1{
  2  "bench_version": 1,
  3  "track": "sequence",
  4  "model": "transformer_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.001,
 10      "epochs": 8,
 11      "rho": 0.0,
 12      "beta": 0.0
 13    },
 14    "sweep": [
 15      {
 16        "cfg": {
 17          "lr": 0.001,
 18          "epochs": 8,
 19          "rho": 0.0,
 20          "beta": 0.0
 21        },
 22        "mean": 0.5321156531572342
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.003,
 27          "epochs": 8,
 28          "rho": 0.0,
 29          "beta": 0.0
 30        },
 31        "mean": 0.6582250595092773
 32      },
 33      {
 34        "cfg": {
 35          "lr": 0.006,
 36          "epochs": 8,
 37          "rho": 0.0,
 38          "beta": 0.0
 39        },
 40        "mean": 0.91780074685812
 41      }
 42    ],
 43    "full": {
 44      "mean": 0.5321156531572342,
 45      "std": 0.03988514268446927,
 46      "per_seed": [
 47        0.5299221873283386,
 48        0.45885390043258667,
 49        0.5064432621002197,
 50        0.5547861456871033,
 51        0.5245285630226135,
 52        0.5539869070053101,
 53        0.6065487861633301,
 54        0.5218554735183716
 55      ],
 56      "n": 8
 57    }
 58  },
 59  "idea": {
 60    "mean": 0.5394741781055927,
 61    "std": 0.0655720338898959,
 62    "per_seed": [
 63      0.6348236799240112,
 64      0.39886191487312317,
 65      0.5056540369987488,
 66      0.5536980628967285,
 67      0.5338740348815918,
 68      0.5505685210227966,
 69      0.602857768535614,
 70      0.5354554057121277
 71    ],
 72    "n": 8
 73  },
 74  "comparison": {
 75    "delta_mean": 0.007358524948358536,
 76    "idea_wins": 5,
 77    "n_pairs": 8,
 78    "per_seed_diffs": [
 79      0.10490149259567261,
 80      -0.0599919855594635,
 81      -0.0007892251014709473,
 82      -0.0010880827903747559,
 83      0.009345471858978271,
 84      -0.0034183859825134277,
 85      -0.0036910176277160645,
 86      0.013599932193756104
 87    ],
 88    "p_value": 0.6281,
 89    "mde": 0.03801589735734751,
 90    "mde_rel_pct": 7.1442922477069555,
 91    "verdict": "no significant win",
 92    "system_worked": false
 93  },
 94  "mechanism_signature": {
 95    "track_rationale": "Sequence forecast contains multi-token temporal correlations and matches the proposed per-token fading-memory transformer gate.",
 96    "observed_best_cfg": {
 97      "lr": 0.001,
 98      "epochs": 8,
 99      "rho": 0.85,
100      "beta": 0.5
101    },
102    "idea_sweep": [
103      {
104        "cfg": {
105          "lr": 0.001,
106          "epochs": 8,
107          "rho": 0.85,
108          "beta": 0.5
109        },
110        "result": {
111          "mean": 0.5394741781055927,
112          "std": 0.0655720338898959,
113          "per_seed": [
114            0.6348236799240112,
115            0.39886191487312317,
116            0.5056540369987488,
117            0.5536980628967285,
118            0.5338740348815918,
119            0.5505685210227966,
120            0.602857768535614,
121            0.5354554057121277
122          ],
123          "n": 8
124        }
125      },
126      {
127        "cfg": {
128          "lr": 0.003,
129          "epochs": 8,
130          "rho": 0.9,
131          "beta": 0.5
132        },
133        "result": {
134          "mean": 0.6679390706121922,
135          "std": 0.11291799270623928,
136          "per_seed": [
137            0.57618248462677,
138            0.4957140386104584,
139            0.6142106652259827,
140            0.6856930255889893,
141            0.6076048016548157,
142            0.8247736692428589,
143            0.6928988695144653,
144            0.846435010433197
145          ],
146          "n": 8
147        }
148      },
149      {
150        "cfg": {
151          "lr": 0.006,
152          "epochs": 8,
153          "rho": 0.9,
154          "beta": 1.0
155        },
156        "result": {
157          "mean": 0.9237265363335609,
158          "std": 0.17665093311633415,
159          "per_seed": [
160            1.304537057876587,
161            0.6689846515655518,
162            0.846419632434845,
163            1.0522245168685913,
164            0.8211356401443481,
165            0.86773282289505,
166            0.9556908011436462,
167            0.8730871677398682
168          ],
169          "n": 8
170        }
171      }
172    ],
173    "mechanism_signature": {
174      "trained_model_seed": 0,
175      "metric": 0.6348236799240112,
176      "initial_gain": 0.9424585700035095,
177      "final_gain": 0.7104766368865967,
178      "mean_state": 0.8186129927635193,
179      "predicted_steady_gain": 0.7095688571417152,
180      "fitted_state_pole": 1.0177967548370361,
181      "rho": 0.85,
182      "beta": 0.5,
183      "confirmed": false
184    },
185    "math_check": {
186      "rows": [
187        {
188          "rho": 0.85,
189          "steady_gain_abs_error": 3.3306690738754696e-16,
190          "half_observed": 5,
191          "half_predicted": 5
192        },
193        {
194          "rho": 0.9,
195          "steady_gain_abs_error": 4.773959005888173e-15,
196          "half_observed": 7,
197          "half_predicted": 7
198        },
199        {
200          "rho": 0.9,
201          "steady_gain_abs_error": 4.385380947269368e-15,
202          "half_observed": 7,
203          "half_predicted": 7
204        }
205      ],
206      "confirmed": true
207    },
208    "protocol": {
209      "seeds": [
210        0,
211        1,
212        2,
213        3,
214        4,
215        5,
216        6,
217        7
218      ],
219      "epochs": 8,
220      "baseline_lr_grid": [
221        0.001,
222        0.003,
223        0.006
224      ],
225      "idea_lr_grid": [
226        0.001,
227        0.003,
228        0.006
229      ]
230    }
231  }
232}