1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "mlp_tiny",
  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": 130.23694229125977
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 17.47633147239685
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 11.130715131759644
 29      }
 30    ],
 31    "full": {
 32      "mean": 11.6078120470047,
 33      "std": 1.1735198758690657,
 34      "per_seed": [
 35        10.810468673706055,
 36        12.079232215881348,
 37        10.855374336242676,
 38        10.777785301208496,
 39        9.649698257446289,
 40        12.84527587890625,
 41        12.719051361083984,
 42        13.1256103515625
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 8.623416543006897,
 49    "std": 1.4506015215736154,
 50    "per_seed": [
 51      7.800422191619873,
 52      11.718671798706055,
 53      10.022994041442871,
 54      6.932704925537109,
 55      7.960085868835449,
 56      8.643534660339355,
 57      8.414508819580078,
 58      7.494410037994385
 59    ],
 60    "n": 8,
 61    "details": [
 62      {
 63        "metric": 7.800422191619873,
 64        "seconds": 7.774140733999957,
 65        "accepted_steps": 12,
 66        "median_alpha": 0.3333333333333333,
 67        "mean_rank": 389.5833333333333,
 68        "max_jacobian_condition": 3202.399169921875,
 69        "signature": {
 70          "predicted_full_to_truncated_amplification": 1000000064.0,
 71          "observed_full_step_norm": 0.04894990101456642,
 72          "observed_truncated_step_norm": 0.04894990101456642,
 73          "observed_amplification": 1.0,
 74          "confirmed": false
 75        }
 76      },
 77      {
 78        "metric": 11.718671798706055,
 79        "seconds": 9.029400838999209,
 80        "accepted_steps": 12,
 81        "median_alpha": 0.3333333333333333,
 82        "mean_rank": 390.6666666666667,
 83        "max_jacobian_condition": 4148.69775390625
 84      },
 85      {
 86        "metric": 10.022994041442871,
 87        "seconds": 9.43387127599999,
 88        "accepted_steps": 12,
 89        "median_alpha": 0.3333333333333333,
 90        "mean_rank": 400.0,
 91        "max_jacobian_condition": 846.0858154296875
 92      },
 93      {
 94        "metric": 6.932704925537109,
 95        "seconds": 9.006122887998572,
 96        "accepted_steps": 12,
 97        "median_alpha": 0.3333333333333333,
 98        "mean_rank": 400.0,
 99        "max_jacobian_condition": 852.4298095703125
100      },
101      {
102        "metric": 7.960085868835449,
103        "seconds": 9.027963689999524,
104        "accepted_steps": 12,
105        "median_alpha": 0.3333333333333333,
106        "mean_rank": 370.5,
107        "max_jacobian_condition": 2142.92919921875
108      },
109      {
110        "metric": 8.643534660339355,
111        "seconds": 8.847281662001478,
112        "accepted_steps": 12,
113        "median_alpha": 0.3333333333333333,
114        "mean_rank": 399.5833333333333,
115        "max_jacobian_condition": 1440.25390625
116      },
117      {
118        "metric": 8.414508819580078,
119        "seconds": 8.764552226000887,
120        "accepted_steps": 12,
121        "median_alpha": 0.3333333333333333,
122        "mean_rank": 109.83333333333333,
123        "max_jacobian_condition": 17170.849609375
124      },
125      {
126        "metric": 7.494410037994385,
127        "seconds": 8.539251098000022,
128        "accepted_steps": 12,
129        "median_alpha": 0.3333333333333333,
130        "mean_rank": 400.0,
131        "max_jacobian_condition": 614.7550048828125
132      }
133    ]
134  },
135  "comparison": {
136    "delta_mean": -2.9843955039978027,
137    "idea_wins": 8,
138    "n_pairs": 8,
139    "per_seed_diffs": [
140      -3.0100464820861816,
141      -0.36056041717529297,
142      -0.8323802947998047,
143      -3.8450803756713867,
144      -1.6896123886108398,
145      -4.2017412185668945,
146      -4.304542541503906,
147      -5.631200313568115
148    ],
149    "p_value": 0.0081,
150    "mde": 1.5535327571979936,
151    "mde_rel_pct": 13.383510612569488,
152    "verdict": "idea better (significant)",
153    "system_worked": true
154  },
155  "method_notes": {
156    "track_justification": "Optimizer idea matched to tabular regression.",
157    "idea_sweep": [
158      {
159        "cfg": {
160          "lr": 0.001
161        },
162        "mean": 9.118698239326477
163      },
164      {
165        "cfg": {
166          "lr": 0.003
167        },
168        "mean": 10.81938362121582
169      },
170      {
171        "cfg": {
172          "lr": 0.01
173        },
174        "mean": 10.81938362121582
175      }
176    ],
177    "idea_best_cfg": {
178      "lr": 0.001
179    },
180    "sanity": {
181      "singular_values": [
182        1.0,
183        9.999999717180685e-10
184      ],
185      "full_step_norm": 100000008.0,
186      "truncated_step_norm": 0.10000000149011612,
187      "amplification_ratio": 1000000064.0,
188      "kept_rank": 1
189    },
190    "epochs": 12,
191    "batch": 128,
192    "tau": 1e-06,
193    "rho": 0.001
194  },
195  "mechanism_signature": {
196    "predicted_full_to_truncated_amplification": 1000000064.0,
197    "observed_full_step_norm": 0.04894990101456642,
198    "observed_truncated_step_norm": 0.04894990101456642,
199    "observed_amplification": 1.0,
200    "confirmed": false
201  }
202}