Certified Temporal Budget for Neural Control / results.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
 1{
 2  "seed": 2734,
 3  "derivative_verification": {
 4    "max_abs_slope_error": 1.4988010832439613e-13,
 5    "max_contract_residual": 0.0,
 6    "predicted_unit_rate": 1.0,
 7    "observed_normalized_slope": 1.0000000000001152
 8  },
 9  "math_verification": {
10    "grid_points": 6480,
11    "conservative_contract_violations": 0,
12    "conservative_violation_rate": 0.0,
13    "underestimated_gamma": 0.75,
14    "underestimate_unsafe_onset_median_L": 1.08675,
15    "expected_physical_boundary_median_L": 1.0867088607594937
16  },
17  "gamma_boundary_sweep": {
18    "predicted_gamma_transition": 0.9090909090909091,
19    "observed_gamma_transition_grid": 0.91,
20    "transition_abs_error": 0.0009090909090909705,
21    "unsafe_permissions_at_conservative_gamma": 0,
22    "unsafe_permissions_at_underestimated_gamma": 51,
23    "predicted_unsafe_onset_gamma": 0.9090909090909091,
24    "observed_first_unsafe_gamma": 0.9
25  },
26  "evaluation_scaling": {
27    "rows": [
28      {
29        "gamma": 0.5,
30        "budget": 1.9,
31        "observed_evals": 53,
32        "predicted_evals": 52.631578947368425,
33        "rate_error": 0.006999999999999928
34      },
35      {
36        "gamma": 1.0,
37        "budget": 0.8999999999999999,
38        "observed_evals": 112,
39        "predicted_evals": 111.11111111111113,
40        "rate_error": 0.007999999999999842
41      },
42      {
43        "gamma": 1.5,
44        "budget": 0.5666666666666665,
45        "observed_evals": 177,
46        "predicted_evals": 176.47058823529414,
47        "rate_error": 0.002999999999999848
48      },
49      {
50        "gamma": 2.0,
51        "budget": 0.4,
52        "observed_evals": 250,
53        "predicted_evals": 250.0,
54        "rate_error": 0.0
55      }
56    ],
57    "rate_ratio_observed_gamma2_over_gamma1": 2.232142857142857,
58    "rate_ratio_predicted_gamma2_over_gamma1": 2.2499999999999996
59  },
60  "scheduler_comparison": {
61    "episodes": 3000,
62    "periodic_unsafe_handoff_rate": 0.288,
63    "naive_event_unsafe_handoff_rate": 0.057666666666666665,
64    "contract_unsafe_handoff_rate": 0.0,
65    "contract_launch_fraction": 0.792,
66    "contract_fallback_fraction": 0.208,
67    "periodic_evaluations_per_episode": 1.0,
68    "contract_evaluations_per_episode": 0.792
69  }
70}