{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.003
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.010543678537942469
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.008196698385290802
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.009198607411235571
      }
    ],
    "full": {
      "mean": 0.007723776740022004,
      "std": 0.0016993505269699558,
      "per_seed": [
        0.0053276014514267445,
        0.00833857711404562,
        0.008281959220767021,
        0.01083865575492382,
        0.008155343122780323,
        0.005130748730152845,
        0.007964092306792736,
        0.007753236219286919
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.008810116676613688,
    "std": 0.0023181915418372715,
    "per_seed": [
      0.009634214453399181,
      0.007696767803281546,
      0.01108845230191946,
      0.013114262372255325,
      0.009547872468829155,
      0.006006436888128519,
      0.006788569502532482,
      0.006604357622563839
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.0010863399365916848,
    "idea_wins": 3,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.004306613001972437,
      -0.0006418093107640743,
      0.002806493081152439,
      0.002275606617331505,
      0.001392529346048832,
      0.0008756881579756737,
      -0.001175522804260254,
      -0.0011488785967230797
    ],
    "p_value": 0.1829,
    "mde": 0.0016717188889132512,
    "mde_rel_pct": 21.643801279896767,
    "verdict": "no significant win",
    "system_worked": false,
    "idea_sweep_results": [
      {
        "cfg": {
          "lr": 0.003,
          "steps": 8,
          "step_size": 0.2
        },
        "result": {
          "mean": 0.010117848869413137,
          "std": 0.002055095696012117,
          "per_seed": [
            0.006481844000518322,
            0.009532518684864044,
            0.011807672679424286,
            0.014011072926223278,
            0.009151139296591282,
            0.010334501974284649,
            0.010534162633121014,
            0.009089878760278225
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.001,
          "steps": 8,
          "step_size": 0.2
        },
        "result": {
          "mean": 0.011166991549544036,
          "std": 0.0018518637822724081,
          "per_seed": [
            0.008419685997068882,
            0.009691255167126656,
            0.01094791665673256,
            0.013981804251670837,
            0.012324392795562744,
            0.011532868258655071,
            0.013251206837594509,
            0.009186802431941032
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.01,
          "steps": 8,
          "step_size": 0.2
        },
        "result": {
          "mean": 0.008810116676613688,
          "std": 0.0023181915418372715,
          "per_seed": [
            0.009634214453399181,
            0.007696767803281546,
            0.01108845230191946,
            0.013114262372255325,
            0.009547872468829155,
            0.006006436888128519,
            0.006788569502532482,
            0.006604357622563839
          ],
          "n": 8
        }
      }
    ]
  },
  "track_justification": "Dynamics is structurally matched: the built-in task is an actuated pendulum rollout whose target depends on coupled state-transition mechanisms.",
  "baseline_sweep_union_lr": [
    0.001,
    0.003,
    0.01
  ],
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.003,
        "steps": 8,
        "step_size": 0.2
      },
      "result": {
        "mean": 0.010117848869413137,
        "std": 0.002055095696012117,
        "per_seed": [
          0.006481844000518322,
          0.009532518684864044,
          0.011807672679424286,
          0.014011072926223278,
          0.009151139296591282,
          0.010334501974284649,
          0.010534162633121014,
          0.009089878760278225
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.001,
        "steps": 8,
        "step_size": 0.2
      },
      "result": {
        "mean": 0.011166991549544036,
        "std": 0.0018518637822724081,
        "per_seed": [
          0.008419685997068882,
          0.009691255167126656,
          0.01094791665673256,
          0.013981804251670837,
          0.012324392795562744,
          0.011532868258655071,
          0.013251206837594509,
          0.009186802431941032
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01,
        "steps": 8,
        "step_size": 0.2
      },
      "result": {
        "mean": 0.008810116676613688,
        "std": 0.0023181915418372715,
        "per_seed": [
          0.009634214453399181,
          0.007696767803281546,
          0.01108845230191946,
          0.013114262372255325,
          0.009547872468829155,
          0.006006436888128519,
          0.006788569502532482,
          0.006604357622563839
        ],
        "n": 8
      }
    }
  ],
  "idea_selected_cfg": {
    "lr": 0.01,
    "steps": 8,
    "step_size": 0.2
  },
  "parameter_counts": {
    "baseline": 3601,
    "idea": 11043
  },
  "mechanism_signature": {
    "prediction": "same xi imposed by different equation replacements yields different downstream state",
    "n_samples": 64,
    "xi": 0.08397173136472702,
    "trained_model_target_error_f0": 0.44350090622901917,
    "trained_model_target_error_f1": 0.44543659687042236,
    "trained_model_downstream_separation": 0.2748295068740845,
    "baseline_vs_observational_prediction_abs_gap": 0.05360101908445358,
    "confirmed": false
  }
}