1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "residual_rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.08261195849627256
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.06586439814418554
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.15003963327035308
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.061237716814503074,
 33      "std": 0.027211549075682028,
 34      "per_seed": [
 35        0.04314250871539116,
 36        0.09609165042638779,
 37        0.08621135354042053,
 38        0.038012079894542694,
 39        0.07434208691120148,
 40        0.028976354748010635,
 41        0.09323041886091232,
 42        0.029895281419157982
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.0020011790038552135,
 49    "std": 0.0007409600802683955,
 50    "per_seed": [
 51      0.0010639947140589356,
 52      0.0032321331091225147,
 53      0.0009753473568707705,
 54      0.0024151953402906656,
 55      0.002823632676154375,
 56      0.0017429819563403726,
 57      0.0018509947694838047,
 58      0.0019051521085202694
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -0.05923653781064786,
 64    "idea_wins": 8,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.04207851400133222,
 68      -0.09285951731726527,
 69      -0.08523600618354976,
 70      -0.03559688455425203,
 71      -0.0715184542350471,
 72      -0.027233372791670263,
 73      -0.09137942409142852,
 74      -0.027990129310637712
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.02416027485033725,
 78    "mde_rel_pct": 39.453258722107215,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "math_check": {
 83    "h": [
 84      0.5,
 85      1.0,
 86      2.0,
 87      4.0,
 88      8.0,
 89      32.0
 90    ],
 91    "q": [
 92      0.4,
 93      0.6666666666666666,
 94      1.0,
 95      1.3333333333333333,
 96      1.6,
 97      1.8823529411764706
 98    ],
 99    "q_bound": 2.0,
100    "euler_amplification": [
101      0.5,
102      0.0,
103      1.0,
104      3.0,
105      7.0,
106      31.0
107    ],
108    "damped_amplification": [
109      0.6,
110      0.33333333333333337,
111      0.0,
112      0.33333333333333326,
113      0.6000000000000001,
114      0.8823529411764706
115    ],
116    "predicted_bounded_effective_step": true,
117    "damped_nonexpansive_test": true,
118    "euler_nonexpansive_test": false
119  },
120  "idea_config": {
121    "lr": 0.001
122  },
123  "idea_sweep": [
124    {
125      "cfg": {
126        "lr": 0.001
127      },
128      "result": {
129        "mean": 0.0020011790038552135,
130        "std": 0.0007409600802683955,
131        "per_seed": [
132          0.0010639947140589356,
133          0.0032321331091225147,
134          0.0009753473568707705,
135          0.0024151953402906656,
136          0.002823632676154375,
137          0.0017429819563403726,
138          0.0018509947694838047,
139          0.0019051521085202694
140        ],
141        "n": 8
142      }
143    },
144    {
145      "cfg": {
146        "lr": 0.003
147      },
148      "result": {
149        "mean": 0.002582564891781658,
150        "std": 0.000729676529186581,
151        "per_seed": [
152          0.0016574665205553174,
153          0.001894353306852281,
154          0.002129645086824894,
155          0.0034991740249097347,
156          0.0038091319147497416,
157          0.003046745667234063,
158          0.0024077408015727997,
159          0.002216261811554432
160        ],
161        "n": 8
162      }
163    },
164    {
165      "cfg": {
166        "lr": 0.01
167      },
168      "result": {
169        "mean": 0.03438545443350449,
170        "std": 0.037436245598765676,
171        "per_seed": [
172          0.009465613402426243,
173          0.010761785320937634,
174          0.054800208657979965,
175          0.007569439243525267,
176          0.007234271615743637,
177          0.014039256609976292,
178          0.12095119804143906,
179          0.05026186257600784
180        ],
181        "n": 8
182      }
183    }
184  ],
185  "mechanism_signature": {
186    "prediction": "denominator block has smaller trained-state update proxy at h=4",
187    "nominal_h": 4.0,
188    "alpha": 0.5,
189    "observed_baseline_update": 21.13020896911621,
190    "observed_idea_update": 4.463923364877701,
191    "observed_update_ratio": 0.21125789013266005,
192    "predicted_upper_ratio_from_q_over_h": 0.3333333333333333,
193    "confirmed": true
194  }
195}