rb125 commited on
Commit
ad6d71e
·
1 Parent(s): 3f8f8eb

added synthetic simulation runner

Browse files
server/results/agent_details.json ADDED
@@ -0,0 +1,157 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "conservative_0": {
3
+ "agent_id": "agent_cf56e1c14e84",
4
+ "model_name": "conservative_0",
5
+ "architecture_hash": "8374b7c44322b61a",
6
+ "status": "active",
7
+ "current_tier": "T3",
8
+ "balance": 0.49150872061921697,
9
+ "total_earned": 0.035680765649084345,
10
+ "total_spent": 0.038000000000000034,
11
+ "total_penalties": 0.0061720450298669545,
12
+ "total_topups": 0.0,
13
+ "contracts_completed": 13,
14
+ "contracts_failed": 7,
15
+ "registration_time": 0.0,
16
+ "audit_cid": null,
17
+ "wallet_address": null,
18
+ "robustness": {
19
+ "cc": 0.9255496218337828,
20
+ "er": 0.7985244829016913,
21
+ "as": 0.772250481405939,
22
+ "ih": 0.8472803268022182
23
+ },
24
+ "strategy": "conservative",
25
+ "true_robustness": {
26
+ "cc": 0.85,
27
+ "er": 0.8,
28
+ "as": 0.75,
29
+ "ih": 0.9
30
+ },
31
+ "decisions_count": 20
32
+ },
33
+ "aggressive_1": {
34
+ "agent_id": "agent_28f8af722342",
35
+ "model_name": "aggressive_1",
36
+ "architecture_hash": "06cf6cd1e0157f7e",
37
+ "status": "active",
38
+ "current_tier": "T1",
39
+ "balance": 0.5277924652461742,
40
+ "total_earned": 0.04979246524617394,
41
+ "total_spent": 0.022000000000000026,
42
+ "total_penalties": 0.0,
43
+ "total_topups": 0.0,
44
+ "contracts_completed": 15,
45
+ "contracts_failed": 0,
46
+ "registration_time": 0.0,
47
+ "audit_cid": null,
48
+ "wallet_address": null,
49
+ "robustness": {
50
+ "cc": 0.4237871475106476,
51
+ "er": 0.43773895056453505,
52
+ "as": 0.34136116942599515,
53
+ "ih": 0.6714820583560304
54
+ },
55
+ "strategy": "aggressive",
56
+ "true_robustness": {
57
+ "cc": 0.35,
58
+ "er": 0.4,
59
+ "as": 0.3,
60
+ "ih": 0.7
61
+ },
62
+ "decisions_count": 20
63
+ },
64
+ "balanced_2": {
65
+ "agent_id": "agent_90781acc6dc0",
66
+ "model_name": "balanced_2",
67
+ "architecture_hash": "ea00df5aae26a51e",
68
+ "status": "active",
69
+ "current_tier": "T1",
70
+ "balance": 0.5403475830647143,
71
+ "total_earned": 0.14271936033504076,
72
+ "total_spent": 0.022,
73
+ "total_penalties": 0.08037177727032704,
74
+ "total_topups": 0.0,
75
+ "contracts_completed": 9,
76
+ "contracts_failed": 11,
77
+ "registration_time": 0.0,
78
+ "audit_cid": null,
79
+ "wallet_address": null,
80
+ "robustness": {
81
+ "cc": 0.6509882462532015,
82
+ "er": 0.529896012603673,
83
+ "as": 0.4386817356571334,
84
+ "ih": 0.8320949702616935
85
+ },
86
+ "strategy": "balanced",
87
+ "true_robustness": {
88
+ "cc": 0.6,
89
+ "er": 0.55,
90
+ "as": 0.5,
91
+ "ih": 0.8
92
+ },
93
+ "decisions_count": 20
94
+ },
95
+ "adaptive_3": {
96
+ "agent_id": "agent_33d2e87d3579",
97
+ "model_name": "adaptive_3",
98
+ "architecture_hash": "b191a86339607a12",
99
+ "status": "active",
100
+ "current_tier": "T1",
101
+ "balance": 0.4866801928830522,
102
+ "total_earned": 0.002893322731259686,
103
+ "total_spent": 0.013999999999999999,
104
+ "total_penalties": 0.0022131298482070774,
105
+ "total_topups": 0.0,
106
+ "contracts_completed": 1,
107
+ "contracts_failed": 2,
108
+ "registration_time": 0.0,
109
+ "audit_cid": null,
110
+ "wallet_address": null,
111
+ "robustness": {
112
+ "cc": 0.5696990952039601,
113
+ "er": 0.5033152153233149,
114
+ "as": 0.42785035192965537,
115
+ "ih": 0.7695601289753685
116
+ },
117
+ "strategy": "adaptive",
118
+ "true_robustness": {
119
+ "cc": 0.55,
120
+ "er": 0.5,
121
+ "as": 0.45,
122
+ "ih": 0.8
123
+ },
124
+ "decisions_count": 20
125
+ },
126
+ "cheater_4": {
127
+ "agent_id": "agent_85b2a0234c59",
128
+ "model_name": "cheater_4",
129
+ "architecture_hash": "73482227d1d7b2b3",
130
+ "status": "active",
131
+ "current_tier": "T0",
132
+ "balance": 0.48599999999999954,
133
+ "total_earned": 0.0,
134
+ "total_spent": 0.013999999999999999,
135
+ "total_penalties": 0.0,
136
+ "total_topups": 0.0,
137
+ "contracts_completed": 0,
138
+ "contracts_failed": 0,
139
+ "registration_time": 0.0,
140
+ "audit_cid": null,
141
+ "wallet_address": null,
142
+ "robustness": {
143
+ "cc": 0.707390265856336,
144
+ "er": 0.2893324248161214,
145
+ "as": 0.6512497059171015,
146
+ "ih": 0.5968103011868765
147
+ },
148
+ "strategy": "cheater",
149
+ "true_robustness": {
150
+ "cc": 0.7,
151
+ "er": 0.25,
152
+ "as": 0.65,
153
+ "ih": 0.6
154
+ },
155
+ "decisions_count": 20
156
+ }
157
+ }
server/results/agent_metrics.json ADDED
@@ -0,0 +1,338 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "balances": {
3
+ "conservative_0": [
4
+ 0.49081387449515695,
5
+ 0.49343559279595983,
6
+ 0.48850458919327977,
7
+ 0.49146401935422707,
8
+ 0.4946749359413105,
9
+ 0.49636326754063087,
10
+ 0.4952029352774256,
11
+ 0.4940040393824217,
12
+ 0.4966972071867608,
13
+ 0.4980884768993407,
14
+ 0.49248891838223513,
15
+ 0.4912914260198051,
16
+ 0.4928912988916253,
17
+ 0.4947083713787377,
18
+ 0.49763670817523475,
19
+ 0.5001232423589304,
20
+ 0.49096027864859226,
21
+ 0.4897716498250345,
22
+ 0.492686327089706,
23
+ 0.49150872061921697
24
+ ],
25
+ "aggressive_1": [
26
+ 0.49479282257430807,
27
+ 0.49808816823059615,
28
+ 0.49778816823059613,
29
+ 0.49300299022864275,
30
+ 0.4927029902286427,
31
+ 0.495411537233485,
32
+ 0.4980393843454029,
33
+ 0.5012766927720629,
34
+ 0.5009766927720629,
35
+ 0.5035887816802933,
36
+ 0.5068851162502124,
37
+ 0.5065851162502124,
38
+ 0.5098062143226315,
39
+ 0.5129977676974148,
40
+ 0.5161602642380944,
41
+ 0.5189561722987545,
42
+ 0.5219453351696965,
43
+ 0.5250587164014742,
44
+ 0.5247587164014742,
45
+ 0.5277924652461742
46
+ ],
47
+ "balanced_2": [
48
+ 0.5088319063260909,
49
+ 0.5234635405742646,
50
+ 0.5140244465183659,
51
+ 0.5047625504836085,
52
+ 0.5204227495801004,
53
+ 0.5123727283326304,
54
+ 0.5286477931330813,
55
+ 0.5220918053639299,
56
+ 0.5380628598027662,
57
+ 0.5537254286259184,
58
+ 0.5459980422068574,
59
+ 0.5403249089008585,
60
+ 0.5335986427751808,
61
+ 0.5259763458300242,
62
+ 0.5167580159101,
63
+ 0.5333363337968263,
64
+ 0.5253121303690602,
65
+ 0.5385243079966603,
66
+ 0.5537207450841788,
67
+ 0.5403475830647143
68
+ ],
69
+ "adaptive_3": [
70
+ 0.49169999999999997,
71
+ 0.49139999999999995,
72
+ 0.4910999999999999,
73
+ 0.4907999999999999,
74
+ 0.4904999999999999,
75
+ 0.49309332273125955,
76
+ 0.49279332273125953,
77
+ 0.4924933227312595,
78
+ 0.49109371302682636,
79
+ 0.49079371302682634,
80
+ 0.4904937130268263,
81
+ 0.4901937130268263,
82
+ 0.48989371302682627,
83
+ 0.48959371302682625,
84
+ 0.4892937130268262,
85
+ 0.4889937130268262,
86
+ 0.4886937130268262,
87
+ 0.48839371302682616,
88
+ 0.48809371302682614,
89
+ 0.4866801928830522
90
+ ],
91
+ "cheater_4": [
92
+ 0.49169999999999997,
93
+ 0.49139999999999995,
94
+ 0.4910999999999999,
95
+ 0.4907999999999999,
96
+ 0.4904999999999999,
97
+ 0.49019999999999986,
98
+ 0.48989999999999984,
99
+ 0.4895999999999998,
100
+ 0.4892999999999998,
101
+ 0.48899999999999977,
102
+ 0.48869999999999975,
103
+ 0.4883999999999997,
104
+ 0.4880999999999997,
105
+ 0.4877999999999997,
106
+ 0.48749999999999966,
107
+ 0.48719999999999963,
108
+ 0.4868999999999996,
109
+ 0.4865999999999996,
110
+ 0.48629999999999957,
111
+ 0.48599999999999954
112
+ ]
113
+ },
114
+ "tiers": {
115
+ "conservative_0": [
116
+ 3,
117
+ 3,
118
+ 3,
119
+ 3,
120
+ 3,
121
+ 3,
122
+ 3,
123
+ 3,
124
+ 3,
125
+ 3,
126
+ 3,
127
+ 3,
128
+ 3,
129
+ 3,
130
+ 3,
131
+ 3,
132
+ 4,
133
+ 4,
134
+ 4,
135
+ 3
136
+ ],
137
+ "aggressive_1": [
138
+ 1,
139
+ 1,
140
+ 1,
141
+ 1,
142
+ 1,
143
+ 1,
144
+ 1,
145
+ 1,
146
