@pga-ai/core 0.4.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (268) hide show
  1. package/README.md +278 -0
  2. package/dist/PGA.d.ts +156 -0
  3. package/dist/PGA.d.ts.map +1 -0
  4. package/dist/PGA.js +636 -0
  5. package/dist/PGA.js.map +1 -0
  6. package/dist/advanced-ai/ModelRouter.d.ts +82 -0
  7. package/dist/advanced-ai/ModelRouter.d.ts.map +1 -0
  8. package/dist/advanced-ai/ModelRouter.js +280 -0
  9. package/dist/advanced-ai/ModelRouter.js.map +1 -0
  10. package/dist/advanced-ai/ThinkingEngine.d.ts +50 -0
  11. package/dist/advanced-ai/ThinkingEngine.d.ts.map +1 -0
  12. package/dist/advanced-ai/ThinkingEngine.js +179 -0
  13. package/dist/advanced-ai/ThinkingEngine.js.map +1 -0
  14. package/dist/core/ContextMemory.d.ts +56 -0
  15. package/dist/core/ContextMemory.d.ts.map +1 -0
  16. package/dist/core/ContextMemory.js +253 -0
  17. package/dist/core/ContextMemory.js.map +1 -0
  18. package/dist/core/DNAProfile.d.ts +19 -0
  19. package/dist/core/DNAProfile.d.ts.map +1 -0
  20. package/dist/core/DNAProfile.js +141 -0
  21. package/dist/core/DNAProfile.js.map +1 -0
  22. package/dist/core/FitnessTracker.d.ts +13 -0
  23. package/dist/core/FitnessTracker.d.ts.map +1 -0
  24. package/dist/core/FitnessTracker.js +101 -0
  25. package/dist/core/FitnessTracker.js.map +1 -0
  26. package/dist/core/GenomeKernel.d.ts +57 -0
  27. package/dist/core/GenomeKernel.d.ts.map +1 -0
  28. package/dist/core/GenomeKernel.js +305 -0
  29. package/dist/core/GenomeKernel.js.map +1 -0
  30. package/dist/core/GenomeManager.d.ts +21 -0
  31. package/dist/core/GenomeManager.d.ts.map +1 -0
  32. package/dist/core/GenomeManager.js +123 -0
  33. package/dist/core/GenomeManager.js.map +1 -0
  34. package/dist/core/LearningAnnouncer.d.ts +16 -0
  35. package/dist/core/LearningAnnouncer.d.ts.map +1 -0
  36. package/dist/core/LearningAnnouncer.js +176 -0
  37. package/dist/core/LearningAnnouncer.js.map +1 -0
  38. package/dist/core/ProactiveSuggestions.d.ts +25 -0
  39. package/dist/core/ProactiveSuggestions.d.ts.map +1 -0
  40. package/dist/core/ProactiveSuggestions.js +238 -0
  41. package/dist/core/ProactiveSuggestions.js.map +1 -0
  42. package/dist/core/PromptAssembler.d.ts +12 -0
  43. package/dist/core/PromptAssembler.d.ts.map +1 -0
  44. package/dist/core/PromptAssembler.js +74 -0
  45. package/dist/core/PromptAssembler.js.map +1 -0
  46. package/dist/enterprise/AuthManager.d.ts +71 -0
  47. package/dist/enterprise/AuthManager.d.ts.map +1 -0
  48. package/dist/enterprise/AuthManager.js +216 -0
  49. package/dist/enterprise/AuthManager.js.map +1 -0
  50. package/dist/enterprise/RateLimiter.d.ts +48 -0
  51. package/dist/enterprise/RateLimiter.d.ts.map +1 -0
  52. package/dist/enterprise/RateLimiter.js +193 -0
  53. package/dist/enterprise/RateLimiter.js.map +1 -0
  54. package/dist/evaluation/BenchmarkSuites.d.ts +27 -0
  55. package/dist/evaluation/BenchmarkSuites.d.ts.map +1 -0
  56. package/dist/evaluation/BenchmarkSuites.js +69 -0
  57. package/dist/evaluation/BenchmarkSuites.js.map +1 -0
  58. package/dist/evaluation/CalibrationManager.d.ts +66 -0
  59. package/dist/evaluation/CalibrationManager.d.ts.map +1 -0
  60. package/dist/evaluation/CalibrationManager.js +117 -0
  61. package/dist/evaluation/CalibrationManager.js.map +1 -0
  62. package/dist/evaluation/Evaluator.d.ts +79 -0
  63. package/dist/evaluation/Evaluator.d.ts.map +1 -0
  64. package/dist/evaluation/Evaluator.js +359 -0
  65. package/dist/evaluation/Evaluator.js.map +1 -0
  66. package/dist/evaluation/EvolutionGuardrails.d.ts +29 -0
