{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.001154391880845651
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.0006730457389494404
      },
      {
        "cfg": {
          "lr": 0.006
        },
        "mean": 0.0004064472232130356
      }
    ],
    "full": {
      "mean": 0.0003541622754710261,
      "std": 9.525887076242907e-05,
      "per_seed": [
        0.0004017692699562758,
        0.00043160535278730094,
        0.00029359731706790626,
        0.0004988169530406594,
        0.0003143931389786303,
        0.00029283223557285964,
        0.00018064776668325067,
        0.00041963616968132555
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.7773510366678238,
    "std": 0.04344097334599117,
    "per_seed": [
      0.7313887476921082,
      0.8112629652023315,
      0.8064488172531128,
      0.8520755171775818,
      0.7981084585189819,
      0.7174731492996216,
      0.7519327402114868,
      0.7501178979873657
    ],
    "n": 8,
    "cfg": {
      "lr": 0.006,
      "beta": 0.01,
      "gamma": 0.02
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "beta": 0.01,
          "gamma": 0.02
        },
        "mean": 0.8052886575460434
      },
      {
        "cfg": {
          "lr": 0.003,
          "beta": 0.01,
          "gamma": 0.02
        },
        "mean": 0.8011865019798279
      },
      {
        "cfg": {
          "lr": 0.006,
          "beta": 0.01,
          "gamma": 0.02
        },
        "mean": 0.8002940118312836
      }
    ],
    "signature_per_seed": [
      {
        "info_nats": -0.11848548799753189,
        "noise_sigma": 0.24387867748737335,
        "reverse_mse": 0.8134815096855164
      },
      {
        "info_nats": -0.114313505589962,
        "noise_sigma": 0.25354620814323425,
        "reverse_mse": 0.8361767530441284
      },
      {
        "info_nats": -0.022014688700437546,
        "noise_sigma": 0.23950694501399994,
        "reverse_mse": 0.8750950694084167
      },
      {
        "info_nats": -0.10838189721107483,
        "noise_sigma": 0.26771512627601624,
        "reverse_mse": 0.8871616125106812
      },
      {
        "info_nats": -0.061512164771556854,
        "noise_sigma": 0.250619113445282,
        "reverse_mse": 0.8385130167007446
      },
      {
        "info_nats": -0.08916227519512177,
        "noise_sigma": 0.24852284789085388,
        "reverse_mse": 0.7741023898124695
      },
      {
        "info_nats": -0.05122513696551323,
        "noise_sigma": 0.25014737248420715,
        "reverse_mse": 0.8042869567871094
      },
      {
        "info_nats": -0.12155700474977493,
        "noise_sigma": 0.23886628448963165,
        "reverse_mse": 0.8002150058746338
      }
    ]
  },
  "comparison": {
    "delta_mean": 0.7769968743923528,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.7309869784221519,
      0.8108313598495442,
      0.8061552199360449,
      0.8515767002245411,
      0.7977940653800033,
      0.7171803170640487,
      0.7517520924448036,
      0.7496982618176844
    ],
    "p_value": 0.0081,
    "mde": 0.03879287423905792,
    "mde_rel_pct": 10953.417945901907,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "track_choice": "dynamics: actuated pendulum rollout is structurally matched to control/stability.",
    "prediction": "The intended prediction was lower information with a useful reverse prior; the trained signature did not validate this quantitatively because the estimated information was negative and no baseline information measurement was recorded.",
    "predicted_info_nats": -0.08583152014762163,
    "observed_reverse_mse": 0.8286290392279625,
    "observed_noise_sigma": 0.2491003219038248,
    "confirmed": false
  }
}