Adaptive conformal safety margins / bench_report.json

Failed on benchmark

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": 0.0005152521189302206
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 4.9621185098658316e-05
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 2.345215943932999e-05
 29      }
 30    ],
 31    "full": {
 32      "mean": 2.282289460708853e-05,
 33      "std": 2.305330446600017e-06,
 34      "per_seed": [
 35        2.4922346710809506e-05,
 36        2.6788266040966846e-05,
 37        1.934224746946711e-05,
 38        2.2755777536076494e-05,
 39        2.3468126528314315e-05,
 40        2.385292282269802e-05,
 41        2.0721481632790528e-05,
 42        2.073198811558541e-05
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 2.4683711671968922e-05,
 49    "std": 2.8119869264952184e-06,
 50    "per_seed": [
 51      2.5581410227459855e-05,
 52      2.4755079721217044e-05,
 53      1.922530645970255e-05,
 54      2.4516248231520876e-05,
 55      2.3418831915478222e-05,
 56      2.4992359612951986e-05,
 57      2.4735531042097136e-05,
 58      3.0244926165323704e-05
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 1.8608170648803934e-06,
 64    "idea_wins": 3,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      6.590635166503489e-07,
 68      -2.0331863197498024e-06,
 69      -1.1694100976455957e-07,
 70      1.7604706954443827e-06,
 71      -4.929461283609271e-08,
 72      1.139436790253967e-06,
 73      4.014049409306608e-06,
 74      9.512938049738295e-06
 75    ],
 76    "p_value": 0.1888,
 77    "mde": 2.9628199598181274e-06,
 78    "mde_rel_pct": 12.981788729366123,
 79    "verdict": "no significant win",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "math_sanity": {
 84      "stationary_target_alpha": 0.2,
 85      "stationary_observed_exceedance": 0.20028,
 86      "stationary_abs_error": 0.00028000000000000247,
 87      "ramp_observed_steps": 19,
 88      "ramp_predicted_steps": 19,
 89      "recovery_observed_steps": 88,
 90      "recovery_predicted_steps": 88,
 91      "quantile_check": 0.8500000000000001
 92    },
 93    "trained_model_signature": {
 94      "test_abs_error_mean": 0.003862829878926277,
 95      "test_abs_error_q90": 0.007725757360458375,
 96      "retested_exceedance": 0.197,
 97      "target_alpha": 0.2,
 98      "final_radius": 0.0,
 99      "confirmed": true,
100      "radius_training_start_end": [
101        0.19,
102        0.0
103      ]
104    },
105    "intervention": "adaptive conformal exceedance-weighted MSE",
106    "structural_match": "controlled pendulum multi-step dynamics"
107  },
108  "idea_sweep": [
109    {
110      "cfg": {
111        "lr": 0.001,
112        "alpha": 0.2,
113        "gamma": 0.05,
114        "strength": 0.5
115      },
116      "result": {
117        "mean": 0.0004968880639353301,
118        "std": 4.8528307616895607e-05,
119        "per_seed": [
120          0.0004283393791411072,
121          0.0005958400433883071,
122          0.0004622905398719013,
123          0.0005341034266166389,
124          0.0004780906019732356,
125          0.0004924809327349067,
126          0.000466031429823488,
127          0.0005179281579330564
128        ],
129        "n": 8
130      }
131    },
132    {
133      "cfg": {
134        "lr": 0.003,
135        "alpha": 0.2,
136        "gamma": 0.05,
137        "strength": 0.5
138      },
139      "result": {
140        "mean": 5.3715010380983585e-05,
141        "std": 1.4501156678207892e-05,
142        "per_seed": [
143          8.585918840253726e-05,
144          6.12151634413749e-05,
145          4.8918511311057955e-05,
146          3.291441680630669e-05,
147          5.7153552916133776e-05,
148          4.41737865912728e-05,
149          4.8585090553388e-05,
150          5.090037302579731e-05
151        ],
152        "n": 8
153      }
154    },
155    {
156      "cfg": {
157        "lr": 0.006,
158        "alpha": 0.2,
159        "gamma": 0.05,
160        "strength": 0.5
161      },
162      "result": {
163        "mean": 2.4683711671968922e-05,
164        "std": 2.8119869264952184e-06,
165        "per_seed": [
166          2.5581410227459855e-05,
167          2.4755079721217044e-05,
168          1.922530645970255e-05,
169          2.4516248231520876e-05,
170          2.3418831915478222e-05,
171          2.4992359612951986e-05,
172          2.4735531042097136e-05,
173          3.0244926165323704e-05
174        ],
175        "n": 8
176      }
177    }
178  ]
179}