{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "cycle_weight": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "cycle_weight": 0.0
        },
        "mean": 0.001864717691205442
      },
      {
        "cfg": {
          "lr": 0.003,
          "cycle_weight": 0.0
        },
        "mean": 0.0010901302302954718
      },
      {
        "cfg": {
          "lr": 0.01,
          "cycle_weight": 0.0
        },
        "mean": 0.0006186682076076977
      }
    ],
    "full": {
      "mean": 0.00054517226817552,
      "std": 0.00020361842727135583,
      "per_seed": [
        0.0004930912400595844,
        0.0004765945195686072,
        0.0004673648509196937,
        0.0010376222198829055,
        0.000511663849465549,
        0.00027233464061282575,
        0.0005571767687797546,
        0.0005455300561152399
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.00045470128861779813,
    "std": 0.00014832940019044596,
    "per_seed": [
      0.0004892470315098763,
      0.0003560753830242902,
      0.0005445053684525192,
      0.0007670727791264653,
      0.00048597707063890994,
      0.00024241428764071316,
      0.0003992895071860403,
      0.0003530288813635707
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": -9.047097955772188e-05,
    "idea_wins": 7,
    "n_pairs": 8,
    "per_seed_diffs": [
      -3.844208549708128e-06,
      -0.00012051913654431701,
      7.714051753282547e-05,
      -0.00027054944075644016,
      -2.5686778826639056e-05,
      -2.9920352972112596e-05,
      -0.00015788726159371436,
      -0.00019250117475166917
    ],
    "p_value": 0.07265,
    "mde": 9.572692629336764e-05,
    "mde_rel_pct": 17.559023428269473,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "cfg": {
      "lr": 0.01,
      "cycle_weight": 0.03
    },
    "behavior": {
      "prediction": "small input perturbations preserve decoded joint state",
      "predicted": 1.0,
      "observed": 1.0,
      "tolerance": 0.1,
      "confirmed": true,
      "n_models": 8
    }
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.01,
        "cycle_weight": 0.03
      },
      "mean": 0.00045470128861779813,
      "std": 0.00014832940019044596,
      "per_seed": [
        0.0004892470315098763,
        0.0003560753830242902,
        0.0005445053684525192,
        0.0007670727791264653,
        0.00048597707063890994,
        0.00024241428764071316,
        0.0003992895071860403,
        0.0003530288813635707
      ]
    },
    {
      "cfg": {
        "lr": 0.001,
        "cycle_weight": 0.03
      },
      "mean": 0.0023312663834076375,
      "std": 0.0007589181455697688,
      "per_seed": [
        0.0011521552223712206,
        0.002354319440200925,
        0.0021086849737912416,
        0.002465922152623534,
        0.0014372181612998247,
        0.0023133959621191025,
        0.0035222345031797886,
        0.0032962006516754627
      ]
    },
    {
      "cfg": {
        "lr": 0.01,
        "cycle_weight": 0.03
      },
      "mean": 0.00045470128861779813,
      "std": 0.00014832940019044596,
      "per_seed": [
        0.0004892470315098763,
        0.0003560753830242902,
        0.0005445053684525192,
        0.0007670727791264653,
        0.00048597707063890994,
        0.00024241428764071316,
        0.0003992895071860403,
        0.0003530288813635707
      ]
    }
  ],
  "notes": "Matched dynamics task; baseline canonical train_model, idea same GRU backbone with differentiable factor-state regularizer."
}