{
  "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": 0.0015
        },
        "mean": 0.33852236717939377
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.024998844717629254
      },
      {
        "cfg": {
          "lr": 0.006
        },
        "mean": 0.0168350946623832
      }
    ],
    "full": {
      "mean": 0.019937668344937265,
      "std": 0.013005039202094617,
      "per_seed": [
        0.03213806077837944,
        0.009990346617996693,
        0.010487839579582214,
        0.014724131673574448,
        0.04623213782906532,
        0.006986701861023903,
        0.011433426290750504,
        0.027508702129125595
      ],
      "n": 8
    }
  },
  "idea": {
    "per_seed": [
      0.032140087336301804,
      0.010026148520410061,
      0.010492031462490559,
      0.01471502985805273,
      0.04622451215982437,
      0.006962496321648359,
      0.011483874171972275,
      0.02751341462135315
    ],
    "mean": 0.019944699306506664,
    "std": 0.013896786279171906
  },
  "comparison": {
    "delta_mean": 7.030961569398642e-06,
    "idea_wins": 3,
    "n_pairs": 8,
    "per_seed_diffs": [
      2.0265579223632812e-06,
      3.5801902413368225e-05,
      4.191882908344269e-06,
      -9.101815521717072e-06,
      -7.625669240951538e-06,
      -2.4205539375543594e-05,
      5.044788122177124e-05,
      4.712492227554321e-06
    ],
    "p_value": 0.45785,
    "mde": 2.0486566875431273e-05,
    "mde_rel_pct": 0.10275307283177568,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "mechanism_signature": {
    "prediction": "For T=1, m >= n/2 is the counting threshold for full rank.",
    "baseline_trained_model": {
      "state_dim": 64,
      "horizon_T": 1,
      "predicted_counting_threshold_m": 32,
      "observed": [
        {
          "m": 8,
          "predicted_rank_upper_bound_T1": 16,
          "observed_rank_T1": 16,
          "observed_smin": 0.03744511306285858,
          "observed_logdet": -383.529296875
        },
        {
          "m": 16,
          "predicted_rank_upper_bound_T1": 32,
          "observed_rank_T1": 32,
          "observed_smin": 0.028290288522839546,
          "observed_logdet": -327.2057189941406
        },
        {
          "m": 32,
          "predicted_rank_upper_bound_T1": 64,
          "observed_rank_T1": 64,
          "observed_smin": 4.912145959679037e-05,
          "observed_logdet": -227.1643829345703
        }
      ],
      "confirmed": true
    },
    "idea_trained_model": {
      "state_dim": 64,
      "horizon_T": 1,
      "predicted_counting_threshold_m": 32,
      "observed": [
        {
          "m": 8,
          "predicted_rank_upper_bound_T1": 16,
          "observed_rank_T1": 16,
          "observed_smin": 0.04268581420183182,
          "observed_logdet": -381.6462097167969
        },
        {
          "m": 16,
          "predicted_rank_upper_bound_T1": 32,
          "observed_rank_T1": 32,
          "observed_smin": 0.02824431285262108,
          "observed_logdet": -325.7181701660156
        },
        {
          "m": 32,
          "predicted_rank_upper_bound_T1": 64,
          "observed_rank_T1": 64,
          "observed_smin": 0.0018923624884337187,
          "observed_logdet": -225.09347534179688
        }
      ],
      "confirmed": true
    }
  },
  "idea_sweep": {
    "0.0015": {
      "per_seed": [
        0.2595526874065399,
        0.3410550355911255,
        0.31123819947242737,
        0.4425976574420929,
        0.19178999960422516,
        0.34441572427749634,
        0.3199017643928528,
        0.3680892884731293
      ],
      "mean": 0.32233004458248615,
      "std": 0.07416895300576867
    },
    "0.003": {
      "per_seed": [
        0.03490198403596878,
        0.014994751662015915,
        0.0031942580826580524,
        0.046722445636987686,
        0.06990749388933182,
        0.01345915999263525,
        0.008423914201557636,
        0.06726980954408646
      ],
      "mean": 0.0323592271306552,
      "std": 0.02652120695991089
    },
    "0.006": {
      "per_seed": [
        0.032140087336301804,
        0.010026148520410061,
        0.010492031462490559,
        0.01471502985805273,
        0.04622451215982437,
        0.006962496321648359,
        0.011483874171972275,
        0.02751341462135315
      ],
      "mean": 0.019944699306506664,
      "std": 0.013896786279171906
    }
  },
  "protocol_notes": "Baseline sweep uses the same three learning rates as the idea; idea adds only finite-horizon logdet loss."
}