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      "weight_decay": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.001864717691205442
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 0.0018400110420770943
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.0010901302302954718
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 0.0010647484887158498
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.01,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 0.0006186682076076977
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.01,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 0.0005735916565754451
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.0005480735380842816,
 58      "std": 0.00014996990905603093,
 59      "per_seed": [
 60        0.00046045987983234227,
 61        0.0004709371132776141,
 62        0.0004756892449222505,
 63        0.0008872803882695735,
 64        0.0004885699599981308,
 65        0.00037048623198643327,
 66        0.0006456486880779266,
 67        0.0005855167983099818
 68      ],
 69      "n": 8
 70    },
 71    "idea_grid": [
 72      {
 73        "cfg": {
 74          "lr": 0.001,
 75          "weight_decay": 0.0001,
 76          "lambda": 0.05
 77        },
 78        "mean": 0.0018400110420770943
 79      },
 80      {
 81        "cfg": {
 82          "lr": 0.001,
 83          "weight_decay": 0.0001,
 84          "lambda": 0.15
 85        },
 86        "mean": 0.0018400110420770943
 87      },
 88      {
 89        "cfg": {
 90          "lr": 0.001,
 91          "weight_decay": 0.0001,
 92          "lambda": 0.3
 93        },
 94        "mean": 0.0018400110420770943
 95      },
 96      {
 97        "cfg": {
 98          "lr": 0.003,
 99          "weight_decay": 0.0001,
100          "lambda": 0.05
101        },
102        "mean": 0.0010647484887158498
103      },
104      {
105        "cfg": {
106          "lr": 0.003,
107          "weight_decay": 0.0001,
108          "lambda": 0.15
109        },
110        "mean": 0.0010647484887158498
111      },
112      {
113        "cfg": {
114          "lr": 0.003,
115          "weight_decay": 0.0001,
116          "lambda": 0.3
117        },
118        "mean": 0.0010647484887158498
119      },
120      {
121        "cfg": {
122          "lr": 0.01,
123          "weight_decay": 0.0001,
124          "lambda": 0.05
125        },
126        "mean": 0.0005735916565754451
127      },
128      {
129        "cfg": {
130          "lr": 0.01,
131          "weight_decay": 0.0001,
132          "lambda": 0.15
133        },
134        "mean": 0.0005735916565754451
135      },
136      {
137        "cfg": {
138          "lr": 0.01,
139          "weight_decay": 0.0001,
140          "lambda": 0.3
141        },
142        "mean": 0.0005735916565754451
143      }
144    ]
145  },
146  "idea": {
147    "mean": 0.0005480735380842816,
148    "std": 0.00014996990905603093,
149    "per_seed": [
150      0.00046045987983234227,
151      0.0004709371132776141,
152      0.0004756892449222505,
153      0.0008872803882695735,
154      0.0004885699599981308,
155      0.00037048623198643327,
156      0.0006456486880779266,
157      0.0005855167983099818
158    ],
159    "n": 8
160  },
161  "comparison": {
162    "delta_mean": 0.0,
163    "idea_wins": 0,
164    "n_pairs": 8,
165    "per_seed_diffs": [
166      0.0,
167      0.0,
168      0.0,
169      0.0,
170      0.0,
171      0.0,
172      0.0,
173      0.0
174    ],
175    "p_value": 1.0,
176    "mde": 0.0,
177    "mde_rel_pct": 0.0,
178    "verdict": "no measurable effect",
179    "system_worked": false
180  },
181  "mechanism_signature": {
182    "prediction": "adding the interval-zonotope loss should change the trained model task metric",
183    "observed_nn": {
184      "baseline_mean_test_mse": 0.0005480735380842816,
185      "idea_mean_test_mse": 0.0005480735380842816,
186      "observed_delta": 0.0,
187      "best_cfg": {
188        "lr": 0.01,
189        "weight_decay": 0.0001,
190        "lambda": 0.05
191      }
192    },
193    "toy_numeric_check": {
194      "max_radius_recurrence_error": 1.6653345369377348e-16,
195      "confirmed": true
196    },
197    "confirmed": false,
198    "note": "The arithmetic radius recurrence is exact, but this does not establish the mechanism on the trained benchmark models."
199  }
200}