{
  "bench_version": 1,
  "track": "piola_surface_operator",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.003
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.398487139493227
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.3586462587118149
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.46151376888155937
      }
    ],
    "full": {
      "mean": 0.33389632031321526,
      "std": 0.010530313211265758,
      "per_seed": [
        0.32944396138191223,
        0.3304601013660431,
        0.3288800120353699,
        0.32970303297042847,
        0.3304898738861084,
        0.3331677317619324,
        0.36147114634513855,
        0.32755470275878906
      ],
      "n": 8
    },
    "parity_grid": [
      {
        "lr": 0.001
      },
      {
        "lr": 0.003
      },
      {
        "lr": 0.01
      }
    ],
    "selected_lr": 0.003,
    "idea_grid": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.33370736241340637
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.24209699407219887
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.2798946388065815
      }
    ]
  },
  "idea": {
    "mean": 0.24209699407219887,
    "std": 0.007792167833569856,
    "per_seed": [
      0.23542946577072144,
      0.23528780043125153,
      0.23226389288902283,
      0.2527761161327362,
      0.23969487845897675,
      0.24286238849163055,
      0.2553972601890564,
      0.24306415021419525
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": -0.09179932624101639,
    "idea_wins": 8,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.0940144956111908,
      -0.09517230093479156,
      -0.09661611914634705,
      -0.07692691683769226,
      -0.09079499542713165,
      -0.09030534327030182,
      -0.10607388615608215,
      -0.08449055254459381
    ],
    "p_value": 0.0081,
    "mde": 0.007214329180313888,
    "mde_rel_pct": 2.160649501481898,
    "verdict": "idea better (significant)",
    "system_worked": true
  },
  "custom_track": {
    "name": "piola_surface_operator",
    "file": "piola_surface_operator.py",
    "domain": "pde"
  },
  "protocol_notes": {
    "epochs": 18,
    "batch": 64,
    "seeds": [
      11,
      29,
      47,
      71,
      89,
      107,
      131,
      149
    ],
    "baseline_and_idea_same_architecture": true,
    "baseline_and_idea_same_lr_union": true
  },
  "mechanism_signature": {
    "quantity": "trained reference prediction RMSE and cross-geometry output consistency",
    "baseline_observed_reference_prediction_rmse": 0.5627477765083313,
    "idea_observed_reference_prediction_rmse": 0.4821018874645233,
    "baseline_geometry_output_consistency": 0.31939321465817877,
    "idea_geometry_output_consistency": 0.03450808376451315,
    "predicted_effect": "inverse-Piola inputs should reduce geometry-induced variation in reference predictions",
    "confirmed": true
  }
}