant_output_change": 0.0,
110      "prediction": "P2 local-correlation inputs and their trained predictions should be unchanged by global rotation and independent apolar flips."
111    },
112    "metric_direction": "lower is better",
113    "model": "mlp_tiny",
114    "n_seeds": 8,
115    "track": "orientation_phase_quotient"
116  },
117  "custom_track": {
118    "domain": "molecular_orientation_symmetry",
119    "file": "orientation_track.py",
120    "name": "orientation_phase_quotient"
121  },
122  "math_check": {
123    "max_transformed_feature_error": 0.0,
124    "predicted": 0.0
125  },
126  "protocol": {
127    "batch": 128,
128    "epochs": 18,
129    "lr_union": [
130      0.001,
131      0.003,
132      0.01
133    ],
134    "paired_seeds": [
135      0,
136      1,
137      2,
138      3,
139      4,
140      5,
141      6,
142      7
143    ]
144  }
145}