  67. package/dist/evaluation/EvolutionGuardrails.d.ts.map +1 -0
  68. package/dist/evaluation/EvolutionGuardrails.js +166 -0
  69. package/dist/evaluation/EvolutionGuardrails.js.map +1 -0
  70. package/dist/evaluation/SandboxSuites.d.ts +26 -0
  71. package/dist/evaluation/SandboxSuites.d.ts.map +1 -0
  72. package/dist/evaluation/SandboxSuites.js +252 -0
  73. package/dist/evaluation/SandboxSuites.js.map +1 -0
  74. package/dist/evaluation/SemanticJudge.d.ts +21 -0
  75. package/dist/evaluation/SemanticJudge.d.ts.map +1 -0
  76. package/dist/evaluation/SemanticJudge.js +139 -0
  77. package/dist/evaluation/SemanticJudge.js.map +1 -0
  78. package/dist/evaluation/fixtures/core-coding-v1.json +68 -0
  79. package/dist/evaluation/fixtures/core-general-v1.json +68 -0
  80. package/dist/evolution/CanaryDeployment.d.ts +77 -0
  81. package/dist/evolution/CanaryDeployment.d.ts.map +1 -0
  82. package/dist/evolution/CanaryDeployment.js +261 -0
  83. package/dist/evolution/CanaryDeployment.js.map +1 -0
  84. package/dist/evolution/DriftAnalyzer.d.ts +63 -0
  85. package/dist/evolution/DriftAnalyzer.d.ts.map +1 -0
  86. package/dist/evolution/DriftAnalyzer.js +283 -0
  87. package/dist/evolution/DriftAnalyzer.js.map +1 -0
  88. package/dist/evolution/FitnessCalculator.d.ts +47 -0
  89. package/dist/evolution/FitnessCalculator.d.ts.map +1 -0
  90. package/dist/evolution/FitnessCalculator.js +177 -0
  91. package/dist/evolution/FitnessCalculator.js.map +1 -0
  92. package/dist/evolution/MutationOperator.d.ts +76 -0
  93. package/dist/evolution/MutationOperator.d.ts.map +1 -0
  94. package/dist/evolution/MutationOperator.js +267 -0
  95. package/dist/evolution/MutationOperator.js.map +1 -0
  96. package/dist/evolution/PromotionGate.d.ts +45 -0
  97. package/dist/evolution/PromotionGate.d.ts.map +1 -0
  98. package/dist/evolution/PromotionGate.js +248 -0
  99. package/dist/evolution/PromotionGate.js.map +1 -0
  100. package/dist/evolution/boost/EvolutionBoostEngine.d.ts +69 -0
  101. package/dist/evolution/boost/EvolutionBoostEngine.d.ts.map +1 -0
  102. package/dist/evolution/boost/EvolutionBoostEngine.js +185 -0
  103. package/dist/evolution/boost/EvolutionBoostEngine.js.map +1 -0
  104. package/dist/evolution/boost/GeneticRecombinator.d.ts +26 -0
  105. package/dist/evolution/boost/GeneticRecombinator.d.ts.map +1 -0
  106. package/dist/evolution/boost/GeneticRecombinator.js +179 -0
  107. package/dist/evolution/boost/GeneticRecombinator.js.map +1 -0
  108. package/dist/evolution/boost/MetaEvolutionEngine.d.ts +48 -0
  109. package/dist/evolution/boost/MetaEvolutionEngine.d.ts.map +1 -0
  110. package/dist/evolution/boost/MetaEvolutionEngine.js +193 -0
  111. package/dist/evolution/boost/MetaEvolutionEngine.js.map +1 -0
  112. package/dist/evolution/boost/ParallelEvolutionEngine.d.ts +44 -0
  113. package/dist/evolution/boost/ParallelEvolutionEngine.d.ts.map +1 -0
  114. package/dist/evolution/boost/ParallelEvolutionEngine.js +135 -0
  115. package/dist/evolution/boost/ParallelEvolutionEngine.js.map +1 -0
  116. package/dist/evolution/boost/ParetoOptimizer.d.ts +42 -0
  117. package/dist/evolution/boost/ParetoOptimizer.d.ts.map +1 -0
  118. package/dist/evolution/boost/ParetoOptimizer.js +167 -0
  119. package/dist/evolution/boost/ParetoOptimizer.js.map +1 -0
  120. package/dist/evolution/boost/operators/BreakthroughOperator.d.ts +22 -0
  121. package/dist/evolution/boost/operators/BreakthroughOperator.d.ts.map +1 -0
  122. package/dist/evolution/boost/operators/BreakthroughOperator.js +217 -0
  123. package/dist/evolution/boost/operators/BreakthroughOperator.js.map +1 -0
