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": 0.0021690295106964186
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.001074885185516905
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.00054517226817552
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.00054517226817552,
 33      "std": 0.00020361842727135583,
 34      "per_seed": [
 35        0.0004930912400595844,
 36        0.0004765945195686072,
 37        0.0004673648509196937,
 38        0.0010376222198829055,
 39        0.000511663849465549,
 40        0.00027233464061282575,
 41        0.0005571767687797546,
 42        0.0005455300561152399
 43      ],
 44      "n": 8
 45    },
 46    "protocol_note": "8 paired seeds; sweep grid shared with idea; baseline is canonical rnn_small."
 47  },
 48  "idea": {
 49    "mean": 0.0016794730763649568,
 50    "std": 0.0005899470001703297,
 51    "per_seed": [
 52      0.00201487448066473,
 53      0.0009998989989981055,
 54      0.0017773952567949891,
 55      0.0030421584378927946,
 56      0.001546183368191123,
 57      0.001343107083812356,
 58      0.0013974033063277602,
 59      0.0013147636782377958
 60    ],
 61    "n": 8,
 62    "cfg": {
 63      "lr": 0.003
 64    }
 65  },
 66  "comparison": {
 67    "delta_mean": 0.0011343008081894368,
 68    "idea_wins": 0,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      0.0015217832406051457,
 72      0.0005233044794294983,
 73      0.0013100304058752954,
 74      0.002004536218009889,
 75      0.001034519518725574,
 76      0.0010707724431995302,
 77      0.0008402265375480056,
 78      0.000769233622122556
 79    ],
 80    "p_value": 0.0081,
 81    "mde": 0.0003928773272119451,
 82    "mde_rel_pct": 72.06480412636415,
 83    "verdict": "idea worse (significant)",
 84    "system_worked": false
 85  },
 86  "method_note": "Braid transport replaces the GRU recurrence; same dynamics data, epochs, batch, and lr grid.",
 87  "mechanism_signature": {
 88    "predicted": "skew residual near zero and transport-only norm drift near zero",
 89    "observed_mean_skew_residual": 0.0,
 90    "observed_mean_transport_norm_drift": 7.996568260337789e-08,
 91    "confirmed": true
 92  },
 93  "idea_sweep": [
 94    {
 95      "cfg": {
 96        "lr": 0.001
 97      },
 98      "mean": 0.002948012319393456,
 99      "per_seed": [
100        0.003085256787016988,
101        0.002189615275710821,
102        0.0050039333291351795,
103        0.0036421462427824736,
104        0.0022776462137699127,
105        0.001872690161690116,
106        0.0024267486296594143,
107        0.003086061915382743
108      ]
109    },
110    {
111      "cfg": {
112        "lr": 0.003
113      },
114      "mean": 0.0016794730763649568,
115      "per_seed": [
116        0.00201487448066473,
117        0.0009998989989981055,
118        0.0017773952567949891,
119        0.0030421584378927946,
120        0.001546183368191123,
121        0.001343107083812356,
122        0.0013974033063277602,
123        0.0013147636782377958
124      ]
125    },
126    {
127      "cfg": {
128        "lr": 0.01
129      },
130      "mean": 0.0018830011467798613,
131      "per_seed": [
132        0.0020360236521810293,
133        0.0009326969156973064,
134        0.00206544017419219,
135        0.0026314484421163797,
136        0.0011524695437401533,
137        0.0020754176657646894,
138        0.001974960323423147,
139        0.002195552457123995
140      ]
141    }
142  ]
143}