1{
  2  "bench_version": 1,
  3  "track": "sequence",
  4  "model": "frozen_random_attention_prompt",
  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": 1.0786777287721634
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 1.0726252794265747
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.921186164021492
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.9356768578290939,
 33      "std": 0.10686743285939519,
 34      "per_seed": [
 35        1.0163524150848389,
 36        0.7307133078575134,
 37        1.0717743635177612,
 38        0.8659045696258545,
 39        0.892680823802948,
 40        1.0664112567901611,
 41        0.9147705435752869,
 42        0.9268075823783875
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 1.7406605532973132,
 49    "std": 0.15897193757942363,
 50    "per_seed": [
 51      2.0928197670190434,
 52      1.6468335788158937,
 53      1.6924736704763956,
 54      1.668518826794135,
 55      1.6360904797108695,
 56      1.7143411545296539,
 57      1.89549804751905,
 58      1.5787089015134637
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.8049836954682192,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      1.0764673519342045,
 68      0.9161202709583802,
 69      0.6206993069586344,
 70      0.8026142571682806,
 71      0.7434096559079215,
 72      0.6479298977394927,
 73      0.9807275039437631,
 74      0.6519013191350762
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.14232114117001493,
 78    "mde_rel_pct": 15.21050135836646,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "track_choice": "sequence: multi-token temporal-window correlations require sequence attention",
 84    "best_sigma": 0.8,
 85    "idea_sweep": [
 86      {
 87        "cfg": {
 88          "sigma": 0.4,
 89          "lr": 0.001
 90        },
 91        "result": {
 92          "mean": 1.8291889989446934,
 93          "std": 0.16079926338695508,
 94          "per_seed": [
 95            2.187105139019409,
 96            1.7452456227531217,
 97            1.7904076713809234,
 98            1.7679971041431175,
 99            1.6905353869220097,
100            1.8289776945690013,
101            1.9642818351456248,
102            1.6589615376243407
103          ],
104          "n": 8
105        }
106      },
107      {
108        "cfg": {
109          "sigma": 0.6,
110          "lr": 0.003
111        },
112        "result": {
113          "mean": 1.7922968691831327,
114          "std": 0.15999984262010897,
115          "per_seed": [
116            2.1471634715202894,
117            1.7083331174848266,
118            1.74349576800059,
119            1.722126782043419,
120            1.669563403023254,
121            1.7793366851501542,
122            1.940943109339341,
123            1.6274126169031866
124          ],
125          "n": 8
126        }
127      },
128      {
129        "cfg": {
130          "sigma": 0.8,
131          "lr": 0.01
132        },
133        "result": {
134          "mean": 1.7406605532973132,
135          "std": 0.15897193757942363,
136          "per_seed": [
137            2.0928197670190434,
138            1.6468335788158937,
139            1.6924736704763956,
140            1.668518826794135,
141            1.6360904797108695,
142            1.7143411545296539,
143            1.89549804751905,
144            1.5787089015134637
145          ],
146          "n": 8
147        }
148      }
149    ],
150    "trained_baseline_behavior": {
151      "0": {
152        "prompt_norm": 19.529014587402344,
153        "attention_entropy": 4.115479946136475
154      },
155      "1": {
156        "prompt_norm": 15.537064552307129,
157        "attention_entropy": 4.137622833251953
158      },
159      "2": {
160        "prompt_norm": 9.14969253540039,
161        "attention_entropy": 4.145155906677246
162      },
163      "3": {
164        "prompt_norm": 17.088436126708984,
165        "attention_entropy": 4.135089874267578
166      },
167      "4": {
168        "prompt_norm": 19.799318313598633,
169        "attention_entropy": 4.124661445617676
170      },
171      "5": {
172        "prompt_norm": 11.53364086151123,
173        "attention_entropy": 4.150853157043457
174      },
175      "6": {
176        "prompt_norm": 18.757871627807617,
177        "attention_entropy": 4.126466751098633
178      },
179      "7": {
180        "prompt_norm": 12.221356391906738,
181        "attention_entropy": 4.149312973022461
182      }
183    },
184    "mechanism_prediction": {
185      "prediction": "prompt norm scales as sigma^-2; affine logit residual is near zero",
186      "observed_prompt_norms": {
187        "0.4": 2036.919040636862,
188        "0.6": 905.2893090361015,
189        "0.8": 509.21965520351125
190      },
191      "predicted_norm_ratio_sigma_0.4_to_0.8": 4.0,
192      "observed_norm_ratio_sigma_0.4_to_0.8": 4.000079376006806,
193      "observed_best_logit_residual": 8.765288996216841e-14,
194      "confirmed": true
195    }
196  },
197  "baseline_sweep_shared_lr_grid": [
198    0.001,
199    0.003,
200    0.01
201  ],
202  "cut": "Only sequence track tested; no CIFAR/MNIST transfer and no latency/FLOP study."
203}