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": 0.0015,
 10      "epochs": 12
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.0015,
 16          "epochs": 12
 17        },
 18        "mean": 0.42323087900877
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "epochs": 12
 24        },
 25        "mean": 0.4927415996789932
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "epochs": 12
 31        },
 32        "mean": 0.721754178404808
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.44760169088840485,
 37      "std": 0.0483900971330472,
 38      "per_seed": [
 39        0.3936149477958679,
 40        0.3890821635723114,
 41        0.40740370750427246,
 42        0.5028226971626282,
 43        0.4323761761188507,
 44        0.4885129928588867,
 45        0.5237836837768555,
 46        0.4432171583175659
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "config": {
 53      "lr": 0.0015,
 54      "epochs": 12
 55    },
 56    "per_seed": [
 57      0.36546337604522705,
 58      0.33078253269195557,
 59      0.3787442743778229,
 60      0.4493727684020996,
 61      0.35211890935897827,
 62      0.4167359173297882,
 63      0.4435553252696991,
 64      0.30875149369239807
 65    ],
 66    "details": [
 67      {
 68        "seed": 0,
 69        "metric": 0.36546337604522705,
 70        "final_train_loss": 0.3290974372625351
 71      },
 72      {
 73        "seed": 1,
 74        "metric": 0.33078253269195557,
 75        "final_train_loss": 0.3341557741165161
 76      },
 77      {
 78        "seed": 2,
 79        "metric": 0.3787442743778229,
 80        "final_train_loss": 0.3714332640171051
 81      },
 82      {
 83        "seed": 3,
 84        "metric": 0.4493727684020996,
 85        "final_train_loss": 0.35627098798751833
 86      },
 87      {
 88        "seed": 4,
 89        "metric": 0.35211890935897827,
 90        "final_train_loss": 0.3012402510643005
 91      },
 92      {
 93        "seed": 5,
 94        "metric": 0.4167359173297882,
 95        "final_train_loss": 0.34340501070022583
 96      },
 97      {
 98        "seed": 6,
 99        "metric": 0.4435553252696991,
100        "final_train_loss": 0.3110033637285233
101      },
102      {
103        "seed": 7,
104        "metric": 0.30875149369239807,
105        "final_train_loss": 0.2961388397216797
106      }
107    ],
108    "mean": 0.3806905746459961
109  },
110  "comparison": {
111    "delta_mean": -0.06691111624240875,
112    "idea_wins": 8,
113    "n_pairs": 8,
114    "per_seed_diffs": [
115      -0.02815157175064087,
116      -0.058299630880355835,
117      -0.028659433126449585,
118      -0.053449928760528564,
119      -0.08025726675987244,
120      -0.07177707552909851,
121      -0.08022835850715637,
122      -0.13446566462516785
123    ],
124    "p_value": 0.0081,
125    "mde": 0.028554757083691817,
126    "mde_rel_pct": 6.379501611581497,
127    "verdict": "idea better (significant)",
128    "system_worked": true
129  },
130  "mechanism_signature": {
131    "quantity": "trained gate vs one-step Fisher-KPP prediction",
132    "predicted_gate_rmse": 0.4042392542224135,
133    "observed_front_index_mean": 0.37890625,
134    "tolerance": 0.15,
135    "confirmed": false
136  },
137  "math_sanity": {
138    "speed_scaling": [
139      {
140        "D": 0.1,
141        "r": 1.0,
142        "predicted": 0.6324555320336759,
143        "observed": 0.6649048996195985,
144        "relative_error": 0.05130695510177757
145      },
146      {
147        "D": 0.25,
148        "r": 1.0,
149        "predicted": 1.0,
150        "observed": 0.9468693074772295,
151        "relative_error": 0.05313069252277047
152      },
153      {
154        "D": 0.5,
155        "r": 1.0,
156        "predicted": 1.4142135623730951,
157        "observed": 1.268869715478862,
158        "relative_error": 0.1027736197426516
159      }
160    ],
161    "stability_boundary": [
162      {
163        "r": -0.5,
164        "predicted": -0.5,
165        "observed": -0.5063520554424232,
166        "abs_error": 0.006352055442423232
167      },
168      {
169        "r": 0.0,
170        "predicted": 0.0,
171        "observed": -1.083053609166525e-14,
172        "abs_error": 1.083053609166525e-14
173      },
174      {
175        "r": 0.5,
176        "predicted": 0.5,
177        "observed": 0.49384613004092737,
178        "abs_error": 0.006153869959072633
179      }
180    ]
181  },
182  "protocol_note": "8 paired seeds; baseline sweep and idea sweep use identical lr/epoch union; n_train=400."
183}