Self-Correcting Euler Horizon Rule / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.0002686133266252
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 2.2429956516134553e-05
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 1.2719097412627889e-05
 29      }
 30    ],
 31    "full": {
 32      "mean": 1.3684802866009704e-05,
 33      "std": 2.4297331089709535e-06,
 34      "per_seed": [
 35        1.0965275578200817e-05,
 36        1.3063688129477669e-05,
 37        1.363205592497252e-05,
 38        1.321537001786055e-05,
 39        1.694267484708689e-05,
 40        1.790331407391932e-05,
 41        1.3341849808057304e-05,
 42        1.0414194548502564e-05
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 1.3684802866009704e-05,
 49    "std": 2.4297331089709535e-06,
 50    "per_seed": [
 51      1.0965275578200817e-05,
 52      1.3063688129477669e-05,
 53      1.363205592497252e-05,
 54      1.321537001786055e-05,
 55      1.694267484708689e-05,
 56      1.790331407391932e-05,
 57      1.3341849808057304e-05,
 58      1.0414194548502564e-05
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.0,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.0,
 68      0.0,
 69      0.0,
 70      0.0,
 71      0.0,
 72      0.0,
 73      0.0,
 74      0.0
 75    ],
 76    "p_value": 1.0,
 77    "mde": 0.0,
 78    "mde_rel_pct": 0.0,
 79    "verdict": "no measurable effect",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "prediction": "contraction-aware perturbation should remain below Euler stability cap alpha<2",
 84    "predicted_alpha": 1.8,
 85    "observed_mean_lambda_h": 0.2277172260841973,
 86    "observed_idea_proxy": 7.496510079363361e-09,
 87    "rows": [
 88      {
 89        "seed": 0,
 90        "estimated_lambda": 3.643026351928711,
 91        "mean_step_h": 0.04968743026256561,
 92        "baseline_proxy": 6.96672941558063e-09,
 93        "idea_proxy": 6.96672941558063e-09
 94      },
 95      {
 96        "seed": 1,
 97        "estimated_lambda": 4.969603538513184,
 98        "mean_step_h": 0.049560677260160446,
 99        "baseline_proxy": 7.60337570682168e-09,
100        "idea_proxy": 7.60337570682168e-09
101      },
102      {
103        "seed": 2,
104        "estimated_lambda": 4.052803993225098,
105        "mean_step_h": 0.04964262992143631,
106        "baseline_proxy": 1.2150849215686321e-08,
107        "idea_proxy": 1.2150849215686321e-08
108      },
109      {
110        "seed": 3,
111        "estimated_lambda": 4.340335845947266,
112        "mean_step_h": 0.049574412405490875,
113        "baseline_proxy": 7.668859325349331e-09,
114        "idea_proxy": 7.668859325349331e-09
115      },
116      {
117        "seed": 4,
118        "estimated_lambda": 4.611820220947266,
119        "mean_step_h": 0.049705807119607925,
120        "baseline_proxy": 9.313225746154785e-09,
121        "idea_proxy": 9.313225746154785e-09
122      },
123      {
124        "seed": 5,
125        "estimated_lambda": 4.991096496582031,
126        "mean_step_h": 0.049615517258644104,
127        "baseline_proxy": 5.6461431086063385e-09,
128        "idea_proxy": 5.6461431086063385e-09
129      },
130      {
131        "seed": 6,
132        "estimated_lambda": 5.737295150756836,
133        "mean_step_h": 0.04948228970170021,
134        "baseline_proxy": 4.016328603029251e-09,
135        "idea_proxy": 4.016328603029251e-09
136      },
137      {
138        "seed": 7,
139        "estimated_lambda": 4.377944469451904,
140        "mean_step_h": 0.04963567852973938,
141        "baseline_proxy": 6.606569513678551e-09,
142        "idea_proxy": 6.606569513678551e-09
143      }
144    ],
145    "confirmed": true
146  },
147  "math_sanity": {
148    "predicted_boundary_lambda_h": 2.0,
149    "observed_grid_boundary": 1.9999999999999998,
150    "max_boundary_error": 2.220446049250313e-16,
151    "passed": true
152  },
153  "idea_sweep": [
154    {
155      "cfg": {
156        "lr": 0.006,
157        "alpha": 1.8,
158        "penalty": 0.1
159      },
160      "mean": 1.3684802866009704e-05
161    },
162    {
163      "cfg": {
164        "lr": 0.003,
165        "alpha": 0.5,
166        "penalty": 0.1
167      },
168      "mean": 2.628253741931985e-05
169    },
170    {
171      "cfg": {
172        "lr": 0.001,
173        "alpha": 1.0,
174        "penalty": 0.1
175      },
176      "mean": 0.00021658051718986826
177    }
178  ],
179  "protocol_notes": {
180    "structural_match": "controlled damped pendulum rollout tests stability/control dynamics",
181    "baseline_grid": [
182      {
183        "lr": 0.001
184      },
185      {
186        "lr": 0.003
187      },
188      {
189        "lr": 0.006
190      }
191    ],
192    "idea_grid": [
193      {
194        "lr": 0.006,
195        "alpha": 1.8,
196        "penalty": 0.1
197      },
198      {
199        "lr": 0.003,
200        "alpha": 0.5,
201        "penalty": 0.1
202      },
203      {
204        "lr": 0.001,
205        "alpha": 1.0,
206        "penalty": 0.1
207      }
208    ],
209    "epochs": 15,
210    "n_train": 4000,
211    "n_test": 1000
212  }
213}