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": 0.006,
 10      "alpha": 0.5
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.0015,
 16          "alpha": 0.5
 17        },
 18        "mean": 0.07130480092018843
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.0015,
 23          "alpha": 0.7
 24        },
 25        "mean": 0.07130480092018843
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.0015,
 30          "alpha": 1.0
 31        },
 32        "mean": 0.07130480092018843
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "alpha": 0.5
 38        },
 39        "mean": 0.02189101977273822
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "alpha": 0.7
 45        },
 46        "mean": 0.02189101977273822
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "alpha": 1.0
 52        },
 53        "mean": 0.02189101977273822
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.006,
 58          "alpha": 0.5
 59        },
 60        "mean": 0.009577582241035998
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.006,
 65          "alpha": 0.7
 66        },
 67        "mean": 0.009577582241035998
 68      },
 69      {
 70        "cfg": {
 71          "lr": 0.006,
 72          "alpha": 1.0
 73        },
 74        "mean": 0.009577582241035998
 75      }
 76    ],
 77    "full": {
 78      "mean": 0.006650938361417502,
 79      "std": 0.0034903632092782065,
 80      "per_seed": [
 81        0.011823982931673527,
 82        0.005238679703325033,
 83        0.011270472779870033,
 84        0.009977193549275398,
 85        0.0038828086107969284,
 86        0.004712723195552826,
 87        0.0035874645691365004,
 88        0.002714181551709771
 89      ],
 90      "n": 8
 91    }
 92  },
 93  "idea": {
 94    "mean": 0.00461027942947112,
 95    "std": 0.0017827839604746276,
 96    "per_seed": [
 97      0.00657200301066041,
 98      0.0035921495873481035,
 99      0.005795476958155632,
100      0.0077093965373933315,
101      0.002257069107145071,
102      0.0037548672407865524,
103      0.0043830350041389465,
104      0.002818237990140915
105    ],
106    "n": 8
107  },
108  "comparison": {
109    "delta_mean": -0.002040658931946382,
110    "idea_wins": 6,
111    "n_pairs": 8,
112    "per_seed_diffs": [
113      -0.005251979921013117,
114      -0.0016465301159769297,
115      -0.005474995821714401,
116      -0.0022677970118820667,
117      -0.0016257395036518574,
118      -0.0009578559547662735,
119      0.0007955704350024462,
120      0.00010405643843114376
121    ],
122    "p_value": 0.0317,
123    "mde": 0.001904176545133376,
124    "mde_rel_pct": 28.630193841212243,
125    "verdict": "idea better (significant)",
126    "system_worked": true
127  },
128  "idea_cfg": {
129    "lr": 0.006,
130    "alpha": 1.0
131  },
132  "protocol_note": "8 paired seeds; baseline 3 learning rates x 3 alpha-parity entries; idea 3 alpha/lr settings",
133  "mechanism_signature": {
134    "predicted_unit_quaternion_norm_error": 0.0,
135    "observed_unit_quaternion_norm_error": 0.0,
136    "predicted_second_order_gain": "moderate tangent spread",
137    "observed_model_mechanism": "unit-norm retraction during recurrent propagation",
138    "confirmed": true
139  }
140}