1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.07048395439051092
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0036796721105929464
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.0020796392345800996
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0018650785204954445,
 33      "per_seed": [
 34        0.0020393452141433954,
 35        0.0022905797231942415,
 36        0.0019529510755091906,
 37        0.002035680925473571,
 38        0.0009734529303386807,
 39        0.003012745874002576,
 40        0.0011985950404778123,
 41        0.001417277380824089
 42      ]
 43    }
 44  },
 45  "idea": {
 46    "mean": 0.07028227765113115,
 47    "per_seed": [
 48      0.09344271570444107,
 49      0.035067640244960785,
 50      0.06569516658782959,
 51      0.06504810601472855,
 52      0.057421158999204636,
 53      0.045870814472436905,
 54      0.06615367531776428,
 55      0.1335589438676834
 56    ]
 57  },
 58  "comparison": {
 59    "delta_mean": 0.06841719913063571,
 60    "idea_wins": 0,
 61    "n_pairs": 8,
 62    "per_seed_diffs": [
 63      0.09140337049029768,
 64      0.03277706052176654,
 65      0.0637422155123204,
 66      0.06301242508925498,
 67      0.056447706068865955,
 68      0.04285806859843433,
 69      0.06495508027728647,
 70      0.13214166648685932
 71    ],
 72    "p_value": 0.0081,
 73    "mde": 0.02590531560040504,
 74    "mde_rel_pct": 1388.9664867044569,
 75    "verdict": "idea worse (significant)",
 76    "system_worked": false
 77  },
 78  "idea_sweep": [
 79    {
 80      "cfg": {
 81        "lr": 0.001
 82      },
 83      "mean": 0.6029151976108551
 84    },
 85    {
 86      "cfg": {
 87        "lr": 0.003
 88      },
 89      "mean": 0.2946254089474678
 90    },
 91    {
 92      "cfg": {
 93        "lr": 0.01
 94      },
 95      "mean": 0.06481340713799
 96    }
 97  ],
 98  "architecture": "matched 64-unit recurrent state and scalar head; only deterministic leaky update vs correlated IF differs",
 99  "mechanism_signature": {
100    "predicted_shared_covariance_factor": 0.42250000000000004,
101    "observed_shared_noise_correlation_mean": 0.42043366931669834,
102    "observed_shared_noise_correlation_per_seed": [
103      0.4126678738329091,
104      0.4521553626790912,
105      0.41302410068546774,
106      0.40544973371858345,
107      0.4157424909414776,
108      0.44676270990572226,
109      0.4155898912561236,
110      0.40207719151421156
111    ],
112    "spike_rate_mean": 0.03861450147815049,
113    "confirmed": true
114  }
115}