Feasible Action Mapping Safety Layer / 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      "epochs": 18
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "epochs": 12
 17        },
 18        "mean": 0.004545392468571663
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "epochs": 18
 24        },
 25        "mean": 0.001921573159052059
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "epochs": 12
 31        },
 32        "mean": 0.00358188062091358
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "epochs": 18
 38        },
 39        "mean": 0.0011179699067724869
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "epochs": 12
 45        },
 46        "mean": 0.0019017510931007564
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "epochs": 18
 52        },
 53        "mean": 0.0007884320366429165
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.0007652444473933429,
 58      "std": 0.00014266442888030616,
 59      "per_seed": [
 60        0.0007294044480659068,
 61        0.0007733436650596559,
 62        0.0005735447048209608,
 63        0.0010774353286251426,
 64        0.0006340121617540717,
 65        0.0008162351441569626,
 66        0.0006983152125030756,
 67        0.0008196649141609669
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.00781346601434052,
 74    "std": 0.0006524094833102025,
 75    "per_seed": [
 76      0.00748194707557559,
 77      0.008646145462989807,
 78      0.007898174226284027,
 79      0.008135655894875526,
 80      0.006896249484270811,
 81      0.006903830915689468,
 82      0.008720104582607746,
 83      0.007825620472431183
 84    ],
 85    "n": 8
 86  },
 87  "comparison": {
 88    "delta_mean": 0.007048221566947177,
 89    "idea_wins": 0,
 90    "n_pairs": 8,
 91    "per_seed_diffs": [
 92      0.006752542627509683,
 93      0.007872801797930151,
 94      0.007324629521463066,
 95      0.007058220566250384,
 96      0.006262237322516739,
 97      0.006087595771532506,
 98      0.00802178937010467,
 99      0.007005955558270216
100    ],
101    "p_value": 0.0081,
102    "mde": 0.0005770632681359283,
103    "mde_rel_pct": 75.40901082021342,
104    "verdict": "idea worse (significant)",
105    "system_worked": false
106  },
107  "mechanism_signature": {
108    "prediction": "feasible predictions have zero projection distance; infeasible predictions have positive distance",
109    "observed_inside_fraction": 0.17499999701976776,
110    "observed_mean_projection_distance": 0.07038450986146927,
111    "observed_max_projection_distance": 0.2860039472579956,
112    "confirmed": true,
113    "idea_sweep": [
114      {
115        "cfg": {
116          "lr": 0.006,
117          "epochs": 18
118        },
119        "mean": 0.00799608719535172
120      },
121      {
122        "cfg": {
123          "lr": 0.003,
124          "epochs": 18
125        },
126        "mean": 0.007877308991737664
127      },
128      {
129        "cfg": {
130          "lr": 0.001,
131          "epochs": 18
132        },
133        "mean": 0.00781346601434052
134      }
135    ],
136    "budget": {
137      "epochs": 18,
138      "batch": 128
139    }
140  }
141}