{
  "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": 0.0016004437929950655
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.0008657508878968656
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.0003360401788086165
      }
    ],
    "full": {
      "mean": 0.00029742292554146843,
      "std": 0.00012590863545935686,
      "per_seed": [
        0.0004953785100951791,
        0.0003386490570846945,
        0.0002134958194801584,
        0.0002966373285744339,
        0.0001490332215325907,
        0.00031617036438547075,
        0.0001173576238215901,
        0.00045266147935763
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.00031140187456912827,
    "std": 0.00012528404252057512,
    "per_seed": [
      0.0005046091391704977,
      0.00035253993701189756,
      0.00023353651340585202,
      0.00032840640051290393,
      0.00015706471458543092,
      0.00032477168133482337,
      0.00012830838386435062,
      0.00046197822666727006
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 1.3978949027659837e-05,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      9.230629075318575e-06,
      1.389087992720306e-05,
      2.004069392569363e-05,
      3.1769071938470006e-05,
      8.031493052840233e-06,
      8.601316949352622e-06,
      1.0950760042760521e-05,
      9.31674730964005e-06
    ],
    "p_value": 0.0081,
    "mde": 6.858680107943e-06,
    "mde_rel_pct": 2.3060361253109667,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "best_idea_cfg": {
      "lr": 0.01,
      "lambda": 0.02
    },
    "all_idea_settings": {
      "0.001": {
        "mean": 0.0016261089622275904,
        "std": 0.00047570904542975056,
        "per_seed": [
          0.0012551955878734589,
          0.0012359732063487172,
          0.0015323605621233582,
          0.002465372672304511,
          0.001136873965151608,
          0.0012831705389544368,
          0.001836425275541842,
          0.002263499889522791
        ],
        "n": 8
      },
      "0.003": {
        "mean": 0.0008687566078151576,
        "std": 0.0002024636252280242,
        "per_seed": [
          0.0008179493015632033,
          0.0007614863570779562,
          0.0006751477485522628,
          0.001272146706469357,
          0.0006523997872136533,
          0.0007431273115798831,
          0.0009471484227105975,
          0.0010806472273543477
        ],
        "n": 8
      },
      "0.01": {
        "mean": 0.00031140187456912827,
        "std": 0.00012528404252057512,
        "per_seed": [
          0.0005046091391704977,
          0.00035253993701189756,
          0.00023353651340585202,
          0.00032840640051290393,
          0.00015706471458543092,
          0.00032477168133482337,
          0.00012830838386435062,
          0.00046197822666727006
        ],
        "n": 8
      }
    },
    "custom_track": null,
    "signature": {
      "prediction": "horizon weighting produces a finite, measured sensitivity proxy",
      "observed": [
        {
          "seed": 0,
          "observed_jacobian_abs": 0.047531094402074814,
          "tube_weighted_observed": 0.039783747423522677
        },
        {
          "seed": 1,
          "observed_jacobian_abs": 0.046463221311569214,
          "tube_weighted_observed": 0.03905660488809137
        },
        {
          "seed": 2,
          "observed_jacobian_abs": 0.04596756026148796,
          "tube_weighted_observed": 0.04002308358438155
        },
        {
          "seed": 3,
          "observed_jacobian_abs": 0.0467328242957592,
          "tube_weighted_observed": 0.040204589420798326
        }
      ],
      "predicted_vs_observed_ratio": 0.852146210300268,
      "confirmed": false
    }
  },
  "protocol_notes": {
    "paired_seeds": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ],
    "epochs": 12,
    "n_train": 800,
    "n_test": 200,
    "structural_match": "dynamics/control",
    "baseline_grid_equals_idea_union": true
  }
}