{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "weight_decay": 0.0001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "mean": 0.17645552940666676
      },
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0001
        },
        "mean": 0.17659775167703629
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 0.028942419216036797
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001
        },
        "mean": 0.028870720881968737
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0
        },
        "mean": 0.013678546762093902
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0001
        },
        "mean": 0.013676861766725779
      }
    ],
    "full": {
      "mean": 0.013490461744368076,
      "std": 0.0036455379141356633,
      "per_seed": [
        0.01656705141067505,
        0.01134894136339426,
        0.013825791887938976,
        0.012965662404894829,
        0.02068343572318554,
        0.014116058126091957,
        0.010448829270899296,
        0.007967923767864704
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.013573068077675998,
    "std": 0.0036332788661090037,
    "per_seed": [
      0.016753576695919037,
      0.011368772946298122,
      0.013876390643417835,
      0.012907849624752998,
      0.020729539915919304,
      0.014167544431984425,
      0.010720721445977688,
      0.008060148917138577
    ],
    "n": 8,
    "cfg": {
      "lr": 0.006,
      "weight_decay": 0.0,
      "lam": 0.001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0,
          "lam": 0.001
        },
        "mean": 0.16410381672903895,
        "per_seed": [
          0.08069120347499847,
          0.20044010877609253,
          0.1486465185880661,
          0.2762965261936188,
          0.060551416128873825,
          0.1775653064250946,
          0.17477571964263916,
          0.19386373460292816
        ]
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0,
          "lam": 0.001
        },
        "mean": 0.02484322735108435,
        "per_seed": [
          0.03566475957632065,
          0.02596493996679783,
          0.02446393296122551,
          0.02981584146618843,
          0.026317916810512543,
          0.01899126172065735,
          0.01727476716041565,
          0.020252399146556854
        ]
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0,
          "lam": 0.001
        },
        "mean": 0.013573068077675998,
        "per_seed": [
          0.016753576695919037,
          0.011368772946298122,
          0.013876390643417835,
          0.012907849624752998,
          0.020729539915919304,
          0.014167544431984425,
          0.010720721445977688,
          0.008060148917138577
        ]
      }
    ]
  },
  "comparison": {
    "delta_mean": 8.260633330792189e-05,
    "idea_wins": 1,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.00018652528524398804,
      1.9831582903862e-05,
      5.059875547885895e-05,
      -5.7812780141830444e-05,
      4.610419273376465e-05,
      5.14863058924675e-05,
      0.00027189217507839203,
      9.222514927387238e-05
    ],
    "p_value": 0.04575,
    "mde": 8.579928619368267e-05,
    "mde_rel_pct": 0.635999625657748,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "protocol_note": "Baseline and idea use identical rnn_small systems; only latent diffusion penalty differs.",
  "idea_grid": [
    {
      "cfg": {
        "lr": 0.001,
        "weight_decay": 0.0,
        "lam": 0.001
      },
      "mean": 0.16410381672903895,
      "per_seed": [
        0.08069120347499847,
        0.20044010877609253,
        0.1486465185880661,
        0.2762965261936188,
        0.060551416128873825,
        0.1775653064250946,
        0.17477571964263916,
        0.19386373460292816
      ]
    },
    {
      "cfg": {
        "lr": 0.003,
        "weight_decay": 0.0,
        "lam": 0.001
      },
      "mean": 0.02484322735108435,
      "per_seed": [
        0.03566475957632065,
        0.02596493996679783,
        0.02446393296122551,
        0.02981584146618843,
        0.026317916810512543,
        0.01899126172065735,
        0.01727476716041565,
        0.020252399146556854
      ]
    },
    {
      "cfg": {
        "lr": 0.006,
        "weight_decay": 0.0,
        "lam": 0.001
      },
      "mean": 0.013573068077675998,
      "per_seed": [
        0.016753576695919037,
        0.011368772946298122,
        0.013876390643417835,
        0.012907849624752998,
        0.020729539915919304,
        0.014167544431984425,
        0.010720721445977688,
        0.008060148917138577
      ]
    }
  ],
  "mechanism_signature": {
    "predicted": "hidden cross diffusion can vary while observed output power is comparatively stable",
    "baseline_observed_output_power_std": 0.15846109952602266,
    "idea_observed_output_power_std": 0.15854744148375174,
    "baseline_hidden_dxy_std": 0.00042474186371430176,
    "idea_hidden_dxy_std": 2.490041131097002e-05,
    "idea_mean_reverse_forward_gap": 0.018812500281407442,
    "confirmed": false
  }
}