{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 130.23694229125977
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 17.47633147239685
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 11.130715131759644
      }
    ],
    "full": {
      "mean": 11.6078120470047,
      "std": 1.1735198758690657,
      "per_seed": [
        10.810468673706055,
        12.079232215881348,
        10.855374336242676,
        10.777785301208496,
        9.649698257446289,
        12.84527587890625,
        12.719051361083984,
        13.1256103515625
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 8.623416543006897,
    "std": 1.4506015215736154,
    "per_seed": [
      7.800422191619873,
      11.718671798706055,
      10.022994041442871,
      6.932704925537109,
      7.960085868835449,
      8.643534660339355,
      8.414508819580078,
      7.494410037994385
    ],
    "n": 8,
    "details": [
      {
        "metric": 7.800422191619873,
        "seconds": 7.774140733999957,
        "accepted_steps": 12,
        "median_alpha": 0.3333333333333333,
        "mean_rank": 389.5833333333333,
        "max_jacobian_condition": 3202.399169921875,
        "signature": {
          "predicted_full_to_truncated_amplification": 1000000064.0,
          "observed_full_step_norm": 0.04894990101456642,
          "observed_truncated_step_norm": 0.04894990101456642,
          "observed_amplification": 1.0,
          "confirmed": false
        }
      },
      {
        "metric": 11.718671798706055,
        "seconds": 9.029400838999209,
        "accepted_steps": 12,
        "median_alpha": 0.3333333333333333,
        "mean_rank": 390.6666666666667,
        "max_jacobian_condition": 4148.69775390625
      },
      {
        "metric": 10.022994041442871,
        "seconds": 9.43387127599999,
        "accepted_steps": 12,
        "median_alpha": 0.3333333333333333,
        "mean_rank": 400.0,
        "max_jacobian_condition": 846.0858154296875
      },
      {
        "metric": 6.932704925537109,
        "seconds": 9.006122887998572,
        "accepted_steps": 12,
        "median_alpha": 0.3333333333333333,
        "mean_rank": 400.0,
        "max_jacobian_condition": 852.4298095703125
      },
      {
        "metric": 7.960085868835449,
        "seconds": 9.027963689999524,
        "accepted_steps": 12,
        "median_alpha": 0.3333333333333333,
        "mean_rank": 370.5,
        "max_jacobian_condition": 2142.92919921875
      },
      {
        "metric": 8.643534660339355,
        "seconds": 8.847281662001478,
        "accepted_steps": 12,
        "median_alpha": 0.3333333333333333,
        "mean_rank": 399.5833333333333,
        "max_jacobian_condition": 1440.25390625
      },
      {
        "metric": 8.414508819580078,
        "seconds": 8.764552226000887,
        "accepted_steps": 12,
        "median_alpha": 0.3333333333333333,
        "mean_rank": 109.83333333333333,
        "max_jacobian_condition": 17170.849609375
      },
      {
        "metric": 7.494410037994385,
        "seconds": 8.539251098000022,
        "accepted_steps": 12,
        "median_alpha": 0.3333333333333333,
        "mean_rank": 400.0,
        "max_jacobian_condition": 614.7550048828125
      }
    ]
  },
  "comparison": {
    "delta_mean": -2.9843955039978027,
    "idea_wins": 8,
    "n_pairs": 8,
    "per_seed_diffs": [
      -3.0100464820861816,
      -0.36056041717529297,
      -0.8323802947998047,
      -3.8450803756713867,
      -1.6896123886108398,
      -4.2017412185668945,
      -4.304542541503906,
      -5.631200313568115
    ],
    "p_value": 0.0081,
    "mde": 1.5535327571979936,
    "mde_rel_pct": 13.383510612569488,
    "verdict": "idea better (significant)",
    "system_worked": true
  },
  "method_notes": {
    "track_justification": "Optimizer idea matched to tabular regression.",
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 9.118698239326477
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 10.81938362121582
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 10.81938362121582
      }
    ],
    "idea_best_cfg": {
      "lr": 0.001
    },
    "sanity": {
      "singular_values": [
        1.0,
        9.999999717180685e-10
      ],
      "full_step_norm": 100000008.0,
      "truncated_step_norm": 0.10000000149011612,
      "amplification_ratio": 1000000064.0,
      "kept_rank": 1
    },
    "epochs": 12,
    "batch": 128,
    "tau": 1e-06,
    "rho": 0.001
  },
  "mechanism_signature": {
    "predicted_full_to_truncated_amplification": 1000000064.0,
    "observed_full_step_norm": 0.04894990101456642,
    "observed_truncated_step_norm": 0.04894990101456642,
    "observed_amplification": 1.0,
    "confirmed": false
  }
}