{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny_factorized_first_layer",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "delta_fraction": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "delta_fraction": 0.0
        },
        "mean": 205.01904678344727
      },
      {
        "cfg": {
          "lr": 0.003,
          "delta_fraction": 0.0
        },
        "mean": 18.50063133239746
      },
      {
        "cfg": {
          "lr": 0.006,
          "delta_fraction": 0.0
        },
        "mean": 16.618733406066895
      }
    ],
    "full": {
      "mean": 16.327228546142578,
      "std": 0.907789506476808,
      "per_seed": [
        17.471485137939453,
        15.701416015625,
        16.48313331604004,
        16.818899154663086,
        16.06818199157715,
        17.633129119873047,
        15.67065715789795,
        14.770926475524902
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 16.319469451904297,
    "std": 0.9038104755432905,
    "per_seed": [
      17.471485137939453,
      15.701416015625,
      16.48313331604004,
      16.756826400756836,
      16.06818199157715,
      17.633129119873047,
      15.67065715789795,
      14.770926475524902
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": -0.00775909423828125,
    "idea_wins": 1,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.0,
      0.0,
      0.0,
      -0.06207275390625,
      0.0,
      0.0,
      0.0,
      0.0
    ],
    "p_value": 1.0,
    "mde": 0.018350257873535156,
    "mde_rel_pct": 0.11239052495452775,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "mechanism_signature": {
    "predicted_quantity": "sum of retained rank-one gradient contributions at pruning",
    "observed_quantity": "sum of retained rank-one absolute responses on held-out inputs",
    "predicted_example": 42.21419620513916,
    "observed_example": 0.8568963706493378,
    "confirmed": false
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001,
        "delta_fraction": 0.15
      },
      "mean": 204.1529769897461
    },
    {
      "cfg": {
        "lr": 0.001,
        "delta_fraction": 0.25
      },
      "mean": 204.1529769897461
    },
    {
      "cfg": {
        "lr": 0.001,
        "delta_fraction": 0.5
      },
      "mean": 204.1120777130127
    },
    {
      "cfg": {
        "lr": 0.003,
        "delta_fraction": 0.15
      },
      "mean": 19.105027437210083
    },
    {
      "cfg": {
        "lr": 0.003,
        "delta_fraction": 0.25
      },
      "mean": 19.105027437210083
    },
    {
      "cfg": {
        "lr": 0.003,
        "delta_fraction": 0.5
      },
      "mean": 19.07418918609619
    },
    {
      "cfg": {
        "lr": 0.006,
        "delta_fraction": 0.15
      },
      "mean": 16.34427547454834
    },
    {
      "cfg": {
        "lr": 0.006,
        "delta_fraction": 0.25
      },
      "mean": 16.34427547454834
    },
    {
      "cfg": {
        "lr": 0.006,
        "delta_fraction": 0.5
      },
      "mean": 16.319469451904297
    }
  ],
  "selected_idea_cfg": {
    "lr": 0.006,
    "delta_fraction": 0.5
  },
  "protocol_note": "Official registered tabular track; baseline sweep and idea use shared lr union, 8 final paired seeds."
}