  124. package/dist/evolution/boost/operators/CrossoverMutationOperator.d.ts +26 -0
  125. package/dist/evolution/boost/operators/CrossoverMutationOperator.d.ts.map +1 -0
  126. package/dist/evolution/boost/operators/CrossoverMutationOperator.js +160 -0
  127. package/dist/evolution/boost/operators/CrossoverMutationOperator.js.map +1 -0
  128. package/dist/evolution/boost/operators/PatternExtractionOperator.d.ts +24 -0
  129. package/dist/evolution/boost/operators/PatternExtractionOperator.d.ts.map +1 -0
  130. package/dist/evolution/boost/operators/PatternExtractionOperator.js +212 -0
  131. package/dist/evolution/boost/operators/PatternExtractionOperator.js.map +1 -0
  132. package/dist/evolution/boost/operators/SemanticRestructuringOperator.d.ts +19 -0
  133. package/dist/evolution/boost/operators/SemanticRestructuringOperator.d.ts.map +1 -0
  134. package/dist/evolution/boost/operators/SemanticRestructuringOperator.js +106 -0
  135. package/dist/evolution/boost/operators/SemanticRestructuringOperator.js.map +1 -0
  136. package/dist/gene-bank/CognitiveGene.d.ts +799 -0
  137. package/dist/gene-bank/CognitiveGene.d.ts.map +1 -0
  138. package/dist/gene-bank/CognitiveGene.js +128 -0
  139. package/dist/gene-bank/CognitiveGene.js.map +1 -0
  140. package/dist/gene-bank/GeneAdopter.d.ts +75 -0
  141. package/dist/gene-bank/GeneAdopter.d.ts.map +1 -0
  142. package/dist/gene-bank/GeneAdopter.js +271 -0
  143. package/dist/gene-bank/GeneAdopter.js.map +1 -0
  144. package/dist/gene-bank/GeneBank.d.ts +124 -0
  145. package/dist/gene-bank/GeneBank.d.ts.map +1 -0
  146. package/dist/gene-bank/GeneBank.js +261 -0
  147. package/dist/gene-bank/GeneBank.js.map +1 -0
  148. package/dist/gene-bank/GeneExtractor.d.ts +59 -0
  149. package/dist/gene-bank/GeneExtractor.d.ts.map +1 -0
  150. package/dist/gene-bank/GeneExtractor.js +311 -0
  151. package/dist/gene-bank/GeneExtractor.js.map +1 -0
  152. package/dist/gene-bank/GeneMatcher.d.ts +82 -0
  153. package/dist/gene-bank/GeneMatcher.d.ts.map +1 -0
  154. package/dist/gene-bank/GeneMatcher.js +215 -0
  155. package/dist/gene-bank/GeneMatcher.js.map +1 -0
  156. package/dist/gene-bank/PGAIntegration.d.ts +53 -0
  157. package/dist/gene-bank/PGAIntegration.d.ts.map +1 -0
  158. package/dist/gene-bank/PGAIntegration.js +139 -0
  159. package/dist/gene-bank/PGAIntegration.js.map +1 -0
  160. package/dist/gene-bank/SandboxTester.d.ts +92 -0
  161. package/dist/gene-bank/SandboxTester.d.ts.map +1 -0
  162. package/dist/gene-bank/SandboxTester.js +262 -0
  163. package/dist/gene-bank/SandboxTester.js.map +1 -0
  164. package/dist/gene-bank/adapters/InMemoryGeneStorage.d.ts +21 -0
  165. package/dist/gene-bank/adapters/InMemoryGeneStorage.d.ts.map +1 -0
  166. package/dist/gene-bank/adapters/InMemoryGeneStorage.js +115 -0
  167. package/dist/gene-bank/adapters/InMemoryGeneStorage.js.map +1 -0
  168. package/dist/gene-bank/adapters/PostgresGeneStorage.d.ts +21 -0
  169. package/dist/gene-bank/adapters/PostgresGeneStorage.d.ts.map +1 -0
  170. package/dist/gene-bank/adapters/PostgresGeneStorage.js +272 -0
  171. package/dist/gene-bank/adapters/PostgresGeneStorage.js.map +1 -0
  172. package/dist/gene-bank/index.d.ts +7 -0
  173. package/dist/gene-bank/index.d.ts.map +1 -0
  174. package/dist/gene-bank/index.js +7 -0
  175. package/dist/gene-bank/index.js.map +1 -0
  176. package/dist/index.d.ts +72 -0
  177. package/dist/index.d.ts.map +1 -0
  178. package/dist/index.js +37 -0
  179. package/dist/index.js.map +1 -0
  180. package/dist/interfaces/LLMAdapter.d.ts +31 -0
