1{
  2  "bench_version": 1,
  3  "track": "braid_garside_classification",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.001
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.508124977350235
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.5193749815225601
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.513124980032444
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.5015624798834324,
 33      "std": 0.0222534158322699,
 34      "per_seed": [
 35        0.48249998688697815,
 36        0.48499998450279236,
 37        0.5374999642372131,
 38        0.5274999737739563,
 39        0.5224999785423279,
 40        0.4899999797344208,
 41        0.4749999940395355,
 42        0.492499977350235
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.5053124837577343,
 49    "std": 0.041011381244236006,
 50    "per_seed": [
 51      0.512499988079071,
 52      0.4899999797344208,
 53      0.5324999690055847,
 54      0.4099999964237213,
 55      0.550000011920929,
 56      0.492499977350235,
 57      0.5374999642372131,
 58      0.5174999833106995
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.0037500038743019104,
 64    "idea_wins": 2,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.030000001192092896,
 68      0.004999995231628418,
 69      -0.004999995231628418,
 70      -0.11749997735023499,
 71      0.027500033378601074,
 72      0.002499997615814209,
 73      0.06249997019767761,
 74      0.025000005960464478
 75    ],
 76    "p_value": 0.85155,
 77    "mde": 0.044624401409542755,
 78    "mde_rel_pct": 8.89707727338653,
 79    "verdict": "no measurable effect",
 80    "system_worked": false
 81  },
 82  "custom_track": {
 83    "name": "braid_garside_classification",
 84    "file": "garside_track.py",
 85    "domain": "algebraic braid presentations"
 86  },
 87  "protocol_note": "Baseline and idea use the same MLP, seed data, epochs, batch size, and shared lr grid; only representation differs.",
 88  "mechanism_signature": {
 89    "trained_factor_prediction_agreement": 1.0,
 90    "observed_generator_to_factor_ratio": 3.489510438346376,
 91    "expected_invariance": 1.0,
 92    "confirmed": true
 93  },
 94  "idea_sweep": [
 95    {
 96      "cfg": {
 97        "lr": 0.001
 98      },
 99      "mean": 0.5181249752640724
100    },
101    {
102      "cfg": {
103        "lr": 0.003
104      },
105      "mean": 0.48624998331069946
106    },
107    {
108      "cfg": {
109        "lr": 0.01
110      },
111      "mean": 0.5106249898672104
112    }
113  ]
114}