{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.0013855277211405337
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.000775343447457999
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.00037718901512562297
      }
    ],
    "full": {
      "mean": 0.00031919028879201505,
      "std": 0.0001565298281272301,
      "per_seed": [
        0.00023150826746132225,
        0.00029249093495309353,
        0.00030515476828441024,
        0.0006796020898036659,
        0.0004163995909038931,
        0.00013048241089563817,
        0.00021296906925272197,
        0.0002849151787813753
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.0003232885937904939,
    "std": 0.00015825787122260757,
    "per_seed": [
      0.00023924534616526216,
      0.00029271835228428245,
      0.0003065642376895994,
      0.0006889366777613759,
      0.0004203166754450649,
      0.00013272969226818532,
      0.00021580554312095046,
      0.00028999222558923066
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 4.098304998478852e-06,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      7.737078703939915e-06,
      2.2741733118891716e-07,
      1.4094694051891565e-06,
      9.334587957710028e-06,
      3.917084541171789e-06,
      2.2472813725471497e-06,
      2.836473868228495e-06,
      5.077046807855368e-06
    ],
    "p_value": 0.0081,
    "mde": 2.622954230385909e-06,
    "mde_rel_pct": 0.8217525164416987,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "lambda_hjb": 0.001
        },
        "result": {
          "mean": 0.0015282349631888792,
          "std": 0.00044214533152655113,
          "per_seed": [
            0.0010978439822793007,
            0.0012095014099031687,
            0.0013402410550042987,
            0.0018931672675535083,
            0.0012229877756908536,
            0.001289954176172614,
            0.00167607213370502,
            0.0024961119052022696
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.003,
          "lambda_hjb": 0.001
        },
        "result": {
          "mean": 0.0007859937832108699,
          "std": 0.00015634093303484574,
          "per_seed": [
            0.0006806133314967155,
            0.0006850846111774445,
            0.0006896749255247414,
            0.0010503793600946665,
            0.000736800953745842,
            0.0006669028662145138,
            0.0007202056003734469,
            0.0010582886170595884
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.01,
          "lambda_hjb": 0.001
        },
        "result": {
          "mean": 0.0003232885937904939,
          "std": 0.00015825787122260757,
          "per_seed": [
            0.00023924534616526216,
            0.00029271835228428245,
            0.0003065642376895994,
            0.0006889366777613759,
            0.0004203166754450649,
            0.00013272969226818532,
            0.00021580554312095046,
            0.00028999222558923066
          ],
          "n": 8
        }
      }
    ],
    "mechanism_signature": {
      "prediction": "fitted control-affine generator predicts observed finite-difference state derivatives and HJB residual is reduced during training",
      "observed_derivative_rms": 0.12187617272138596,
      "observed_hjb_residual_rms": 0.5597086548805237,
      "baseline_test_mse": 0.0006821716669946909,
      "idea_test_mse": 0.0006843233713880181,
      "confirmed": true
    }
  },
  "idea_config": {
    "lr": 0.01,
    "lambda_hjb": 0.001
  }
}