  181. package/dist/interfaces/LLMAdapter.d.ts.map +1 -0
  182. package/dist/interfaces/LLMAdapter.js +2 -0
  183. package/dist/interfaces/LLMAdapter.js.map +1 -0
  184. package/dist/interfaces/StorageAdapter.d.ts +49 -0
  185. package/dist/interfaces/StorageAdapter.d.ts.map +1 -0
  186. package/dist/interfaces/StorageAdapter.js +2 -0
  187. package/dist/interfaces/StorageAdapter.js.map +1 -0
  188. package/dist/memory/LayeredMemory.d.ts +112 -0
  189. package/dist/memory/LayeredMemory.d.ts.map +1 -0
  190. package/dist/memory/LayeredMemory.js +405 -0
  191. package/dist/memory/LayeredMemory.js.map +1 -0
  192. package/dist/memory-compaction/MemoryCompactor.d.ts +18 -0
  193. package/dist/memory-compaction/MemoryCompactor.d.ts.map +1 -0
  194. package/dist/memory-compaction/MemoryCompactor.js +156 -0
  195. package/dist/memory-compaction/MemoryCompactor.js.map +1 -0
  196. package/dist/memory-compaction/index.d.ts +6 -0
  197. package/dist/memory-compaction/index.d.ts.map +1 -0
  198. package/dist/memory-compaction/index.js +5 -0
  199. package/dist/memory-compaction/index.js.map +1 -0
  200. package/dist/memory-compaction/strategies/BaseStrategy.d.ts +9 -0
  201. package/dist/memory-compaction/strategies/BaseStrategy.d.ts.map +1 -0
  202. package/dist/memory-compaction/strategies/BaseStrategy.js +50 -0
  203. package/dist/memory-compaction/strategies/BaseStrategy.js.map +1 -0
  204. package/dist/memory-compaction/strategies/ImportanceBasedStrategy.d.ts +9 -0
  205. package/dist/memory-compaction/strategies/ImportanceBasedStrategy.d.ts.map +1 -0
  206. package/dist/memory-compaction/strategies/ImportanceBasedStrategy.js +101 -0
  207. package/dist/memory-compaction/strategies/ImportanceBasedStrategy.js.map +1 -0
  208. package/dist/memory-compaction/strategies/SlidingWindowStrategy.d.ts +9 -0
  209. package/dist/memory-compaction/strategies/SlidingWindowStrategy.d.ts.map +1 -0
  210. package/dist/memory-compaction/strategies/SlidingWindowStrategy.js +87 -0
  211. package/dist/memory-compaction/strategies/SlidingWindowStrategy.js.map +1 -0
  212. package/dist/memory-compaction/types.d.ts +78 -0
  213. package/dist/memory-compaction/types.d.ts.map +1 -0
  214. package/dist/memory-compaction/types.js +2 -0
  215. package/dist/memory-compaction/types.js.map +1 -0
  216. package/dist/monitoring/AlertWebhooks.d.ts +57 -0
  217. package/dist/monitoring/AlertWebhooks.d.ts.map +1 -0
  218. package/dist/monitoring/AlertWebhooks.js +205 -0
  219. package/dist/monitoring/AlertWebhooks.js.map +1 -0
  220. package/dist/monitoring/MetricsCollector.d.ts +120 -0
  221. package/dist/monitoring/MetricsCollector.d.ts.map +1 -0
  222. package/dist/monitoring/MetricsCollector.js +274 -0
  223. package/dist/monitoring/MetricsCollector.js.map +1 -0
  224. package/dist/monitoring/MonitoringDashboard.d.ts +38 -0
  225. package/dist/monitoring/MonitoringDashboard.d.ts.map +1 -0
  226. package/dist/monitoring/MonitoringDashboard.js +271 -0
  227. package/dist/monitoring/MonitoringDashboard.js.map +1 -0
  228. package/dist/plugins/PluginManager.d.ts +61 -0
  229. package/dist/plugins/PluginManager.d.ts.map +1 -0
  230. package/dist/plugins/PluginManager.js +154 -0
  231. package/dist/plugins/PluginManager.js.map +1 -0
  232. package/dist/rag/RAGEngine.d.ts +54 -0
  233. package/dist/rag/RAGEngine.d.ts.map +1 -0
  234. package/dist/rag/RAGEngine.js +162 -0
  235. package/dist/rag/RAGEngine.js.map +1 -0
  236. package/dist/rag/VectorStoreAdapter.d.ts +40 -0
  237. package/dist/rag/VectorStoreAdapter.d.ts.map +1 -0
  238. package/dist/rag/VectorStoreAdapter.js +106 -0