+ 1,
147
+ 1,
148
+ 1,
149
+ 1,
150
+ 1,
151
+ 1,
152
+ 1,
153
+ 1,
154
+ 1,
155
+ 1,
156
+ 1,
157
+ 1
158
+ ],
159
+ "balanced_2": [
160
+ 2,
161
+ 2,
162
+ 2,
163
+ 2,
164
+ 2,
165
+ 2,
166
+ 2,
167
+ 2,
168
+ 2,
169
+ 2,
170
+ 2,
171
+ 2,
172
+ 2,
173
+ 2,
174
+ 2,
175
+ 2,
176
+ 2,
177
+ 2,
178
+ 2,
179
+ 1
180
+ ],
181
+ "adaptive_3": [
182
+ 1,
183
+ 1,
184
+ 1,
185
+ 1,
186
+ 1,
187
+ 1,
188
+ 1,
189
+ 1,
190
+ 1,
191
+ 1,
192
+ 1,
193
+ 1,
194
+ 1,
195
+ 1,
196
+ 1,
197
+ 1,
198
+ 1,
199
+ 1,
200
+ 1,
201
+ 1
202
+ ],
203
+ "cheater_4": [
204
+ 0,
205
+ 0,
206
+ 0,
207
+ 0,
208
+ 0,
209
+ 0,
210
+ 0,
211
+ 0,
212
+ 0,
213
+ 0,
214
+ 0,
215
+ 0,
216
+ 0,
217
+ 0,
218
+ 0,
219
+ 0,
220
+ 0,
221
+ 0,
222
+ 0,
223
+ 0
224
+ ]
225
+ },
226
+ "earnings": {
227
+ "conservative_0": [
228
+ 0.0,
229
+ 0.002921718300802926,
230
+ 0.006290714698122897,
231
+ 0.009550144859070195,
232
+ 0.013061061446153663,
233
+ 0.015049393045474056,
234
+ 0.015049393045474056,
235
+ 0.015049393045474056,
236
+ 0.018042560849813163,
237
+ 0.0197338305623931,
238
+ 0.022434272045287508,
239
+ 0.022434272045287508,
240
+ 0.02433414491710776,
241
+ 0.026451217404220177,
242
+ 0.029679554200717252,
243
+ 0.032466088384412846,
244
+ 0.032466088384412846,
245
+ 0.032466088384412846,
246
+ 0.035680765649084345,
247
+ 0.035680765649084345
248
+ ],
249
+ "aggressive_1": [
250
+ 0.003092822574308115,
251
+ 0.006688168230596231,
252
+ 0.006688168230596231,
253
+ 0.010202990228642848,
254
+ 0.010202990228642848,
255
+ 0.013211537233485171,
256
+ 0.016139384345403053,
257
+ 0.019676692772063034,
258
+ 0.019676692772063034,
259
+ 0.02258878168029342,
260
+ 0.026185116250212498,
261
+ 0.026185116250212498,
262
+ 0.029706214322631524,
263
+ 0.03319776769741471,
264
+ 0.03666026423809421,
265
+ 0.03975617229875429,
266
+ 0.04304533516969639,
267
+ 0.04645871640147403,
268
+ 0.04645871640147403,
269
+ 0.04979246524617394
270
+ ],
271
+ "balanced_2": [
272
+ 0.017131906326090882,
273
+ 0.03206354057426457,
274
+ 0.03206354057426457,
275
+ 0.03206354057426457,
276
+ 0.04802373967075634,
277
+ 0.04802373967075634,
278
+ 0.06459880447120725,
279
+ 0.06459880447120725,
280
+ 0.08086985891004358,
281
+ 0.09683242773319581,
282
+ 0.09683242773319581,
283
+ 0.09683242773319581,
284
+ 0.09683242773319581,
285
+ 0.09683242773319581,
286
+ 0.09683242773319581,
287
+ 0.11371074561992217,
288
+ 0.11371074561992217,
289
+ 0.12722292324752224,
290
+ 0.14271936033504076,
291
+ 0.14271936033504076
292
+ ],
293
+ "adaptive_3": [
294
+ 0.0,
295
+ 0.0,
296
+ 0.0,
297
+ 0.0,
298
+ 0.0,
299
+ 0.002893322731259686,
300
+ 0.002893322731259686,
301
+ 0.002893322731259686,
302
+ 0.002893322731259686,
303
+ 0.002893322731259686,
304
+ 0.002893322731259686,
305
+ 0.002893322731259686,
306
+ 0.002893322731259686,
307
+ 0.002893322731259686,
308
+ 0.002893322731259686,
309
+ 0.002893322731259686,
310
+ 0.002893322731259686,
311
+ 0.002893322731259686,
312
+ 0.002893322731259686,
313
+ 0.002893322731259686
314
+ ],
315
+ "cheater_4": [
316
+ 0.0,
317
+ 0.0,
318
+ 0.0,
319
+ 0.0,
320
+ 0.0,
321
+ 0.0,
322
+ 0.0,
323
+ 0.0,
324
+ 0.0,
325
+ 0.0,
326
+ 0.0,
327
+ 0.0,
328
+ 0.0,
329
+ 0.0,
330
+ 0.0,
331
+ 0.0,
332
+ 0.0,
333
+ 0.0,
334
+ 0.0,
335
+ 0.0
336
+ ]
337
+ }
338
+ }
server/results/economy_state.json ADDED
@@ -0,0 +1,144 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "timestamp": 20.0,
3
+ "config": {
4
+ "decay_rate": 0.005,
5
+ "ih_threshold": 0.45,
6
+ "initial_balance": 0.5
7
+ },
8
+ "agents": {
9
+ "agent_cf56e1c14e84": {
10
+ "agent_id": "agent_cf56e1c14e84",
11
+ "model_name": "conservative_0",
12
+ "architecture_hash": "8374b7c44322b61a",
13
+ "status": "active",
14
+ "current_tier": "T3",
15
+ "balance": 0.49150872061921697,
16
+ "total_earned": 0.035680765649084345,
17
+ "total_spent": 0.038000000000000034,
18
+ "total_penalties": 0.0061720450298669545,
19
+ "total_topups": 0.0,
20
+ "contracts_completed": 13,
21
+ "contracts_failed": 7,
22
+ "registration_time": 0.0,
23
+ "audit_cid": null,
24
+ "wallet_address": null,
25
+ "robustness": {
26
+ "cc": 0.9255496218337828,
27
+ "er": 0.7985244829016913,
28
+ "as": 0.772250481405939,
29
+ "ih": 0.8472803268022182
30
+ }
31
+ },
32
+ "agent_28f8af722342": {
33
+ "agent_id": "agent_28f8af722342",
34
+ "model_name": "aggressive_1",
35
+ "architecture_hash": "06cf6cd1e0157f7e",
36
+ "status": "active",
37
+ "current_tier": "T1",
38
+ "balance": 0.5277924652461742,
39
+ "total_earned": 0.04979246524617394,
40
+ "total_spent": 0.022000000000000026,
41
+ "total_penalties": 0.0,
42
+ "total_topups": 0.0,
43
+ "contracts_completed": 15,
44
+ "contracts_failed": 0,
45
+ "registration_time": 0.0,
46
+ "audit_cid": null,
47
+ "wallet_address": null,
48
+ "robustness": {
49
+ "cc": 0.4237871475106476,
50
+ "er": 0.43773895056453505,
51
+ "as": 0.34136116942599515,
52
+ "ih": 0.6714820583560304
53
+ }
54
+ },
55
+ "agent_90781acc6dc0": {
56
+ "agent_id": "agent_90781acc6dc0",
57
+ "model_name": "balanced_2",
58
+ "architecture_hash": "ea00df5aae26a51e",
59
+ "status": "active",
60
+ "current_tier": "T1",
61
+ "balance": 0.5403475830647143,
62
+ "total_earned": 0.14271936033504076,
63
+ "total_spent": 0.022,
64
+ "total_penalties": 0.08037177727032704,
65
+ "total_topups": 0.0,
66
+ "contracts_completed": 9,
67
+ "contracts_failed": 11,
68
+ "registration_time": 0.0,
69
+ "audit_cid": null,
70
+ "wallet_address": null,
71
+ "robustness": {
72
+ "cc": 0.6509882462532015,
73
+ "er": 0.529896012603673,
74
+ "as": 0.4386817356571334,
75
+ "ih": 0.8320949702616935
76
+ }
77
+ },
78
+ "agent_33d2e87d3579": {
79
+ "agent_id": "agent_33d2e87d3579",
80
+ "model_name": "adaptive_3",
81
+ "architecture_hash": "b191a86339607a12",
82
+ "status": "active",
83
+ "current_tier": "T1",
84
+ "balance": 0.4866801928830522,
85
+ "total_earned": 0.002893322731259686,
86
+ "total_spent": 0.013999999999999999,
87
+ "total_penalties": 0.0022131298482070774,
88
+ "total_topups": 0.0,
89
+ "contracts_completed": 1,
90
+ "contracts_failed": 2,
91
+ "registration_time": 0.0,
92
+ "audit_cid": null,
93
+ "wallet_address": null,
94
+ "robustness": {
95
+ "cc": 0.5696990952039601,
96
+ "er": 0.5033152153233149,
97
+ "as": 0.42785035192965537,
98
+ "ih": 0.7695601289753685
99
+ }
100
+ },
101
+ "agent_85b2a0234c59": {
102
+ "agent_id": "agent_85b2a0234c59",
103
+ "model_name": "cheater_4",
104
+ "architecture_hash": "73482227d1d7b2b3",
105
+ "status": "active",
106
+ "current_tier": "T0",
107
+ "balance": 0.48599999999999954,
108
+ "total_earned": 0.0,
109
+ "total_spent": 0.013999999999999999,
110
+ "total_penalties": 0.0,
111
+ "total_topups": 0.0,
112
+ "contracts_completed": 0,
113
+ "contracts_failed": 0,
114
+ "registration_time": 0.0,
115
+ "audit_cid": null,
116
+ "wallet_address": null,
117
+ "robustness": {
118
+ "cc": 0.707390265856336,
119
+ "er": 0.2893324248161214,
120
+ "as": 0.6512497059171015,
121
+ "ih": 0.5968103011868765
122
+ }
123
+ }
124
+ },
125
+ "contracts": {
126
+ "total_contracts": 220,
127
+ "status_distribution": {
128
+ "failed": 20,
129
+ "open": 162,
130
+ "completed": 38
131
+ },
132
+ "total_rewards_paid": 0.23108591396155878,
133
+ "total_penalties_collected": 0.08875695214840107,
134
+ "total_escrowed": 54.72028111841184,
135
+ "active_exposures": {
136
+ "agent_cf56e1c14e84": 0,
137
+ "agent_28f8af722342": 0,
138
+ "agent_90781acc6dc0": 0,
139
+ "agent_33d2e87d3579": 0
140
+ }
141
+ },
142
+ "aggregate_safety": 0.7608347665972391,
143
+ "total_test_eth_topups": 0.0
144
+ }
server/results/protocol_events.json ADDED
@@ -0,0 +1 @@
 
