{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "mean": 0.0720269184675999
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 0.0028564660606207326
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0
        },
        "mean": 0.00228094776684884
      }
    ],
    "full": {
      "mean": 0.00228094776684884,
      "std": 0.0008615682293713639,
      "per_seed": [
        0.002221147296950221,
        0.001970731420442462,
        0.0013024760410189629,
        0.0042177895084023476,
        0.0020735678263008595,
        0.0015317687066271901,
        0.002943465718999505,
        0.0019866356160491705
      ],
      "n": 8
    }
  },
  "idea": {
    "cfg": {
      "lr": 0.006,
      "weight_decay": 0.0,
      "kappa": 8.0
    },
    "mean": 0.002070821574307047,
    "std": 0.0007668985617216761,
    "per_seed": [
      0.0024689047131687403,
      0.0014718936290591955,
      0.0017854231409728527,
      0.0037137719336897135,
      0.0013458451721817255,
      0.0012669704155996442,
      0.0025451004039496183,
      0.0019686631858348846
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": -0.000210126192541793,
    "delta_std": 0.0003877800102460661,
    "p_value": 0.1702,
    "diffs": [
      0.0002477574162185192,
      -0.0004988377913832664,
      0.00048294709995388985,
      -0.0005040175747126341,
      -0.000727722654119134,
      -0.00026479829102754593,
      -0.0003983653150498867,
      -1.797243021428585e-05
    ]
  },
  "mechanism_signature": {
    "math_check": {
      "kappas": [
        1,
        2,
        4,
        8,
        16
      ],
      "delays": [
        0.9989999999999988,
        0.4989999999999988,
        0.24899999999999878,
        0.12399999999999878,
        0.062000000000001165
      ],
      "predicted_offset_slope": 1.0,
      "fit_slope": 0.9997204301075259,
      "r2": 0.9999997339246215,
      "passed": true
    },
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0,
          "kappa": 2.0
        },
        "mean": 0.002092966591590084,
        "std": 0.0007941076298383792,
        "per_seed": [
          0.0025941352359950542,
          0.0013959517236799002,
          0.0019340378930792212,
          0.0037612225860357285,
          0.0012745162239298224,
          0.0012861500727012753,
          0.0025039087049663067,
          0.0019938102923333645
        ],
        "n": 8
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0,
          "kappa": 4.0
        },
        "mean": 0.0020771025301655754,
        "std": 0.0007748204332654841,
        "per_seed": [
          0.002521086484193802,
          0.0014459786470979452,
          0.0018341594841331244,
          0.003721740795299411,
          0.0013138779904693365,
          0.001272903406061232,
          0.0025291924830526114,
          0.0019778809510171413
        ],
        "n": 8
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0,
          "kappa": 8.0
        },
        "mean": 0.002070821574307047,
        "std": 0.0007668985617216761,
        "per_seed": [
          0.0024689047131687403,
          0.0014718936290591955,
          0.0017854231409728527,
          0.0037137719336897135,
          0.0013458451721817255,
          0.0012669704155996442,
          0.0025451004039496183,
          0.0019686631858348846
        ],
        "n": 8
      }
    ],
    "comparison_recomputed": {
      "delta_mean": -0.000210126192541793,
      "delta_std": 0.0003877800102460661,
      "p_value": 0.1702,
      "diffs": [
        0.0002477574162185192,
        -0.0004988377913832664,
        0.00048294709995388985,
        -0.0005040175747126341,
        -0.000727722654119134,
        -0.00026479829102754593,
        -0.0003983653150498867,
        -1.797243021428585e-05
      ]
    },
    "mechanism_signature": {
      "prediction": "lag delay scales as 1/kappa and positive feed-forward compensation reduces it",
      "trained_model_measurements": [
        {
          "kappa": 2.0,
          "baseline_cross_step": 1.0,
          "idea_cross_step": 1.0,
          "observed_mean_effective_gap": 0.0
        },
        {
          "kappa": 4.0,
          "baseline_cross_step": 1.0,
          "idea_cross_step": 1.0,
          "observed_mean_effective_gap": 0.0
        },
        {
          "kappa": 8.0,
          "baseline_cross_step": 1.0,
          "idea_cross_step": 1.0,
          "observed_mean_effective_gap": 0.0
        }
      ],
      "confirmed": false,
      "reason": "No reliable local-growth threshold crossing was observed consistently in the trained dynamics runs; therefore the NN-scale quantitative claim is not confirmed."
    }
  }
}