  239. package/dist/rag/VectorStoreAdapter.js.map +1 -0
  240. package/dist/realtime/EventEmitter.d.ts +86 -0
  241. package/dist/realtime/EventEmitter.d.ts.map +1 -0
  242. package/dist/realtime/EventEmitter.js +173 -0
  243. package/dist/realtime/EventEmitter.js.map +1 -0
  244. package/dist/realtime/StreamingManager.d.ts +26 -0
  245. package/dist/realtime/StreamingManager.d.ts.map +1 -0
  246. package/dist/realtime/StreamingManager.js +175 -0
  247. package/dist/realtime/StreamingManager.js.map +1 -0
  248. package/dist/reasoning/ReasoningEngine.d.ts +57 -0
  249. package/dist/reasoning/ReasoningEngine.d.ts.map +1 -0
  250. package/dist/reasoning/ReasoningEngine.js +316 -0
  251. package/dist/reasoning/ReasoningEngine.js.map +1 -0
  252. package/dist/resilience/CircuitBreaker.d.ts +41 -0
  253. package/dist/resilience/CircuitBreaker.d.ts.map +1 -0
  254. package/dist/resilience/CircuitBreaker.js +108 -0
  255. package/dist/resilience/CircuitBreaker.js.map +1 -0
  256. package/dist/resilience/RetryManager.d.ts +14 -0
  257. package/dist/resilience/RetryManager.d.ts.map +1 -0
  258. package/dist/resilience/RetryManager.js +35 -0
  259. package/dist/resilience/RetryManager.js.map +1 -0
  260. package/dist/types/GenomeV2.d.ts +226 -0
  261. package/dist/types/GenomeV2.d.ts.map +1 -0
  262. package/dist/types/GenomeV2.js +2 -0
  263. package/dist/types/GenomeV2.js.map +1 -0
  264. package/dist/types/index.d.ts +205 -0
  265. package/dist/types/index.d.ts.map +1 -0
  266. package/dist/types/index.js +2 -0
  267. package/dist/types/index.js.map +1 -0
  268. package/package.json +75 -0
@@ -0,0 +1,117 @@
1
+ export class CalibrationManager {
2
+ storage;
3
+ cache = new Map();
4
+ constructor(storage) {
5
+ this.storage = storage;
6
+ void this.storage;
7
+ }
8
+ async getCalibratedThreshold(context) {
9
+ const cacheKey = this.getContextKey(context);
10
+ if (this.cache.has(cacheKey)) {
11
+ const history = this.cache.get(cacheKey);
12
+ return {
13
+ threshold: history.recommendedThreshold,
14
+ confidence: history.confidence,
15
+ source: 'cached',
16
+ };
17
+ }
18
+ const history = await this.loadCalibrationHistory(context);
19
+ if (history && history.points.length >= 10) {
20
+ this.cache.set(cacheKey, history);
21
+ return {
22
+ threshold: history.recommendedThreshold,
23
+ confidence: history.confidence,
24
+ source: 'calibrated',
25
+ };
26
+ }
27
+ const defaultThreshold = this.getDefaultThreshold(context);
28
+ return {
29
+ threshold: defaultThreshold,
30
+ confidence: 0.5,
31
+ source: 'default',
32
+ };
33
+ }
34
+ async recordCalibrationPoint(context, metrics, currentThreshold) {
35
+ const falsePositiveRate = metrics.passedSandbox > 0
36
+ ? metrics.rolledBack / metrics.passedSandbox
37
+ : 0;
38
+ const deploymentSuccessRate = metrics.deployedSuccessfully > 0
39
+ ? metrics.deployedSuccessfully / metrics.passedSandbox
40
+ : 0;
41
+ const falseNegativeRate = deploymentSuccessRate > 0.8
42
+ ? Math.max(0, 0.2 - (1 - currentThreshold))
43
+ : 0;
44
+ const optimalThreshold = this.calculateOptimalThreshold(falsePositiveRate, falseNegativeRate, currentThreshold);
45
+ const calibrationData = {
46
+ timestamp: new Date(),
47
+ context,
48
+ threshold: currentThreshold,
49
+ metrics,
50
+ performance: {
51
+ falsePositiveRate,
52
+ falseNegativeRate,
53
+ optimalThreshold,
54
+ },
55
+ };
56
+ console.log('📊 Calibration point recorded:', calibrationData);
57
+ this.cache.delete(this.getContextKey(context));
58
+ }
59
+ calculateOptimalThreshold(falsePositiveRate, falseNegativeRate, currentThreshold) {