 
1
+ []
server/results/strategy_summary.json ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "survival": {
3
+ "conservative": 1,
4
+ "aggressive": 1,
5
+ "balanced": 1,
6
+ "adaptive": 1,
7
+ "cheater": 1
8
+ },
9
+ "total_earned": {
10
+ "conservative": 0.035680765649084345,
11
+ "aggressive": 0.04979246524617394,
12
+ "balanced": 0.14271936033504076,
13
+ "adaptive": 0.002893322731259686,
14
+ "cheater": 0.0
15
+ },
16
+ "final_tier": {
17
+ "conservative": 3,
18
+ "aggressive": 1,
19
+ "balanced": 1,
20
+ "adaptive": 1,
21
+ "cheater": 0
22
+ }
23
+ }
server/results/task_results.json ADDED
@@ -0,0 +1,1055 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "agent": "conservative_0",
4
+ "task_id": "contract_a762c024c23b",
5
+ "tier": "T1",
6
+ "domain": "data_labeling",
7
+ "proof_cid": "0xe3788e9a12befa04d89985ec593eebf7",
8
+ "verification": {
9
+ "overall_pass": false,
10
+ "constraints_passed": [],
11
+ "constraints_failed": [
12
+ "format_compliance",
13
+ "completeness"
14
+ ]
15
+ },
16
+ "settlement": {
17
+ "reward": 0.0,
18
+ "penalty": 0.0008861255048430355
19
+ },
20
+ "output_preview": "Synthetic execution of contract_a762c024c23b: FAILURE"
21
+ },
22
+ {
23
+ "agent": "aggressive_1",
24
+ "task_id": "contract_6a04b2ca71e6",
25
+ "tier": "T1",
26
+ "domain": "summarization",
27
+ "proof_cid": "0x0de9d6b5ac920e35a143b7de2828138d",
28
+ "verification": {
29
+ "overall_pass": true,
30
+ "constraints_passed": [],
31
+ "constraints_failed": []
32
+ },
33
+ "settlement": {
34
+ "reward": 0.003092822574308115,
35
+ "penalty": 0.0
36
+ },
37
+ "output_preview": "Synthetic execution of contract_6a04b2ca71e6: SUCCESS"
38
+ },
39
+ {
40
+ "agent": "balanced_2",
41
+ "task_id": "contract_e254b22fba41",
42
+ "tier": "T2",
43
+ "domain": "analysis",
44
+ "proof_cid": "0x72bb23b6fddbea88e0dd14a2e40f95cb",
45
+ "verification": {
46
+ "overall_pass": true,
47
+ "constraints_passed": [],
48
+ "constraints_failed": []
49
+ },
50
+ "settlement": {
51
+ "reward": 0.017131906326090882,
52
+ "penalty": 0.0
53
+ },
54
+ "output_preview": "Synthetic execution of contract_e254b22fba41: SUCCESS"
55
+ },
56
+ {
57
+ "agent": "conservative_0",
58
+ "task_id": "contract_9d11e654a1af",
59
+ "tier": "T1",
60
+ "domain": "summarization",
61
+ "proof_cid": "0x10f8962cb9e6f69fbb4c32c81eaf51fa",
62
+ "verification": {
63
+ "overall_pass": true,
64
+ "constraints_passed": [],
65
+ "constraints_failed": []
66
+ },
67
+ "settlement": {
68
+ "reward": 0.002921718300802926,
69
+ "penalty": 0.0
70
+ },
71
+ "output_preview": "Synthetic execution of contract_9d11e654a1af: SUCCESS"
72
+ },
73
+ {
74
+ "agent": "aggressive_1",
75
+ "task_id": "contract_ea289c611e45",
76
+ "tier": "T1",
77
+ "domain": "summarization",
78
+ "proof_cid": "0xafdbb688086437a054dda6aafea164d5",
79
+ "verification": {
80
+ "overall_pass": true,
81
+ "constraints_passed": [],
82
+ "constraints_failed": []
83
+ },
84
+ "settlement": {
85
+ "reward": 0.003595345656288116,
86
+ "penalty": 0.0
87
+ },
88
+ "output_preview": "Synthetic execution of contract_ea289c611e45: SUCCESS"
89
+ },
90
+ {
91
+ "agent": "balanced_2",
92
+ "task_id": "contract_b97498b9138f",
93
+ "tier": "T2",
94
+ "domain": "analysis",
95
+ "proof_cid": "0xfd719cb4f2e7cde733c8bdfbb1ba22a3",
96
+ "verification": {
97
+ "overall_pass": true,
98
+ "constraints_passed": [],
99
+ "constraints_failed": []
100
+ },
101
+ "settlement": {
102
+ "reward": 0.014931634248173686,
103
+ "penalty": 0.0
104
+ },
105
+ "output_preview": "Synthetic execution of contract_b97498b9138f: SUCCESS"
106
+ },
107
+ {
108
+ "agent": "conservative_0",
109
+ "task_id": "contract_8f83ccb818b2",
110
+ "tier": "T1",
111
+ "domain": "summarization",
112
+ "proof_cid": "0x881a1b4fd780c04cf4eb02e07349fa16",
113
+ "verification": {
114
+ "overall_pass": true,
115
+ "constraints_passed": [],
116
+ "constraints_failed": []
117
+ },
118
+ "settlement": {
119
+ "reward": 0.0033689963973199716,
120
+ "penalty": 0.0
121
+ },
122
+ "output_preview": "Synthetic execution of contract_8f83ccb818b2: SUCCESS"
123
+ },
124
+ {
125
+ "agent": "balanced_2",
126
+ "task_id": "contract_d6f27d89c8ef",
127
+ "tier": "T2",
128
+ "domain": "analysis",
129
+ "proof_cid": "0xab250c0c87d8aab0cd891579832c533e",
130
+ "verification": {
131
+ "overall_pass": false,
132
+ "constraints_passed": [],
133
+ "constraints_failed": [
134
+ "format_compliance",
135
+ "accuracy",
136
+ "completeness"
137
+ ]
138
+ },
139
+ "settlement": {
140
+ "reward": 0.0,
141
+ "penalty": 0.009139094055898793
142
+ },
143
+ "output_preview": "Synthetic execution of contract_d6f27d89c8ef: FAILURE"
144
+ },
145
+ {
146
+ "agent": "conservative_0",
147
+ "task_id": "contract_26627eeeea32",
148
+ "tier": "T1",
149
+ "domain": "summarization",
150
+ "proof_cid": "0x393ded3cb91de8d0659b65697eb40420",
151
+ "verification": {
152
+ "overall_pass": true,
153
+ "constraints_passed": [],
154
+ "constraints_failed": []
155
+ },
156
+ "settlement": {
157
+ "reward": 0.003259430160947298,
158
+ "penalty": 0.0
159
+ },
160
+ "output_preview": "Synthetic execution of contract_26627eeeea32: SUCCESS"
161
+ },
162
+ {
163
+ "agent": "aggressive_1",
164
+ "task_id": "contract_5044b25e64ba",
165
+ "tier": "T1",
166
+ "domain": "summarization",
167
+ "proof_cid": "0x717dc9278dbae3187d2accd765517646",
168
+ "verification": {
169
+ "overall_pass": true,
170
+ "constraints_passed": [],
171
+ "constraints_failed": []
172
+ },
173
+ "settlement": {
174
+ "reward": 0.003514821998046617,
175
+ "penalty": 0.0
176
+ },
177
+ "output_preview": "Synthetic execution of contract_5044b25e64ba: SUCCESS"
178
+ },
179
+ {
180
+ "agent": "balanced_2",
181
+ "task_id": "contract_8a5344725b1b",
182
+ "tier": "T2",
183
+ "domain": "analysis",
184
+ "proof_cid": "0xb44a4ca6377c11611d8ecf66ca433482",
185
+ "verification": {
186
+ "overall_pass": false,
187
+ "constraints_passed": [],
188
+ "constraints_failed": [
189
+ "format_compliance",
190
+ "accuracy",
191
+ "completeness"
192
+ ]
193
+ },
194
+ "settlement": {
195
+ "reward": 0.0,
196
+ "penalty": 0.008961896034757435
197
+ },
198
+ "output_preview": "Synthetic execution of contract_8a5344725b1b: FAILURE"
199
+ },
200
+ {
201
+ "agent": "conservative_0",
202
+ "task_id": "contract_da1cc69e5fcf",
203
+ "tier": "T1",
204
+ "domain": "summarization",
205
+ "proof_cid": "0x3b18ea0666a093e5910678a11a119a2a",
206
+ "verification": {
207
+ "overall_pass": true,
208
+ "constraints_passed": [],
209
+ "constraints_failed": []
210
+ },
211
+ "settlement": {
212
+ "reward": 0.003510916587083468,
213
+ "penalty": 0.0
214
+ },
215
+ "output_preview": "Synthetic execution of contract_da1cc69e5fcf: SUCCESS"
216
+ },
217
+ {
218
+ "agent": "balanced_2",
219
+ "task_id": "contract_c24512d81eb2",
220
+ "tier": "T2",
221
+ "domain": "analysis",
222
