{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "weight_decay": 0.0001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "mean": 16.501991748809814
      },
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0001
        },
        "mean": 16.475047826766968
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 10.8477942943573
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001
        },
        "mean": 10.81019401550293
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0
        },
        "mean": 7.244522333145142
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0001
        },
        "mean": 7.222299218177795
      }
    ],
    "full": {
      "mean": 7.190596640110016,
      "std": 0.44163564514728276,
      "per_seed": [
        7.42201042175293,
        7.114084720611572,
        7.479846477508545,
        6.873255252838135,
        6.192978382110596,
        7.314598083496094,
        7.447637557983398,
        7.680362224578857
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 14.991608619689941,
    "std": 1.4869862368564055,
    "per_seed": [
      15.754435539245605,
      14.699921607971191,
      14.950590133666992,
      13.512487411499023,
      12.468204498291016,
      15.603882789611816,
      15.10988712310791,
      17.833459854125977
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 7.8010119795799255,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      8.332425117492676,
      7.585836887359619,
      7.470743656158447,
      6.639232158660889,
      6.27522611618042,
      8.289284706115723,
      7.662249565124512,
      10.15309762954712
    ],
    "p_value": 0.0081,
    "mde": 0.9946617786244648,
    "mde_rel_pct": 13.832812886153583,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "prediction": "local nonlinear affine-model violation increases superlinearly with trusted radius",
    "trained_model_measurements": [
      {
        "radius_scale": 0.004526921082288027,
        "violation_raw": 0.0009449819917790592,
        "accepted_scale": 0.004526921082288027,
        "tolerance": 0.0013580762315541506
      },
      {
        "radius_scale": 0.004727812483906746,
        "violation_raw": 0.0003354753425810486,
        "accepted_scale": 0.004727812483906746,
        "tolerance": 0.0014183437451720238
      },
      {
        "radius_scale": 0.004053650889545679,
        "violation_raw": 0.0005828518187627196,
        "accepted_scale": 0.004053650889545679,
        "tolerance": 0.0012160951737314463
      },
      {
        "radius_scale": 0.003885602578520775,
        "violation_raw": 0.00023847918782848865,
        "accepted_scale": 0.003885602578520775,
        "tolerance": 0.0011656807037070394
      },
      {
        "radius_scale": 0.005713270511478186,
        "violation_raw": 0.0006674728938378394,
        "accepted_scale": 0.005713270511478186,
        "tolerance": 0.0017139810370281339
      },
      {
        "radius_scale": 0.0043457550927996635,
        "violation_raw": 0.00025109699345193803,
        "accepted_scale": 0.0043457550927996635,
        "tolerance": 0.001303726457990706
      },
      {
        "radius_scale": 0.005000402685254812,
        "violation_raw": 0.0010114216711372137,
        "accepted_scale": 0.005000402685254812,
        "tolerance": 0.0015001207357272506
      },
      {
        "radius_scale": 0.004544075112789869,
        "violation_raw": 0.0003719122614711523,
        "accepted_scale": 0.004544075112789869,
        "tolerance": 0.0013632223708555102
      },
      {
        "radius_scale": 0.013580761849880219,
        "violation_raw": 0.008852558210492134,
        "accepted_scale": 0.009213240817189217,
        "tolerance": 0.004074228461831808
      },
      {
        "radius_scale": 0.014183436520397663,
        "violation_raw": 0.003163333749398589,
        "accepted_scale": 0.014183436520397663,
        "tolerance": 0.004255030769854784
      },
      {
        "radius_scale": 0.01216095220297575,
        "violation_raw": 0.005176236387342215,
        "accepted_scale": 0.010209511034190655,
        "tolerance": 0.003648285521194339
      },
      {
        "radius_scale": 0.0116568086668849,
        "violation_raw": 0.0022630963940173388,