60
+ if (falsePositiveRate > 0.1) {
61
+ return Math.min(1.0, currentThreshold + 0.05);
62
+ }
63
+ if (falseNegativeRate > 0.2 && falsePositiveRate < 0.05) {
64
+ return Math.max(0.5, currentThreshold - 0.05);
65
+ }
66
+ return currentThreshold;
67
+ }
68
+ async loadCalibrationHistory(context) {
69
+ void context;
70
+ return null;
71
+ }
72
+ getContextKey(context) {
73
+ return `${context.layer ?? 'any'}_${context.operator ?? 'any'}_${context.taskType ?? 'any'}`;
74
+ }
75
+ getDefaultThreshold(context) {
76
+ if (context.layer === 0)
77
+ return 1.0;
78
+ if (context.layer === 1)
79
+ return 0.75;
80
+ if (context.layer === 2)
81
+ return 0.60;
82
+ if (context.operator === 'safety_reinforcement')
83
+ return 0.85;
84
+ if (context.operator === 'compress_instructions')
85
+ return 0.65;
86
+ if (context.operator === 'reorder_constraints')
87
+ return 0.70;
88
+ if (context.operator === 'tool_selection_bias')
89
+ return 0.70;
90
+ if (context.taskType === 'coding')
91
+ return 0.75;
92
+ if (context.taskType === 'general')
93
+ return 0.65;
94
+ return 0.65;
95
+ }
96
+ async getCalibrationReport(context) {
97
+ const lines = [];
98
+ lines.push('# 📊 Sandbox Calibration Report\n');
99
+ if (context) {
100
+ const result = await this.getCalibratedThreshold(context);
101
+ const key = this.getContextKey(context);
102
+ lines.push(`**Context**: ${key}`);
103
+ lines.push(`**Threshold**: ${(result.threshold * 100).toFixed(1)}%`);
104
+ lines.push(`**Confidence**: ${(result.confidence * 100).toFixed(1)}%`);
105
+ lines.push(`**Source**: ${result.source}`);
106
+ lines.push('');
107
+ }
108
+ else {
109
+ lines.push('**Status**: Calibration system active');
110
+ lines.push(`**Cached contexts**: ${this.cache.size}`);
111
+ lines.push('');
112
+ lines.push('Use specific context for detailed calibration data.');
113
+ }
114
+ return lines.join('\n');
115
+ }
116
+ }
117
+ //# sourceMappingURL=CalibrationManager.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"CalibrationManager.js","sourceRoot":"","sources":["../../src/evaluation/CalibrationManager.ts"],"names":[],"mappings":"AAmDA,MAAM,OAAO,kBAAkB;IAGP;IAFZ,KAAK,GAAG,IAAI,GAAG,EAA8B,CAAC;IAEtD,YAAoB,OAAuB;QAAvB,YAAO,GAAP,OAAO,CAAgB;QAEvC,KAAK,IAAI,CAAC,OAAO,CAAC;IACtB,CAAC;IAOD,KAAK,CAAC,sBAAsB,CAAC,OAI5B;QAKG,MAAM,QAAQ,GAAG,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;QAG7C,IAAI,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC3B,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAE,CAAC;YAC1C,OAAO;gBACH,SAAS,EAAE,OAAO,CAAC,oBAAoB;gBACvC,UAAU,EAAE,OAAO,CAAC,UAAU;gBAC9B,MAAM,EAAE,QAAQ;aACnB,CAAC;QACN,CAAC;QAGD,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,sBAAsB,CAAC,OAAO,CAAC,CAAC;QAE3D,IAAI,OAAO,IAAI,OAAO,CAAC,MAAM,CAAC,MAAM,IAAI,EAAE,EAAE,CAAC;YAEzC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;YAClC,OAAO;gBACH,SAAS,EAAE,OAAO,CAAC,oBAAoB;gBACvC,UAAU,EAAE,OAAO,CAAC,UAAU;gBAC9B,MAAM,EAAE,YAAY;aACvB,CAAC;QACN,CAAC;QAGD,MAAM,gBAAgB,GAAG,IAAI,CAAC,mBAAmB,CAAC,OAAO,CAAC,CAAC;QAC3D,OAAO;YACH,SAAS,EAAE,gBAAgB;YAC3B,UAAU,EAAE,GAAG;YACf,MAAM,EAAE,SAAS;SACpB,CAAC;IACN,CAAC;IAOD,KAAK,CAAC,sBAAsB,CACxB,OAIC,EACD,OAKC,EACD,gBAAwB;QAGxB,MAAM,iBAAiB,GAAG,OAAO,CAAC,aAAa,GAAG,CAAC;YAC/C,CAAC,CAAC,OAAO,CAAC,UAAU,GAAG,OAAO,CAAC,aAAa;YAC5C,CAAC,CAAC,CAAC,CAAC;QAIR,MAAM,qBAAqB,GAAG,OAAO,CAAC,oBAAoB,GAAG,CAAC;YAC1D,CAAC,CAAC,OAAO,CAAC,oBAAoB,GAAG,OAAO,CAAC,aAAa;YACtD,CAAC,CAAC,CAAC,CAAC;QAER,MAAM,iBAAiB,GAAG,qBAAqB,GAAG,GAAG;YACjD,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,GAAG,CAAC,CAAC,GAAG,gBAAgB,CAAC,CAAC;YAC3C,CAAC,CAAC,CAAC,CAAC;QAIR,