+ "proof_cid": "0xa207087fb8f4d91e65a3246c12d7b671",
223
+ "verification": {
224
+ "overall_pass": true,
225
+ "constraints_passed": [],
226
+ "constraints_failed": []
227
+ },
228
+ "settlement": {
229
+ "reward": 0.015960199096491774,
230
+ "penalty": 0.0
231
+ },
232
+ "output_preview": "Synthetic execution of contract_c24512d81eb2: SUCCESS"
233
+ },
234
+ {
235
+ "agent": "conservative_0",
236
+ "task_id": "contract_774f2afce35d",
237
+ "tier": "T1",
238
+ "domain": "data_labeling",
239
+ "proof_cid": "0xf2c334d0b1eb999eeea1c89175518109",
240
+ "verification": {
241
+ "overall_pass": true,
242
+ "constraints_passed": [],
243
+ "constraints_failed": []
244
+ },
245
+ "settlement": {
246
+ "reward": 0.0019883315993203927,
247
+ "penalty": 0.0
248
+ },
249
+ "output_preview": "Synthetic execution of contract_774f2afce35d: SUCCESS"
250
+ },
251
+ {
252
+ "agent": "aggressive_1",
253
+ "task_id": "contract_30bae7d33b63",
254
+ "tier": "T1",
255
+ "domain": "summarization",
256
+ "proof_cid": "0x7b266462496c6df47778d5473897a8ec",
257
+ "verification": {
258
+ "overall_pass": true,
259
+ "constraints_passed": [],
260
+ "constraints_failed": []
261
+ },
262
+ "settlement": {
263
+ "reward": 0.003008547004842323,
264
+ "penalty": 0.0
265
+ },
266
+ "output_preview": "Synthetic execution of contract_30bae7d33b63: SUCCESS"
267
+ },
268
+ {
269
+ "agent": "balanced_2",
270
+ "task_id": "contract_267621db5452",
271
+ "tier": "T2",
272
+ "domain": "analysis",
273
+ "proof_cid": "0xd75813e6bbdd205ea4900a153f4f0ba2",
274
+ "verification": {
275
+ "overall_pass": false,
276
+ "constraints_passed": [],
277
+ "constraints_failed": [
278
+ "format_compliance",
279
+ "accuracy",
280
+ "completeness"
281
+ ]
282
+ },
283
+ "settlement": {
284
+ "reward": 0.0,
285
+ "penalty": 0.007750021247470094
286
+ },
287
+ "output_preview": "Synthetic execution of contract_267621db5452: FAILURE"
288
+ },
289
+ {
290
+ "agent": "adaptive_3",
291
+ "task_id": "contract_832d70fc55c6",
292
+ "tier": "T1",
293
+ "domain": "summarization",
294
+ "proof_cid": "0x7e0aa4424e9a6ac90ced69357f454c9c",
295
+ "verification": {
296
+ "overall_pass": true,
297
+ "constraints_passed": [],
298
+ "constraints_failed": []
299
+ },
300
+ "settlement": {
301
+ "reward": 0.002893322731259686,
302
+ "penalty": 0.0
303
+ },
304
+ "output_preview": "Synthetic execution of contract_832d70fc55c6: SUCCESS"
305
+ },
306
+ {
307
+ "agent": "conservative_0",
308
+ "task_id": "contract_abb67f7418ec",
309
+ "tier": "T1",
310
+ "domain": "summarization",
311
+ "proof_cid": "0xb80745d02c2cae2b14de66af7ced8639",
312
+ "verification": {
313
+ "overall_pass": false,
314
+ "constraints_passed": [],
315
+ "constraints_failed": [
316
+ "length_constraint",
317
+ "accuracy"
318
+ ]
319
+ },
320
+ "settlement": {
321
+ "reward": 0.0,
322
+ "penalty": 0.0008603322632052347
323
+ },
324
+ "output_preview": "Synthetic execution of contract_abb67f7418ec: FAILURE"
325
+ },
326
+ {
327
+ "agent": "aggressive_1",
328
+ "task_id": "contract_9b870ef54e9d",
329
+ "tier": "T1",
330
+ "domain": "summarization",
331
+ "proof_cid": "0x5ae50e8a2c6560a194cc21cefcb442b0",
332
+ "verification": {
333
+ "overall_pass": true,
334
+ "constraints_passed": [],
335
+ "constraints_failed": []
336
+ },
337
+ "settlement": {
338
+ "reward": 0.0029278471119178833,
339
+ "penalty": 0.0
340
+ },
341
+ "output_preview": "Synthetic execution of contract_9b870ef54e9d: SUCCESS"
342
+ },
343
+ {
344
+ "agent": "balanced_2",
345
+ "task_id": "contract_585e86184d7e",
346
+ "tier": "T2",
347
+ "domain": "analysis",
348
+ "proof_cid": "0x8a5fc294c236b93ecac0d4c870a4c398",
349
+ "verification": {
350
+ "overall_pass": true,
351
+ "constraints_passed": [],
352
+ "constraints_failed": []
353
+ },
354
+ "settlement": {
355
+ "reward": 0.016575064800450907,
356
+ "penalty": 0.0
357
+ },
358
+ "output_preview": "Synthetic execution of contract_585e86184d7e: SUCCESS"
359
+ },
360
+ {
361
+ "agent": "conservative_0",
362
+ "task_id": "contract_8405e50a5ebc",
363
+ "tier": "T1",
364
+ "domain": "data_labeling",
365
+ "proof_cid": "0x5f3870dc8ac36788dd6e4a60515f9fe3",
366
+ "verification": {
367
+ "overall_pass": false,
368
+ "constraints_passed": [],
369
+ "constraints_failed": [
370
+ "format_compliance",
371
+ "completeness"
372
+ ]
373
+ },
374
+ "settlement": {
375
+ "reward": 0.0,
376
+ "penalty": 0.0008988958950038638
377
+ },
378
+ "output_preview": "Synthetic execution of contract_8405e50a5ebc: FAILURE"
379
+ },
380
+ {
381
+ "agent": "aggressive_1",
382
+ "task_id": "contract_60e2abeab26f",
383
+ "tier": "T1",
384
+ "domain": "summarization",
385
+ "proof_cid": "0xb6d703438a944c5b6912fa5495ea950f",
386
+ "verification": {
387
+ "overall_pass": true,
388
+ "constraints_passed": [],
389
+ "constraints_failed": []
390
+ },
391
+ "settlement": {
392
+ "reward": 0.003537308426659982,
393
+ "penalty": 0.0
394
+ },
395
+ "output_preview": "Synthetic execution of contract_60e2abeab26f: SUCCESS"
396
+ },
397
+ {
398
+ "agent": "balanced_2",
399
+ "task_id": "contract_911afcedcc61",
400
+ "tier": "T2",
401
+ "domain": "translation",
402
+ "proof_cid": "0xd77412a07afea5bc9b32c6b4c9654928",
403
+ "verification": {
404
+ "overall_pass": false,
405
+ "constraints_passed": [],
406
+ "constraints_failed": [
407
+ "accuracy",
408
+ "terminology_consistency"
409
+ ]
410
+ },
411
+ "settlement": {
412
+ "reward": 0.0,
413
+ "penalty": 0.00625598776915143
414
+ },
415
+ "output_preview": "Synthetic execution of contract_911afcedcc61: FAILURE"
416
+ },
417
+ {
418
+ "agent": "conservative_0",
419
+ "task_id": "contract_47a2465077bf",
420
+ "tier": "T1",
421
+ "domain": "summarization",
422
+ "proof_cid": "0x962cc7636be73ebf96af5019fe8d96d0",
423
+ "verification": {
424
+ "overall_pass": true,
425
+ "constraints_passed": [],
426
+ "constraints_failed": []
427
+ },
428
+ "settlement": {
429
+ "reward": 0.0029931678043391063,
430
+ "penalty": 0.0
431
+ },
432
+ "output_preview": "Synthetic execution of contract_47a2465077bf: SUCCESS"
433
+ },
434
+ {
435
+ "agent": "balanced_2",
436
+ "task_id": "contract_74c517e69b09",
437
+ "tier": "T2",
438
+ "domain": "analysis",
439
+ "proof_cid": "0x715788df4865bb682a9f2df8d4143c76",
440
+ "verification": {
441
+ "overall_pass": true,
442
+ "constraints_passed": [],
443
+ "constraints_failed": []
444
+ },
445
+ "settlement": {
446
+ "reward": 0.016271054438836335,
447
+ "penalty": 0.0
448
+ },
449
+ "output_preview": "Synthetic execution of contract_74c517e69b09: SUCCESS"
450
+ },
451
+ {
452
+ "agent": "adaptive_3",
453
+ "task_id": "contract_28e3ff49574c",
454
+ "tier": "T1",
455
+ "domain": "summarization",
456
+ "proof_cid": "0xac2b041fa9422c6c6fad4780f4ec5aba",
457
+ "verification": {
458
+ "overall_pass": false,
459
+ "constraints_passed": [],
460
+ "constraints_failed": [
461
+ "length_constraint",
