Bellman Stopping Controller for Self-Refinement / results.json
Mechanism confirmed, baseline not beaten
1{
2 "seed": 2092,
3 "environment": {
4 "score_states": 21,
5 "max_refinements": 5,
6 "transition": "score increases one grid step with p=0.10+0.75*(1-score)",
7 "payoff": "score"
8 },
9 "math_check": {
10 "max_bellman_residual": 0.0,
11 "delta_gt_cost_matches_policy": true,
12 "value_at_initial_state_by_cost": {
13 "0.0": 0.19714903424072264,
14 "0.02": 0.09714903424072266,
15 "0.03": 0.047149034240722656,
16 "0.035": 0.02214903424072262,
17 "0.038": 0.009432143371582019,
18 "0.04": 0.0037074088745117084,
19 "0.045": 0.0,
20 "0.1": 0.0
21 }
22 },
23 "predictions": {
24 "P1_initial_continue_decreases_with_cost": {
25 "predicted": "monotone non-increasing",
26 "costs": [
27 0.0,
28 0.02,
29 0.03,
30 0.035,
31 0.038,
32 0.04,
33 0.045,
34 0.1
35 ],
36 "observed_continue": [
37 true,
38 true,
39 true,
40 true,
41 true,
42 true,
43 false,
44 false
45 ],
46 "observed_threshold_score": [
47 1.0,
48 0.6,
49 0.35,
50 0.2,
51 0.15,
52 0.1,
53 0.0,
54 0.0
55 ],
56 "confirmed": true
57 },
58 "P2_stopping_boundary_is_upper_score_region_and_moves_lower_with_cost": {
59 "predicted": "stop for sufficiently high score; boundary non-increasing in cost",
60 "boundary_monotone_each_cost": [
61 true,
62 true,
63 true,
64 true,
65 true,
66 true,
67 true,
68 true
69 ],
70 "finite_boundaries": [
71 1.0,
72 0.6,
73 0.35,
74 0.2,
75 0.15,
76 0.1,
77 0.0,
78 0.0
79 ],
80 "confirmed": true
81 },
82 "P3_bellman_value_decreases_with_cost": {
83 "predicted": "non-increasing value and exact max recursion",
84 "values": [
85 0.19714903424072264,
86 0.09714903424072266,
87 0.047149034240722656,
88 0.02214903424072262,
89 0.009432143371582019,
90 0.0037074088745117084,
91 0.0,
92 0.0
93 ],
94 "residuals": [
95 0.0,
96 0.0,
97 0.0,
98 0.0,
99 0.0,
100 0.0,
101 0.0,
102 0.0
103 ],
104 "confirmed": true
105 }
106 },
107 "simulation_cost_0.038": {
108 "adaptive": {
109 "calls": 3.62601,
110 "quality": 0.14725649999999998,
111 "objective": 0.009468120000000002,
112 "pass_at_0.75": 0.0,
113 "stop_state_mean": 2.94513
114 },
115 "fixed": {
116 "1": {
117 "calls": 1.0,
118 "quality": 0.042446500000000005,
119 "objective": 0.004446500000000005,
120 "pass_at_0.75": 0.0,
121 "stop_state_mean": 0.84893
122 },
123 "3": {
124 "calls": 3.0,
125 "quality": 0.12281,
126 "objective": 0.00881000000000001,
127 "pass_at_0.75": 0.0,
128 "stop_state_mean": 2.4562
129 },
130 "5": {
131 "calls": 5.0,
132 "quality": 0.1969025,
133 "objective": 0.0069025000000000015,
134 "pass_at_0.75": 0.0,
135 "stop_state_mean": 3.93805
136 }
137 }
138 },
139 "simulation_cost_sweep": {
140 "0.0": {
141 "adaptive": {
142 "calls": 5.0,
143 "quality": 0.19711666666666666,
144 "objective": 0.19711666666666666,
145 "pass_at_0.75": 0.0,
146 "stop_state_mean": 3.9423333333333335
147 }
148 },
149 "0.02": {
150 "adaptive": {
151 "calls": 5.0,
152 "quality": 0.197255,
153 "objective": 0.09725500000000001,
154 "pass_at_0.75": 0.0,
155 "stop_state_mean": 3.9451
156 }
157 },
158 "0.03": {
159 "adaptive": {
160 "calls": 5.0,
161 "quality": 0.19720166666666666,
162 "objective": 0.04720166666666668,
163 "pass_at_0.75": 0.0,
164 "stop_state_mean": 3.9440333333333335
165 }
166 },
167 "0.035": {
168 "adaptive": {
169 "calls": 4.603133333333333,
170 "quality": 0.18343333333333334,
171 "objective": 0.022323666666666662,
172 "pass_at_0.75": 0.0,
173 "stop_state_mean": 3.6686666666666667
174 }
175 },
176 "0.038": {
177 "adaptive": {
178 "calls": 3.6234333333333333,
179 "quality": 0.14737166666666665,
180 "objective": 0.009681200000000001,
181 "pass_at_0.75": 0.0,
182 "stop_state_mean": 2.947433333333333
183 }
184 },
185 "0.04": {
186 "adaptive": {
187 "calls": 2.3984666666666667,
188 "quality": 0.09983500000000001,
189 "objective": 0.0038963333333333385,
190 "pass_at_0.75": 0.0,
191 "stop_state_mean": 1.9967
192 }
193 },
194 "0.045": {
195 "adaptive": {
196 "calls": 0.0,
197 "quality": 0.0,
198 "objective": 0.0,
199 "pass_at_0.75": 0.0,
200 "stop_state_mean": 0.0
201 }
202 },
203 "0.1": {
204 "adaptive": {
205 "calls": 0.0,
206 "quality": 0.0,
207 "objective": 0.0,
208 "pass_at_0.75": 0.0,
209 "stop_state_mean": 0.0
210 }
211 }
212 },
213 "worked": true
214}