MAAM,gBAAgB,GAAG,IAAI,CAAC,yBAAyB,CACnD,iBAAiB,EACjB,iBAAiB,EACjB,gBAAgB,CACnB,CAAC;QAIF,MAAM,eAAe,GAAG;YACpB,SAAS,EAAE,IAAI,IAAI,EAAE;YACrB,OAAO;YACP,SAAS,EAAE,gBAAgB;YAC3B,OAAO;YACP,WAAW,EAAE;gBACT,iBAAiB;gBACjB,iBAAiB;gBACjB,gBAAgB;aACnB;SACJ,CAAC;QAGF,OAAO,CAAC,GAAG,CAAC,gCAAgC,EAAE,eAAe,CAAC,CAAC;QAG/D,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC;IACnD,CAAC;IAQO,yBAAyB,CAC7B,iBAAyB,EACzB,iBAAyB,EACzB,gBAAwB;QAMxB,IAAI,iBAAiB,GAAG,GAAG,EAAE,CAAC;YAC1B,OAAO,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,gBAAgB,GAAG,IAAI,CAAC,CAAC;QAClD,CAAC;QAGD,IAAI,iBAAiB,GAAG,GAAG,IAAI,iBAAiB,GAAG,IAAI,EAAE,CAAC;YACtD,OAAO,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,gBAAgB,GAAG,IAAI,CAAC,CAAC;QAClD,CAAC;QAGD,OAAO,gBAAgB,CAAC;IAC5B,CAAC;IAKO,KAAK,CAAC,sBAAsB,CAAC,OAIpC;QAGG,KAAK,OAAO,CAAC;QACb,OAAO,IAAI,CAAC;IAChB,CAAC;IAKO,aAAa,CAAC,OAIrB;QACG,OAAO,GAAG,OAAO,CAAC,KAAK,IAAI,KAAK,IAAI,OAAO,CAAC,QAAQ,IAAI,KAAK,IAAI,OAAO,CAAC,QAAQ,IAAI,KAAK,EAAE,CAAC;IACjG,CAAC;IAKO,mBAAmB,CAAC,OAI3B;QAEG,IAAI,OAAO,CAAC,KAAK,KAAK,CAAC;YAAE,OAAO,GAAG,CAAC;QACpC,IAAI,OAAO,CAAC,KAAK,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC;QACrC,IAAI,OAAO,CAAC,KAAK,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC;QAErC,IAAI,OAAO,CAAC,QAAQ,KAAK,sBAAsB;YAAE,OAAO,IAAI,CAAC;QAC7D,IAAI,OAAO,CAAC,QAAQ,KAAK,uBAAuB;YAAE,OAAO,IAAI,CAAC;QAC9D,IAAI,OAAO,CAAC,QAAQ,KAAK,qBAAqB;YAAE,OAAO,IAAI,CAAC;QAC5D,IAAI,OAAO,CAAC,QAAQ,KAAK,qBAAqB;YAAE,OAAO,IAAI,CAAC;QAE5D,IAAI,OAAO,CAAC,QAAQ,KAAK,QAAQ;YAAE,OAAO,IAAI,CAAC;QAC/C,IAAI,OAAO,CAAC,QAAQ,KAAK,SAAS;YAAE,OAAO,IAAI,CAAC;QAEhD,OAAO,IAAI,CAAC;IAChB,CAAC;IAOD,KAAK,CAAC,oBAAoB,CAAC,OAI1B;QACG,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,mCAAmC,CAAC,CAAC;QAEhD,IAAI,OAAO,EAAE,CAAC;YACV,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,sBAAsB,CAAC,OAAO,CAAC,CAAC;YAC1D,MAAM,GAAG,GAAG,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;YAExC,KAAK,CAAC,IAAI,CAAC,gBAAgB,GAAG,EAAE,CAAC,CAAC;YAClC,KAAK,CAAC,IAAI,CAAC,kBAAkB,CAAC,MAAM,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;YACrE,KAAK,CAAC,IAAI,CAAC,mBAAmB,CAAC,MAAM,CAAC,UAAU,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;YACvE,KAAK,CAAC,IAAI,CAAC,eAAe,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;YAC3C,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACnB,CAAC;aAAM,CAAC;YACJ,KAAK,CAAC,IAAI,CAAC,uCAAuC,CAAC,CAAC;YACpD,KAAK,CAAC,IAAI,CAAC,wBAAwB,IAAI,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC;YACtD,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACf,KAAK,CAAC,IAAI,CAAC,qDAAqD,CAAC,CAAC;QACtE,CAAC;QAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC5B,CAAC;CACJ"}
@@ -0,0 +1,79 @@
1
+ import type { BenchmarkSuiteId } from './BenchmarkSuites.js';
2
+ import type { LLMAdapter } from '../interfaces/LLMAdapter.js';
3
+ export interface EvaluatableGenome {
4
+ chat(userMessage: string, context: {
5
+ userId?: string;
6
+ }): Promise<string>;
7
+ chatWithMetrics?(userMessage: string, context: {
8
+ userId?: string;
9
+ }): Promise<{
10
+ content: string;
11
+ metrics: {
12
+ tokensUsed: number;
13
+ responseTime: number;
14
+ fitnessScore: number;
15
+ };
16
+ }>;
17
+ }
18
+ export interface EvaluationTask {
19
+ id: string;
20
+ name: string;
21
+ description: string;
22
+ userMessage: string;
23
+ expectedOutcome: {
24
+ keywords?: string[];
25
+ minLength?: number;
26
+ maxLength?: number;
27
+ successCriteria?: (response: string) => boolean;
28
+ };
29
+ difficulty: 'easy' | 'medium' | 'hard';
30
+ }
31
+ export interface EvaluationResult {
32
+ taskId: string;
33
+ taskName: string;
34