462
+ "accuracy"
463
+ ]
464
+ },
465
+ "settlement": {
466
+ "reward": 0.0,
467
+ "penalty": 0.0010996097044331404
468
+ },
469
+ "output_preview": "Synthetic execution of contract_28e3ff49574c: FAILURE"
470
+ },
471
+ {
472
+ "agent": "conservative_0",
473
+ "task_id": "contract_292b64d4eb34",
474
+ "tier": "T1",
475
+ "domain": "data_labeling",
476
+ "proof_cid": "0x0d9859a99ed15eab50872a8511ef3f4a",
477
+ "verification": {
478
+ "overall_pass": true,
479
+ "constraints_passed": [],
480
+ "constraints_failed": []
481
+ },
482
+ "settlement": {
483
+ "reward": 0.0016912697125799397,
484
+ "penalty": 0.0
485
+ },
486
+ "output_preview": "Synthetic execution of contract_292b64d4eb34: SUCCESS"
487
+ },
488
+ {
489
+ "agent": "aggressive_1",
490
+ "task_id": "contract_8bacab0d4544",
491
+ "tier": "T1",
492
+ "domain": "summarization",
493
+ "proof_cid": "0xaebf575423009fa0bb556d8f07851973",
494
+ "verification": {
495
+ "overall_pass": true,
496
+ "constraints_passed": [],
497
+ "constraints_failed": []
498
+ },
499
+ "settlement": {
500
+ "reward": 0.0029120889082303865,
501
+ "penalty": 0.0
502
+ },
503
+ "output_preview": "Synthetic execution of contract_8bacab0d4544: SUCCESS"
504
+ },
505
+ {
506
+ "agent": "balanced_2",
507
+ "task_id": "contract_12555ae63e88",
508
+ "tier": "T2",
509
+ "domain": "analysis",
510
+ "proof_cid": "0x3a02751becc8df4c6ca5035e784ba75b",
511
+ "verification": {
512
+ "overall_pass": true,
513
+ "constraints_passed": [],
514
+ "constraints_failed": []
515
+ },
516
+ "settlement": {
517
+ "reward": 0.01596256882315222,
518
+ "penalty": 0.0
519
+ },
520
+ "output_preview": "Synthetic execution of contract_12555ae63e88: SUCCESS"
521
+ },
522
+ {
523
+ "agent": "conservative_0",
524
+ "task_id": "contract_fa617f94d520",
525
+ "tier": "T1",
526
+ "domain": "summarization",
527
+ "proof_cid": "0x755485295416028ccd6c26d151d75609",
528
+ "verification": {
529
+ "overall_pass": true,
530
+ "constraints_passed": [],
531
+ "constraints_failed": []
532
+ },
533
+ "settlement": {
534
+ "reward": 0.002700441482894405,
535
+ "penalty": 0.0
536
+ },
537
+ "output_preview": "Synthetic execution of contract_fa617f94d520: SUCCESS"
538
+ },
539
+ {
540
+ "agent": "aggressive_1",
541
+ "task_id": "contract_88fd92bae684",
542
+ "tier": "T1",
543
+ "domain": "summarization",
544
+ "proof_cid": "0xa17f835b71f8220f03516b1a3b4c7f5a",
545
+ "verification": {
546
+ "overall_pass": true,
547
+ "constraints_passed": [],
548
+ "constraints_failed": []
549
+ },
550
+ "settlement": {
551
+ "reward": 0.0035963345699190756,
552
+ "penalty": 0.0
553
+ },
554
+ "output_preview": "Synthetic execution of contract_88fd92bae684: SUCCESS"
555
+ },
556
+ {
557
+ "agent": "balanced_2",
558
+ "task_id": "contract_58a6cf8906eb",
559
+ "tier": "T2",
560
+ "domain": "analysis",
561
+ "proof_cid": "0xdb3a9312247f8d1dbf9324f70165bbd2",
562
+ "verification": {
563
+ "overall_pass": false,
564
+ "constraints_passed": [],
565
+ "constraints_failed": [
566
+ "format_compliance",
567
+ "accuracy",
568
+ "completeness"
569
+ ]
570
+ },
571
+ "settlement": {
572
+ "reward": 0.0,
573
+ "penalty": 0.007427386419061029
574
+ },
575
+ "output_preview": "Synthetic execution of contract_58a6cf8906eb: FAILURE"
576
+ },
577
+ {
578
+ "agent": "conservative_0",
579
+ "task_id": "contract_0d0c5c4043dd",
580
+ "tier": "T1",
581
+ "domain": "summarization",
582
+ "proof_cid": "0xac1b7ebde0853facfcccd71df1e6c12a",
583
+ "verification": {
584
+ "overall_pass": false,
585
+ "constraints_passed": [],
586
+ "constraints_failed": [
587
+ "length_constraint",
588
+ "accuracy"
589
+ ]
590
+ },
591
+ "settlement": {
592
+ "reward": 0.0,
593
+ "penalty": 0.0008974923624300554
594
+ },
595
+ "output_preview": "Synthetic execution of contract_0d0c5c4043dd: FAILURE"
596
+ },
597
+ {
598
+ "agent": "balanced_2",
599
+ "task_id": "contract_2f565829d3bb",
600
+ "tier": "T2",
601
+ "domain": "translation",
602
+ "proof_cid": "0xf359c2dae17fab61b523d962e160893f",
603
+ "verification": {
604
+ "overall_pass": false,
605
+ "constraints_passed": [],
606
+ "constraints_failed": [
607
+ "accuracy",
608
+ "terminology_consistency"
609
+ ]
610
+ },
611
+ "settlement": {
612
+ "reward": 0.0,
613
+ "penalty": 0.005373133305998998
614
+ },
615
+ "output_preview": "Synthetic execution of contract_2f565829d3bb: FAILURE"
616
+ },
617
+ {
618
+ "agent": "conservative_0",
619
+ "task_id": "contract_eee77bdc994a",
620
+ "tier": "T1",
621
+ "domain": "data_labeling",
622
+ "proof_cid": "0x96c179c1294253c64178491b45cd2965",
623
+ "verification": {
624
+ "overall_pass": true,
625
+ "constraints_passed": [],
626
+ "constraints_failed": []
627
+ },
628
+ "settlement": {
629
+ "reward": 0.0018998728718202518,
630
+ "penalty": 0.0
631
+ },
632
+ "output_preview": "Synthetic execution of contract_eee77bdc994a: SUCCESS"
633
+ },
634
+ {
635
+ "agent": "aggressive_1",
636
+ "task_id": "contract_ee6438a35eb2",
637
+ "tier": "T1",
638
+ "domain": "summarization",
639
+ "proof_cid": "0x66ec2a457304d380e1219388b1d3f0cc",
640
+ "verification": {
641
+ "overall_pass": true,
642
+ "constraints_passed": [],
643
+ "constraints_failed": []
644
+ },
645
+ "settlement": {
646
+ "reward": 0.0035210980724190246,
647
+ "penalty": 0.0
648
+ },
649
+ "output_preview": "Synthetic execution of contract_ee6438a35eb2: SUCCESS"
650
+ },
651
+ {
652
+ "agent": "balanced_2",
653
+ "task_id": "contract_2df338eb83dc",
654
+ "tier": "T2",
655
+ "domain": "translation",
656
+ "proof_cid": "0x687de0274a2f0b6e66f5c47459872e19",
657
+ "verification": {
658
+ "overall_pass": false,
659
+ "constraints_passed": [],
660
+ "constraints_failed": [
661
+ "accuracy",
662
+ "terminology_consistency"
663
+ ]
664
+ },
665
+ "settlement": {
666
+ "reward": 0.0,
667
+ "penalty": 0.00642626612567779
668
+ },
669
+ "output_preview": "Synthetic execution of contract_2df338eb83dc: FAILURE"
670
+ },
671
+ {
672
+ "agent": "conservative_0",
673
+ "task_id": "contract_3971377c6105",
674
+ "tier": "T1",
675
+ "domain": "data_labeling",
676
+ "proof_cid": "0xa375d7b10b82ba39f65dc4a0f4f717d5",
677
+ "verification": {
678
+ "overall_pass": true,
679
+ "constraints_passed": [],
680
+ "constraints_failed": []
681
+ },
682
+ "settlement": {
683
+ "reward": 0.0021170724871124183,
684
+ "penalty": 0.0
685
+ },
686
+ "output_preview": "Synthetic execution of contract_3971377c6105: SUCCESS"
687
+ },
688
+ {
689
+ "agent": "aggressive_1",
690
+ "task_id": "contract_2637144a151c",
691
+ "tier": "T1",
692
+ "domain": "summarization",
693
+ "proof_cid": "0x9f244b6e27a1f78ab7a2dc20cb571a22",
694
+ "verification": {
695
+ "overall_pass": true,
696
+ "constraints_passed": [],
697
+ "constraints_failed": []
698
+ },
699
+ "settlement": {
700
+ "reward": 0.003491553374783192,
701
+ "penalty": 0.0
702
+ },
703