        "accepted_scale": 0.0116568086668849,
        "tolerance": 0.0034970424603670835
      },
      {
        "radius_scale": 0.01713980920612812,
        "violation_raw": 0.005821220576763153,
        "accepted_scale": 0.01610877737402916,
        "tolerance": 0.005141942761838436
      },
      {
        "radius_scale": 0.013037264347076416,
        "violation_raw": 0.0011820535873994231,
        "accepted_scale": 0.013037264347076416,
        "tolerance": 0.003911179024726152
      },
      {
        "radius_scale": 0.015001206658780575,
        "violation_raw": 0.009202888235449791,
        "accepted_scale": 0.0104902982711792,
        "tolerance": 0.004500361625105143
      },
      {
        "radius_scale": 0.01363222487270832,
        "violation_raw": 0.003228261601179838,
        "accepted_scale": 0.01363222487270832,
        "tolerance": 0.004089667461812496
      },
      {
        "radius_scale": 0.045269206166267395,
        "violation_raw": 0.09564003348350525,
        "accepted_scale": 0.017058666795492172,
        "tolerance": 0.013580761849880219
      },
      {
        "radius_scale": 0.04727812111377716,
        "violation_raw": 0.03237629309296608,
        "accepted_scale": 0.03129229322075844,
        "tolerance": 0.014183435589075089
      },
      {
        "radius_scale": 0.04053650423884392,
        "violation_raw": 0.05683043599128723,
        "accepted_scale": 0.018751652911305428,
        "tolerance": 0.012160951271653175
      },
      {
        "radius_scale": 0.03885602578520775,
        "violation_raw": 0.02356799878180027,
        "accepted_scale": 0.02732667699456215,
        "tolerance": 0.01165680680423975
      },
      {
        "radius_scale": 0.057132698595523834,
        "violation_raw": 0.06333432346582413,
        "accepted_scale": 0.029721327126026154,
        "tolerance": 0.01713980920612812
      },
      {
        "radius_scale": 0.04345754534006119,
        "violation_raw": 0.008958065882325172,
        "accepted_scale": 0.04345754534006119,
        "tolerance": 0.013037262484431267
      },
      {
        "radius_scale": 0.0500040166079998,
        "violation_raw": 0.09854104369878769,
        "accepted_scale": 0.019510094076395035,
        "tolerance": 0.015001204796135426
      },
      {
        "radius_scale": 0.04544074460864067,
        "violation_raw": 0.03523782640695572,
        "accepted_scale": 0.028263390064239502,
        "tolerance": 0.013632223010063171
      }
    ],
    "confirmed": true,
    "loglog_slope_radius_violation": 1.9431904882117401,
    "spearman_like_rank_correlation": 0.9486956521739129,
    "confirmation_rule": "slope > 1 and rank correlation > 0.5"
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001,
        "radius": 0.1
      },
      "result": {
        "mean": 211.3460350036621,
        "std": 6.958365136348233,
        "per_seed": [
          216.36398315429688,
          211.4189453125,
          219.05775451660156,
          203.28623962402344,
          208.64508056640625,
          212.2245330810547,
          199.26976013183594,
          220.50198364257812
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.003,
        "radius": 0.1
      },
      "result": {
        "mean": 72.21579933166504,
        "std": 26.963887190478438,
        "per_seed": [
          75.80757904052734,
          56.88224411010742,
          93.34162139892578,
          92.83026123046875,
          23.559711456298828,
          91.82359313964844,
          40.557342529296875,
          102.92404174804688
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01,
        "radius": 0.1
      },
      "result": {
        "mean": 14.991608619689941,
        "std": 1.4869862368564055,
        "per_seed": [
          15.754435539245605,
          14.699921607971191,
          14.950590133666992,
          13.512487411499023,
          12.468204498291016,
          15.603882789611816,
          15.10988712310791,
          17.833459854125977
        ],
        "n": 8
      }
    }
  ],
  "protocol_notes": {
    "epochs": 10,
    "n_train": 1200,
    "n_test": 400,
    "device": "cuda",
    "shared_architecture": true,
    "track_reason": "optimizer idea mapped to Friedman tabular regression"
  }
}