{
  "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": 6.268094512051903e-05
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 2.4053396373346914e-05
      },
      {
        "cfg": {
          "lr": 0.006
        },
        "mean": 1.9922963019780582e-05
      }
    ],
    "full": {
      "mean": 1.6942617776294355e-05,
      "std": 7.0294610370873945e-06,
      "per_seed": [
        2.646726898092311e-05,
        1.3497274267137982e-05,
        1.7921185644809157e-05,
        2.180612318625208e-05,
        1.1675516361719929e-05,
        9.150396181212272e-06,
        2.697196396184154e-05,
        8.051213626458775e-06
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 6.218317793127426e-06,
    "std": 2.61984541800953e-06,
    "per_seed": [
      2.8394540549925296e-06,
      8.368528142455034e-06,
      4.065665507368976e-06,
      6.483904144261032e-06,
      2.6967427402269095e-06,
      9.669754945207387e-06,
      9.356464033771772e-06,
      6.266028776735766e-06
    ],
    "n": 8,
    "chosen_cfg": {
      "lr": 0.006,
      "tmax": 8
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "tmax": 8
        },
        "mean": 3.799157411776832e-05,
        "per_seed": [
          2.9381331842159852e-05,
          6.0935002693440765e-05,
          3.8257920095929876e-05,
          2.33920418395428e-05
        ]
      },
      {
        "cfg": {
          "lr": 0.003,
          "tmax": 8
        },
        "mean": 1.020754018554726e-05,
        "per_seed": [
          6.3169459281198215e-06,
          1.2884396710433066e-05,
          1.2727699868264608e-05,
          8.901118235371541e-06
        ]
      },
      {
        "cfg": {
          "lr": 0.006,
          "tmax": 8
        },
        "mean": 5.439387962269393e-06,
        "per_seed": [
          2.8394540549925296e-06,
          8.368528142455034e-06,
          4.065665507368976e-06,
          6.483904144261032e-06
        ]
      }
    ]
  },
  "comparison": {
    "delta_mean": -1.072429998316693e-05,
    "idea_wins": 7,
    "n_pairs": 8,
    "per_seed_diffs": [
      -2.362781492593058e-05,
      -5.128746124682948e-06,
      -1.385552013744018e-05,
      -1.5322219041991048e-05,
      -8.97877362149302e-06,
      5.193587639951147e-07,
      -1.7615499928069767e-05,
      -1.7851848497230094e-06
    ],
    "p_value": 0.0163,
    "mde": 6.964423506000914e-06,
    "mde_rel_pct": 41.105947132593315,
    "verdict": "idea better (significant)",
    "system_worked": true
  },
  "mechanism_signature": {
    "prediction": "shared adaptive recurrence performs fewer than Tmax weighted updates on trained inputs",
    "predicted": {
      "weighted_steps": "< 8"
    },
    "observed": {
      "mean_weighted_steps": 1.24119733273983,
      "mean_mass": 1.0,
      "per_seed_steps": [
        1.2474790811538696,
        1.2404111623764038,
        1.2472835779190063,
        1.2294979095458984,
        1.252071499824524,
        1.2432061433792114,
        1.2319114208221436,
        1.237717866897583
      ]
    },
    "confirmed": true,
    "parameter_counts": {
      "baseline": 135297,
      "idea": 18434
    }
  },
  "corrected_protocol": true
}