+ "output_preview": "Synthetic execution of contract_2637144a151c: SUCCESS"
704
+ },
705
+ {
706
+ "agent": "balanced_2",
707
+ "task_id": "contract_9205ea050b7a",
708
+ "tier": "T2",
709
+ "domain": "analysis",
710
+ "proof_cid": "0x6da58ecd76cd5116a3f1e1b7c49e1661",
711
+ "verification": {
712
+ "overall_pass": false,
713
+ "constraints_passed": [],
714
+ "constraints_failed": [
715
+ "format_compliance",
716
+ "accuracy",
717
+ "completeness"
718
+ ]
719
+ },
720
+ "settlement": {
721
+ "reward": 0.0,
722
+ "penalty": 0.0073222969451566585
723
+ },
724
+ "output_preview": "Synthetic execution of contract_9205ea050b7a: FAILURE"
725
+ },
726
+ {
727
+ "agent": "conservative_0",
728
+ "task_id": "contract_0f3319c6939c",
729
+ "tier": "T1",
730
+ "domain": "summarization",
731
+ "proof_cid": "0x9f998a50c6209104f43a76d6d0e42916",
732
+ "verification": {
733
+ "overall_pass": true,
734
+ "constraints_passed": [],
735
+ "constraints_failed": []
736
+ },
737
+ "settlement": {
738
+ "reward": 0.0032283367964970762,
739
+ "penalty": 0.0
740
+ },
741
+ "output_preview": "Synthetic execution of contract_0f3319c6939c: SUCCESS"
742
+ },
743
+ {
744
+ "agent": "aggressive_1",
745
+ "task_id": "contract_d287fc801e29",
746
+ "tier": "T1",
747
+ "domain": "summarization",
748
+ "proof_cid": "0xdc5f0585641df466fcf27a2d3a0b85ea",
749
+ "verification": {
750
+ "overall_pass": true,
751
+ "constraints_passed": [],
752
+ "constraints_failed": []
753
+ },
754
+ "settlement": {
755
+ "reward": 0.003462496540679495,
756
+ "penalty": 0.0
757
+ },
758
+ "output_preview": "Synthetic execution of contract_d287fc801e29: SUCCESS"
759
+ },
760
+ {
761
+ "agent": "balanced_2",
762
+ "task_id": "contract_5a3f9486d106",
763
+ "tier": "T2",
764
+ "domain": "analysis",
765
+ "proof_cid": "0x3640e5366d7bb6e6c8effdb253495184",
766
+ "verification": {
767
+ "overall_pass": false,
768
+ "constraints_passed": [],
769
+ "constraints_failed": [
770
+ "format_compliance",
771
+ "accuracy",
772
+ "completeness"
773
+ ]
774
+ },
775
+ "settlement": {
776
+ "reward": 0.0,
777
+ "penalty": 0.008918329919924212
778
+ },
779
+ "output_preview": "Synthetic execution of contract_5a3f9486d106: FAILURE"
780
+ },
781
+ {
782
+ "agent": "conservative_0",
783
+ "task_id": "contract_8e0f8276ec96",
784
+ "tier": "T1",
785
+ "domain": "summarization",
786
+ "proof_cid": "0xd36516b07eb0023c9c137bf47109f7c5",
787
+ "verification": {
788
+ "overall_pass": true,
789
+ "constraints_passed": [],
790
+ "constraints_failed": []
791
+ },
792
+ "settlement": {
793
+ "reward": 0.0027865341836955915,
794
+ "penalty": 0.0
795
+ },
796
+ "output_preview": "Synthetic execution of contract_8e0f8276ec96: SUCCESS"
797
+ },
798
+ {
799
+ "agent": "aggressive_1",
800
+ "task_id": "contract_067db520271d",
801
+ "tier": "T1",
802
+ "domain": "summarization",
803
+ "proof_cid": "0xf7277144f75d7e1506a28bd20afaab7c",
804
+ "verification": {
805
+ "overall_pass": true,
806
+ "constraints_passed": [],
807
+ "constraints_failed": []
808
+ },
809
+ "settlement": {
810
+ "reward": 0.003095908060660085,
811
+ "penalty": 0.0
812
+ },
813
+ "output_preview": "Synthetic execution of contract_067db520271d: SUCCESS"
814
+ },
815
+ {
816
+ "agent": "balanced_2",
817
+ "task_id": "contract_36f16cd4c323",
818
+ "tier": "T2",
819
+ "domain": "analysis",
820
+ "proof_cid": "0x629030341cbe0e7a5f84c0a4313b7c33",
821
+ "verification": {
822
+ "overall_pass": true,
823
+ "constraints_passed": [],
824
+ "constraints_failed": []
825
+ },
826
+ "settlement": {
827
+ "reward": 0.016878317886726356,
828
+ "penalty": 0.0
829
+ },
830
+ "output_preview": "Synthetic execution of contract_36f16cd4c323: SUCCESS"
831
+ },
832
+ {
833
+ "agent": "conservative_0",
834
+ "task_id": "contract_6dd7a4c98a14",
835
+ "tier": "T1",
836
+ "domain": "data_labeling",
837
+ "proof_cid": "0x37a294db63ae2a5c16e66e3558d201f4",
838
+ "verification": {
839
+ "overall_pass": false,
840
+ "constraints_passed": [],
841
+ "constraints_failed": [
842
+ "format_compliance",
843
+ "completeness"
844
+ ]
845
+ },
846
+ "settlement": {
847
+ "reward": 0.0,
848
+ "penalty": 0.0008629637103380356
849
+ },
850
+ "output_preview": "Synthetic execution of contract_6dd7a4c98a14: FAILURE"
851
+ },
852
+ {
853
+ "agent": "aggressive_1",
854
+ "task_id": "contract_58271e8e7624",
855
+ "tier": "T1",
856
+ "domain": "summarization",
857
+ "proof_cid": "0x1c76c89a21147c831205cee5b0102f04",
858
+ "verification": {
859
+ "overall_pass": true,
860
+ "constraints_passed": [],
861
+ "constraints_failed": []
862
+ },
863
+ "settlement": {
864
+ "reward": 0.003289162870942104,
865
+ "penalty": 0.0
866
+ },
867
+ "output_preview": "Synthetic execution of contract_58271e8e7624: SUCCESS"
868
+ },
869
+ {
870
+ "agent": "balanced_2",
871
+ "task_id": "contract_32cd9acee782",
872
+ "tier": "T2",
873
+ "domain": "analysis",
874
+ "proof_cid": "0x471176d4ac6d4ed3a3735c35278ee5f0",
875
+ "verification": {
876
+ "overall_pass": false,
877
+ "constraints_passed": [],
878
+ "constraints_failed": [
879
+ "format_compliance",
880
+ "accuracy",
881
+ "completeness"
882
+ ]
883
+ },
884
+ "settlement": {
885
+ "reward": 0.0,
886
+ "penalty": 0.0077242034277661235
887
+ },
888
+ "output_preview": "Synthetic execution of contract_32cd9acee782: FAILURE"
889
+ },
890
+ {
891
+ "agent": "conservative_0",
892
+ "task_id": "contract_aa1a90a8659f",
893
+ "tier": "T1",
894
+ "domain": "data_labeling",
895
+ "proof_cid": "0x65e0363ece76458545f2bb0944cff748",
896
+ "verification": {
897
+ "overall_pass": false,
898
+ "constraints_passed": [],
899
+ "constraints_failed": [
900
+ "format_compliance",
901
+ "completeness"
902
+ ]
903
+ },
904
+ "settlement": {
905
+ "reward": 0.0,
906
+ "penalty": 0.0008886288235577006
907
+ },
908
+ "output_preview": "Synthetic execution of contract_aa1a90a8659f: FAILURE"
909
+ },
910
+ {
911
+ "agent": "aggressive_1",
912
+ "task_id": "contract_8efbb7f13ac3",
913
+ "tier": "T1",
914
+ "domain": "summarization",
915
+ "proof_cid": "0x15e8f78016312c17dedd87590b4a2f8f",
916
+ "verification": {
917
+ "overall_pass": true,
918
+ "constraints_passed": [],
919
+ "constraints_failed": []
920
+ },
921
+ "settlement": {
922
+ "reward": 0.003413381231777636,
923
+ "penalty": 0.0
924
+ },
925
+ "output_preview": "Synthetic execution of contract_8efbb7f13ac3: SUCCESS"
926
+ },
927
+ {
928
+ "agent": "balanced_2",
929
+ "task_id": "contract_703bc5dcee48",
930
+ "tier": "T2",
931
+ "domain": "translation",
932
+ "proof_cid": "0xda3f3aa1ada8a769100280d73c08adcf",
933
+ "verification": {
934
+ "overall_pass": true,
935
+ "constraints_passed": [],
936
+ "constraints_failed": []
937
+ },
938
+ "settlement": {
939
+ "reward": 0.013512177627600068,
940
+ "penalty": 0.0
941
+ },
942
+ "output_preview": "Synthetic execution of contract_703bc5dcee48: SUCCESS"
943
+ },
944
+ {
945
+ "agent": "conservative_0",
