{
  "bench_version": 1,
  "track": "poisson_boundary",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.06875268183648586
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.05776503775268793
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.04322445672005415
      }
    ],
    "full": {
      "mean": 0.043870826717466116,
      "std": 0.002839556331423274,
      "per_seed": [
        0.04028090834617615,
        0.044616829603910446,
        0.04462457448244095,
        0.043375514447689056,
        0.038494233042001724,
        0.04666280001401901,
        0.046755287796258926,
        0.046156466007232666
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.06319879228249192,
    "std": 0.011146065364718148,
    "per_seed": [
      0.04028090834617615,
      0.07205105572938919,
      0.07206139713525772,
      0.055017322301864624,
      0.05459468439221382,
      0.07074009627103806,
      0.07260646671056747,
      0.06823840737342834
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.019327965565025806,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.0,
      0.027434226125478745,
      0.027436822652816772,
      0.011641807854175568,
      0.016100451350212097,
      0.024077296257019043,
      0.025851178914308548,
      0.02208194136619568
    ],
    "p_value": 0.01485,
    "mde": 0.008040142777977925,
    "mde_rel_pct": 18.326854950232647,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "mechanism_signature": {
      "selected_index": 5,
      "scores": [
        3.491577856996373,
        3.515875502902411,
        3.537788184248112,
        3.5629319410966094,
        3.474729111946311,
        3.402205193851991
      ],
      "nested_probe": {
        "2": 3.402205193851991,
        "3": 3.536084817812452,
        "4": 3.5380539738679375
      },
      "predicted_monotone": true,
      "confirmed": true
    },
    "custom_track": {
      "name": "poisson_boundary",
      "file": "poisson_track.py",
      "domain": "pde"
    }
  },
  "idea_config_sweep": [
    {
      "cfg": {
        "lr": 0.01,
        "level": 2
      },
      "mean": 0.06319879228249192,
      "per_seed": [
        0.04028090834617615,
        0.07205105572938919,
        0.07206139713525772,
        0.055017322301864624,
        0.05459468439221382,
        0.07074009627103806,
        0.07260646671056747,
        0.06823840737342834
      ]
    },
    {
      "cfg": {
        "lr": 0.01,
        "level": 3
      },
      "mean": 0.06653260113671422,
      "per_seed": [
        0.06695137917995453,
        0.07205105572938919,
        0.07206139713525772,
        0.055017322301864624,
        0.05459468439221382,
        0.07074009627103806,
        0.07260646671056747,
        0.06823840737342834
      ]
    },
    {
      "cfg": {
        "lr": 0.01,
        "level": 4
      },
      "mean": 0.06653260113671422,
      "per_seed": [
        0.06695137917995453,
        0.07205105572938919,
        0.07206139713525772,
        0.055017322301864624,
        0.05459468439221382,
        0.07074009627103806,
        0.07260646671056747,
        0.06823840737342834
      ]
    }
  ],
  "runtime_seconds": 117.34046530723572,
  "protocol_note": "Custom PDE track required; baseline and idea share mlp_tiny and Adam, differing only in archived checkpoint selector."
}