{
  "baseline": {
    "best_cfg": {
      "epochs": 35,
      "lr": 0.01,
      "width": 48
    },
    "full": {
      "mean": 0.1913430094718933,
      "n": 8,
      "per_seed": [
        0.21796821057796478,
        0.18948782980442047,
        0.18038326501846313,
        0.17539961636066437,
        0.2016671746969223,
        0.19406883418560028,
        0.18906868994235992,
        0.18270045518875122
      ],
      "std": 0.01265175674434216
    },
    "sweep": [
      {
        "cfg": {
          "epochs": 35,
          "lr": 0.001,
          "width": 48
        },
        "mean": 0.2922794222831726
      },
      {
        "cfg": {
          "epochs": 35,
          "lr": 0.003,
          "width": 48
        },
        "mean": 0.2332044392824173
      },
      {
        "cfg": {
          "epochs": 35,
          "lr": 0.01,
          "width": 48
        },
        "mean": 0.1908097304403782
      }
    ]
  },
  "bench_version": 1,
  "comparison": {
    "delta_mean": -0.10181683763221372,
    "idea_wins": 6,
    "mde": 0.08970938035328954,
    "mde_rel_pct": 46.88406469663429,
    "n_pairs": 8,
    "p_value": 0.03805,
    "per_seed_diffs": [
      -0.13966041058301926,
      -0.18819955282378942,
      -0.1548399105668068,
      -0.17267675884068012,
      0.06887666881084442,
      0.06751273572444916,
      -0.11764338612556458,
      -0.17790408665314317
    ],
    "system_worked": true,
    "verdict": "idea better (significant)"
  },
  "custom_track": {
    "domain": "graph",
    "file": "holonomy_track.py",
    "name": "directed_path_holonomy"
  },
  "idea": {
    "mean": 0.08952617183967959,
    "n": 8,
    "per_seed": [
      0.07830779999494553,
      0.0012882769806310534,
      0.02554335445165634,
      0.0027228575199842453,
      0.2705438435077667,
      0.26158156991004944,
      0.07142530381679535,
      0.004796368535608053
    ],
    "std": 0.10572229702516253
  },
  "idea_best_cfg": {
    "epochs": 35,
    "lr": 0.01,
    "width": 48
  },
  "idea_sweep": [
    {
      "cfg": {
        "epochs": 35,
        "lr": 0.001,
        "width": 48
      },
      "mean": 0.33307167142629623
    },
    {
      "cfg": {
        "epochs": 35,
        "lr": 0.003,
        "width": 48
      },
      "mean": 0.1860583908855915
    },
    {
      "cfg": {
        "epochs": 35,
        "lr": 0.01,
        "width": 48
      },
      "mean": 0.02696557223680429
    }
  ],
  "mechanism_signature": {
    "confirmed": false,
    "observed_mean": {
      "baseline_test_rmse": 0.4371945932507515,
      "idea_test_rmse": 0.2369789406657219,
      "order_sensitivity": 0.36837229039520025,
      "system_gap": 0.3969488888978958
    },
    "prediction": "ordered holonomy should retain order: reversal sensitivity exceeds additive baseline gap"
  },
  "metric_direction": "lower is better",
  "model": "matched_matrix_mlp",
  "n_seeds": 8,
  "notes": "Baseline and idea share edge encoder and head; only sum versus ordered matrix product differs.",
  "track": "directed_path_holonomy"
}