946
+ "task_id": "contract_40435ef07179",
947
+ "tier": "T1",
948
+ "domain": "summarization",
949
+ "proof_cid": "0xe46a41ebb316b37aa5588d29cf4ea185",
950
+ "verification": {
951
+ "overall_pass": true,
952
+ "constraints_passed": [],
953
+ "constraints_failed": []
954
+ },
955
+ "settlement": {
956
+ "reward": 0.0032146772646714974,
957
+ "penalty": 0.0
958
+ },
959
+ "output_preview": "Synthetic execution of contract_40435ef07179: SUCCESS"
960
+ },
961
+ {
962
+ "agent": "balanced_2",
963
+ "task_id": "contract_a37bf191435e",
964
+ "tier": "T2",
965
+ "domain": "analysis",
966
+ "proof_cid": "0xcc164736fab643f47206c2a8c796984b",
967
+ "verification": {
968
+ "overall_pass": true,
969
+ "constraints_passed": [],
970
+ "constraints_failed": []
971
+ },
972
+ "settlement": {
973
+ "reward": 0.015496437087518517,
974
+ "penalty": 0.0
975
+ },
976
+ "output_preview": "Synthetic execution of contract_a37bf191435e: SUCCESS"
977
+ },
978
+ {
979
+ "agent": "conservative_0",
980
+ "task_id": "contract_9d22852235bd",
981
+ "tier": "T1",
982
+ "domain": "data_labeling",
983
+ "proof_cid": "0x656c161d92f950dad015528889fe52bc",
984
+ "verification": {
985
+ "overall_pass": false,
986
+ "constraints_passed": [],
987
+ "constraints_failed": [
988
+ "format_compliance",
989
+ "completeness"
990
+ ]
991
+ },
992
+ "settlement": {
993
+ "reward": 0.0,
994
+ "penalty": 0.000877606470489028
995
+ },
996
+ "output_preview": "Synthetic execution of contract_9d22852235bd: FAILURE"
997
+ },
998
+ {
999
+ "agent": "aggressive_1",
1000
+ "task_id": "contract_ef2cb460c2c2",
1001
+ "tier": "T1",
1002
+ "domain": "summarization",
1003
+ "proof_cid": "0x139e54cdb1ba712213c401e295a3ec59",
1004
+ "verification": {
1005
+ "overall_pass": true,
1006
+ "constraints_passed": [],
1007
+ "constraints_failed": []
1008
+ },
1009
+ "settlement": {
1010
+ "reward": 0.0033337488446999112,
1011
+ "penalty": 0.0
1012
+ },
1013
+ "output_preview": "Synthetic execution of contract_ef2cb460c2c2: SUCCESS"
1014
+ },
1015
+ {
1016
+ "agent": "balanced_2",
1017
+ "task_id": "contract_6bf723ac3a92",
1018
+ "tier": "T2",
1019
+ "domain": "translation",
1020
+ "proof_cid": "0xba0267f54737fb6a8656055389919db6",
1021
+ "verification": {
1022
+ "overall_pass": false,
1023
+ "constraints_passed": [],
1024
+ "constraints_failed": [
1025
+ "accuracy",
1026
+ "terminology_consistency"
1027
+ ]
1028
+ },
1029
+ "settlement": {
1030
+ "reward": 0.0,
1031
+ "penalty": 0.005073162019464485
1032
+ },
1033
+ "output_preview": "Synthetic execution of contract_6bf723ac3a92: FAILURE"
1034
+ },
1035
+ {
1036
+ "agent": "adaptive_3",
1037
+ "task_id": "contract_eca507c33880",
1038
+ "tier": "T1",
1039
+ "domain": "summarization",
1040
+ "proof_cid": "0x147ec0fe2fd131f24e0042d69b6c2f73",
1041
+ "verification": {
1042
+ "overall_pass": false,
1043
+ "constraints_passed": [],
1044
+ "constraints_failed": [
1045
+ "length_constraint",
1046
+ "accuracy"
1047
+ ]
1048
+ },
1049
+ "settlement": {
1050
+ "reward": 0.0,
1051
+ "penalty": 0.001113520143773937
1052
+ },
1053
+ "output_preview": "Synthetic execution of contract_eca507c33880: FAILURE"
1054
+ }
1055
+ ]
server/results/time_series.json ADDED
@@ -0,0 +1,178 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "timestamps": [
3
+ 1.0,
4
+ 2.0,
5
+ 3.0,
6
+ 4.0,
7
+ 5.0,
8
+ 6.0,
9
+ 7.0,
10
+ 8.0,
11
+ 9.0,
12
+ 10.0,
13
+ 11.0,
14
+ 12.0,
15
+ 13.0,
16
+ 14.0,
17
+ 15.0,
18
+ 16.0,
19
+ 17.0,
20
+ 18.0,
21
+ 19.0,
22
+ 20.0
23
+ ],
24
+ "aggregate_safety": [
25
+ 0.7149303399599031,
26
+ 0.7113646100135703,
27
+ 0.7332981455002783,
28
+ 0.7299758899151536,
29
+ 0.7263351199753616,
30
+ 0.722712508451399,
31
+ 0.7191079647777887,
32
+ 0.7155213988407516,
33
+ 0.7119527209759524,
34
+ 0.7084018419662584,
35
+ 0.7017113756760207,
36
+ 0.6982115755891051,
37
+ 0.6947292308279442,
38
+ 0.6912642543337382,
39
+ 0.6878165594818937,
40
+ 0.6843860600798599,
41
+ 0.7799318073119693,
42
+ 0.7760418811947121,
43
+ 0.7721713561649035,
44
+ 0.7608347665972391
45
+ ],
46
+ "total_balance": [
47
+ 2.477838603395556,
48
+ 2.4977873016008205,
49
+ 2.4825172039422414,
50
+ 2.4708295600664782,
51
+ 2.4888006757500536,
52
+ 2.4874408558380057,
53
+ 2.504583435487169,
54
+ 2.499465860249674,
55
+ 2.516130472788416,
56
+ 2.535196400232379,
57
+ 2.524565789866131,
58
+ 2.5167951641977018,
59
+ 2.5142898690162636,
60
+ 2.5110761979330025,
61
+ 2.507348701350255,
62
+ 2.528609461481337,
63
+ 2.513811457214175,
64
+ 2.5283483872499946,
65
+ 2.5455595016021846,
66
+ 2.5323289618131573
67
+ ],
68
+ "active_agent_count": [
69
+ 5,
70
+ 5,
71
+ 5,
72
+ 5,
73
+ 5,
74
+ 5,
75
+ 5,
76
+ 5,
77
+ 5,
78
+ 5,
79
+ 5,
80
+ 5,
81
+ 5,
82
+ 5,
83
+ 5,
84
+ 5,
85
+ 5,
86
+ 5,
87
+ 5,
88
+ 5
89
+ ],
90
+ "contracts_completed": [
91
+ 2,
92
+ 5,
93
+ 6,
94
+ 8,
95
+ 10,
96
+ 13,
97
+ 15,
98
+ 16,
99
+ 18,
100
+ 21,
101
+ 23,
102
+ 23,
103
+ 25,
104
+ 27,
105
+ 29,
106
+ 32,
107
+ 33,
108
+ 35,
109
+ 37,
110
+ 38
111
+ ],
112
+ "contracts_failed": [
113
+ 1,
114
+ 1,
115
+ 2,
116
+ 3,
117
+ 3,
118
+ 4,
119
+ 5,
120
+ 7,
121
+ 8,
122
+ 8,
123
+ 9,
124
+ 11,
125
+ 12,
126
+ 13,
127
+ 14,
128
+ 14,
129
+ 16,
130
+ 17,
131
+ 17,
132
+ 20
133
+ ],
134
+ "rewards_paid": [
135
+ 0.020224728900398996,
136
+ 0.04167342710566373,
137
+ 0.0450424235029837,
138
+ 0.05181667566197762,
139
+ 0.07128779134555285,
140
+ 0.07917799268097525,
141
+ 0.09868090459334405,
142
+ 0.10221821302000404,
143
+ 0.12148243526317948,
144
+ 0.14204836270714202,
145
+ 0.1483451387599555,
146
+ 0.1483451387599555,
147
+ 0.15376610970419477,
148
+ 0.1593747355660904,
149
+ 0.16606556890326696,
150
+ 0.188826329034349,
151
+ 0.19211549190529112,
152
+ 0.20904105076466883,
153
+ 0.22775216511685886,
154
+ 0.23108591396155878
155
+ ],
156
+ "penalties_collected": [
157
+ 0.0008861255048430355,
158
+ 0.0008861255048430355,
159
+ 0.010025219560741828,
160
+ 0.018987115595499263,
161
+ 0.018987115595499263,
162
+ 0.026737136842969356,
163
+ 0.02759746910617459,
164
+ 0.03475235277032988,
165
+ 0.03585196247476302,
166
+ 0.03585196247476302,
167
+ 0.04327934889382405,
168
+ 0.04954997456225311,
169
+ 0.055976240687930896,
170
+ 0.06329853763308756,
171
+ 0.07221686755301177,
172
+ 0.07221686755301177,
173
+ 0.08080403469111593,
174
+ 0.08169266351467362,
175
+ 0.08169266351467362,
176
+ 0.08875695214840107
177
+ ]
178
+ }
server/runner.py CHANGED
@@ -33,7 +33,6 @@ from cgae_engine.registry import AgentRegistry, AgentStatus
33
  from cgae_engine.contracts import ContractManager, ContractStatus
34
  from cgae_engine.economy import Economy, EconomyConfig, EconomySnapshot
35
  from cgae_engine.marketplace import TaskMarketplace
36
- from cgae_engine.audit import AuditOrchestrator
37
  from agents.base import BaseAgent, AgentDecision
38
  from agents.strategies import create_agent_cohort
39
 
