{
  "bench_version": 1,
  "track": "budgeted_route_costs",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "wd": 0.0001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "wd": 0.0
        },
        "mean": 0.024585966020822525
      },
      {
        "cfg": {
          "lr": 0.001,
          "wd": 0.0001
        },
        "mean": 0.02389408415183425
      },
      {
        "cfg": {
          "lr": 0.003,
          "wd": 0.0
        },
        "mean": 0.004350764560513198
      },
      {
        "cfg": {
          "lr": 0.003,
          "wd": 0.0001
        },
        "mean": 0.004227392724715173
      },
      {
        "cfg": {
          "lr": 0.01,
          "wd": 0.0
        },
        "mean": 0.002535940962843597
      },
      {
        "cfg": {
          "lr": 0.01,
          "wd": 0.0001
        },
        "mean": 0.0022327409242279828
      }
    ],
    "full": {
      "mean": 0.0023824743693694472,
      "std": 0.0003298540492088312,
      "per_seed": [
        0.002391217043623328,
        0.0022307508625090122,
        0.0022920493502169847,
        0.002016946440562606,
        0.0019817666616290808,
        0.003033609827980399,
        0.0027303267270326614,
        0.0023831280414015055
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.12442152807489038,
    "std": 0.06743830819346265,
    "per_seed": [
      0.09885232150554657,
      0.07384644448757172,
      0.14970475435256958,
      0.25081515312194824,
      0.05385749414563179,
      0.21027871966362,
      0.08540374785661697,
      0.07261358946561813
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.12203905370552093,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.09646110446192324,
      0.0716156936250627,
      0.1474127050023526,
      0.24879820668138564,
      0.05187572748400271,
      0.2072451098356396,
      0.08267342112958431,
      0.07023046142421663
    ],
    "p_value": 0.0081,
    "mde": 0.06023392098974088,
    "mde_rel_pct": 2528.208561827365,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "custom_track": {
    "name": "budgeted_route_costs",
    "file": "route_track.py",
    "domain": "graph-decision"
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.01,
        "wd": 0.0001
      },
      "result": {
        "mean": 0.12442152807489038,
        "std": 0.06743830819346265,
        "per_seed": [
          0.09885232150554657,
          0.07384644448757172,
          0.14970475435256958,
          0.25081515312194824,
          0.05385749414563179,
          0.21027871966362,
          0.08540374785661697,
          0.07261358946561813
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.001,
        "wd": 0.0001
      },
      "result": {
        "mean": 0.27160821482539177,
        "std": 0.05016213483864423,
        "per_seed": [
          0.3274783790111542,
          0.2356351912021637,
          0.31766167283058167,
          0.16266748309135437,
          0.25924399495124817,
          0.2929310202598572,
          0.2722110152244568,
          0.3050369620323181
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01,
        "wd": 0.0001
      },
      "result": {
        "mean": 0.12442152807489038,
        "std": 0.06743830819346265,
        "per_seed": [
          0.09885232150554657,
          0.07384644448757172,
          0.14970475435256958,
          0.25081515312194824,
          0.05385749414563179,
          0.21027871966362,
          0.08540374785661697,
          0.07261358946561813
        ],
        "n": 8
      }
    }
  ],
  "mechanism_signature": {
    "baseline_worst_regret_mean": 0.006290760776028037,
    "idea_worst_regret_mean": 0.006380649516358972,
    "predicted_effect": "adversarial training reduces budgeted path flips/regret",
    "confirmed": false
  }
}