{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "wd": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "wd": 0.0
        },
        "mean": 0.00032828053372213617
      },
      {
        "cfg": {
          "lr": 0.0015,
          "wd": 0.0001
        },
        "mean": 0.0004147848885622807
      },
      {
        "cfg": {
          "lr": 0.003,
          "wd": 0.0
        },
        "mean": 4.9621185098658316e-05
      },
      {
        "cfg": {
          "lr": 0.003,
          "wd": 0.0001
        },
        "mean": 0.00020817978656850755
      },
      {
        "cfg": {
          "lr": 0.006,
          "wd": 0.0
        },
        "mean": 2.345215943932999e-05
      },
      {
        "cfg": {
          "lr": 0.006,
          "wd": 0.0001
        },
        "mean": 0.00010142060364159988
      }
    ],
    "full": {
      "mean": 2.282289460708853e-05,
      "std": 2.305330446600017e-06,
      "per_seed": [
        2.4922346710809506e-05,
        2.6788266040966846e-05,
        1.934224746946711e-05,
        2.2755777536076494e-05,
        2.3468126528314315e-05,
        2.385292282269802e-05,
        2.0721481632790528e-05,
        2.073198811558541e-05
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 4.757533133670222e-05,
    "std": 2.1643697598220244e-05,
    "per_seed": [
      9.76939991232939e-05,
      3.8501457311213017e-05,
      3.3765736588975415e-05,
      2.9315116989891976e-05,
      6.082628169679083e-05,
      5.370224607759155e-05,
      3.0736955523025244e-05,
      3.606085738283582e-05
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 2.475243672961369e-05,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      7.27716524124844e-05,
      1.171319127024617e-05,
      1.4423489119508304e-05,
      6.5593394538154826e-06,
      3.7358155168476515e-05,
      2.9849323254893534e-05,
      1.0015473890234716e-05,
      1.532886926725041e-05
    ],
    "p_value": 0.0081,
    "mde": 1.8439870679175355e-05,
    "mde_rel_pct": 80.79549503527105,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "signature": {
      "prediction": "idea lowers trained recurrent cross-block q and keeps it near <=0.8",
      "baseline_q_mean": 2.9282636810198936,
      "idea_q_mean": 0.7976552316857172,
      "baseline_q_per_seed": [
        2.7491974371903396,
        2.8032834877024646,
        2.7482760381642044,
        3.334465893518113,
        2.8231071925105216,
        3.0133113893982655,
        2.993095970408364,
        2.9613720392668768
      ],
      "idea_q_per_seed": [
        0.7953895769092711,
        0.7993655229735372,
        0.7991708584055264,
        0.79743240619884,
        0.7974510193922437,
        0.798977436754247,
        0.7961304629450436,
        0.7973245699070284
      ],
      "observed_reduction": 2.1306084493341766,
      "target_q": 0.8,
      "confirmed": true
    },
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.006,
          "lam": 0.1
        },
        "result": {
          "mean": 4.757533133670222e-05,
          "std": 2.1643697598220244e-05,
          "per_seed": [
            9.76939991232939e-05,
            3.8501457311213017e-05,
            3.3765736588975415e-05,
            2.9315116989891976e-05,
            6.082628169679083e-05,
            5.370224607759155e-05,
            3.0736955523025244e-05,
            3.606085738283582e-05
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.0015,
          "lam": 0.1
        },
        "result": {
          "mean": 0.00037327232348616235,
          "std": 3.6520547820565044e-05,
          "per_seed": [
            0.0003330762847326696,
            0.00043403543531894684,
            0.00033968265051953495,
            0.000401037948904559,
            0.00037536423769779503,
            0.0003695053164847195,
            0.0003257567877881229,
            0.00040771992644295096
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.006,
          "lam": 0.1
        },
        "result": {
          "mean": 4.757533133670222e-05,
          "std": 2.1643697598220244e-05,
          "per_seed": [
            9.76939991232939e-05,
            3.8501457311213017e-05,
            3.3765736588975415e-05,
            2.9315116989891976e-05,
            6.082628169679083e-05,
            5.370224607759155e-05,
            3.0736955523025244e-05,
            3.606085738283582e-05
          ],
          "n": 8
        }
      }
    ],
    "bdd_delta": 0.2
  },
  "protocol_notes": "Dynamics is structurally matched: controlled pendulum rollout and recurrent state coupling. Baseline and idea share rnn_small, data, Adam, epochs, batch, lr/weight-decay grids."
}