@@ -123,12 +122,15 @@ class SimulationRunner:
123
  self.economy.contracts,
124
  contracts_per_step=self.config.contracts_per_step,
125
  )
126
- self.audit = AuditOrchestrator()
127
-
128
- # Create agent cohort
129
  self.agents: dict[str, BaseAgent] = {}
130
  self.metrics = SimulationMetrics()
131
 
 
 
 
 
 
 
132
  def setup(self):
133
  """Register agents and run initial audits."""
134
  cohort = create_agent_cohort(self.config.agent_strategies)
@@ -142,14 +144,10 @@ class SimulationRunner:
142
  self.agents[record.agent_id] = agent
143
 
144
  # Initial audit with true robustness (+ small noise)
145
- audit_result = self.audit.synthetic_audit(
146
- record.agent_id,
147
- base_robustness=agent.true_robustness,
148
- noise_scale=0.03,
149
- )
150
  self.economy.audit_agent(
151
  record.agent_id,
152
- audit_result.robustness,
153
  audit_type="registration",
154
  )
155
 
@@ -281,15 +279,11 @@ class SimulationRunner:
281
  record.total_spent += cost
282
  new_r = agent.invest_robustness(dim, amount)
283
  # Re-audit with improved robustness
284
- audit_result = self.audit.synthetic_audit(
285
- agent_id,
286
- base_robustness=new_r,
287
- noise_scale=0.02,
288
- )
289
  old_tier = record.current_tier
290
  self.economy.audit_agent(
291
  agent_id,
292
- audit_result.robustness,
293
  audit_type="upgrade",
294
  )
295
  new_tier = record.current_tier
@@ -305,10 +299,7 @@ class SimulationRunner:
305
  def audit_callback(aid):
306
  agent = self.agents.get(aid)
307
  if agent:
308
- result = self.audit.synthetic_audit(
309
- aid, base_robustness=agent.true_robustness, noise_scale=0.04
310
- )
311
- return result.robustness
312
  return None
313
 
314
  self.economy.step(audit_callback=audit_callback)
 
33
  from cgae_engine.contracts import ContractManager, ContractStatus
34
  from cgae_engine.economy import Economy, EconomyConfig, EconomySnapshot
35
  from cgae_engine.marketplace import TaskMarketplace
 
36
  from agents.base import BaseAgent, AgentDecision
37
  from agents.strategies import create_agent_cohort
38
 
 
122
  self.economy.contracts,
123
  contracts_per_step=self.config.contracts_per_step,
124
  )
 
 
 
125
  self.agents: dict[str, BaseAgent] = {}
126
  self.metrics = SimulationMetrics()
127
 
128
+ @staticmethod
129
+ def _noisy_robustness(r: RobustnessVector, noise: float = 0.03) -> RobustnessVector:
130
+ """Add small noise to a robustness vector for synthetic audits."""
131
+ def clamp(v): return max(0.0, min(1.0, v + random.gauss(0, noise)))
132
+ return RobustnessVector(cc=clamp(r.cc), er=clamp(r.er), as_=clamp(r.as_), ih=clamp(r.ih))
133
+
134
  def setup(self):
135
  """Register agents and run initial audits."""
136
  cohort = create_agent_cohort(self.config.agent_strategies)
 
144
  self.agents[record.agent_id] = agent
145
 
146
  # Initial audit with true robustness (+ small noise)
147
+ noisy_r = self._noisy_robustness(agent.true_robustness)
 
 
 
 
148
  self.economy.audit_agent(
149
  record.agent_id,
150
+ noisy_r,
151
  audit_type="registration",
152
  )
153
 
 
279
  record.total_spent += cost
280
  new_r = agent.invest_robustness(dim, amount)
281
  # Re-audit with improved robustness
282
+ noisy_r = self._noisy_robustness(new_r, 0.02)
 
 
 
 
283
  old_tier = record.current_tier
284
  self.economy.audit_agent(
285
  agent_id,
286
+ noisy_r,
287
  audit_type="upgrade",
288
  )
289
  new_tier = record.current_tier
 
299
  def audit_callback(aid):
300
  agent = self.agents.get(aid)
301
  if agent:
302
+ return self._noisy_robustness(agent.true_robustness, 0.04)
 
 
 
303
  return None
304
 
305
  self.economy.step(audit_callback=audit_callback)