+ success: boolean;
35
+ response: string;
36
+ tokensUsed: number;
37
+ responseTime: number;
38
+ qualityScore: number;
39
+ failureReason?: string;
40
+ }
41
+ export interface BenchmarkResult {
42
+ totalTasks: number;
43
+ successfulTasks: number;
44
+ successRate: number;
45
+ avgTokensPerTask: number;
46
+ avgResponseTime: number;
47
+ avgQualityScore: number;
48
+ results: EvaluationResult[];
49
+ timestamp: Date;
50
+ }
51
+ export interface ComparisonResult {
52
+ withPGA: BenchmarkResult;
53
+ withoutPGA: BenchmarkResult;
54
+ improvements: {
55
+ successRate: number;
56
+ tokenEfficiency: number;
57
+ responseTime: number;
58
+ qualityScore: number;
59
+ };
60
+ verdict: 'PGA_WINS' | 'BASELINE_WINS' | 'TIE';
61
+ }
62
+ export declare class Evaluator {
63
+ private semanticJudge?;
64
+ constructor(options?: {
65
+ llm?: LLMAdapter;
66
+ enableSemanticJudge?: boolean;
67
+ });
68
+ evaluate(genome: EvaluatableGenome, tasks: EvaluationTask[], userId: string): Promise<BenchmarkResult>;
69
+ private evaluateTask;
70
+ private checkSuccess;
71
+ private checkSemanticValidation;
72
+ private calculateQuality;
73
+ compareWithSuite(genome: EvaluatableGenome, suiteId: BenchmarkSuiteId, userId: string): Promise<BenchmarkResult>;
74
+ compare(genomeWithPGA: EvaluatableGenome, genomeBaseline: EvaluatableGenome, tasks: EvaluationTask[], userId: string): Promise<ComparisonResult>;
75
+ formatReport(benchmark: BenchmarkResult): string;
76
+ formatComparisonReport(comparison: ComparisonResult): string;
77
+ }
78
+ export declare const STANDARD_TASKS: EvaluationTask[];
79
+ //# sourceMappingURL=Evaluator.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"Evaluator.d.ts","sourceRoot":"","sources":["../../src/evaluation/Evaluator.ts"],"names":[],"mappings":"AAcA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAI7D,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AAQ9D,MAAM,WAAW,iBAAiB;IAC9B,IAAI,CAAC,WAAW,EAAE,MAAM,EAAE,OAAO,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IACzE,eAAe,CAAC,CAAC,WAAW,EAAE,MAAM,EAAE,OAAO,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC;QACzE,OAAO,EAAE,MAAM,CAAC;QAChB,OAAO,EAAE;YAAE,UAAU,EAAE,MAAM,CAAC;YAAC,YAAY,EAAE,MAAM,CAAC;YAAC,YAAY,EAAE,MAAM,CAAA;SAAE,CAAC;KAC/E,CAAC,CAAC;CACN;AAED,MAAM,WAAW,cAAc;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,eAAe,EAAE;QACb,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAC;QACpB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,eAAe,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,OAAO,CAAC;KACnD,CAAC;IACF,UAAU,EAAE,MAAM,GAAG,QAAQ,GAAG,MAAM,CAAC;CAC1C;AAED,MAAM,WAAW,gBAAgB;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,OAAO,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,eAAe;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,eAAe,EAAE,MAAM,CAAC;IACxB,WAAW,EAAE,MAAM,CAAC;IACpB,gBAAgB,EAAE,MAAM,CAAC;IACzB,eAAe,EAAE,MAAM,CAAC;IACxB,eAAe,EAAE,MAAM,CAAC;IACxB,OAAO,EAAE,gBAAgB,EAAE,CAAC;IAC5B,SAAS,EAAE,IAAI,CAAC;CACnB;AAED,MAAM,WAAW,gBAAgB;IAC7B,OAAO,EAAE,eAAe,CAAC;IACzB,UAAU,EAAE,eAAe,CAAC;IAC5B,YAAY,EAAE;QACV,WAAW,EAAE,MAAM,CAAC;QACpB,eAAe,EAAE,MAAM,CAAC;QACxB,YAAY,EAAE,MAAM,CAAC;QACrB,YAAY,EAAE,MAAM,CAAC;KACxB,CAAC;IACF,OAAO,EAAE,UAAU,GAAG,eAAe,GAAG,KAAK,CAAC;CACjD;AAED,qBAAa,SAAS;IAClB,OAAO,CAAC,aAAa,CAAC,CAAgB;gBAE1B,OAAO,CAAC,EAAE;QAAE,GAAG,CAAC,EAAE,UAAU,CAAC;QAAC,mBAAmB,CAAC,EAAE,OAAO,CAAA;KAAE;IAYnE,QAAQ,CACV,MAAM,EAAE,iBAAiB,EACzB,KAAK,EAAE,cAAc,EAAE,EACvB,MAAM,EAAE,MAAM,GACf,OAAO,CAAC,eAAe,CAAC;YA+Bb,YAAY;IA2E1B,OAAO,CAAC,YAAY;IAwDpB,OAAO,CAAC,uBAAuB;IAyD/B,OAAO,CAAC,gBAAgB;IA4BlB,gBAAgB,CAClB,MAAM,EAAE,iBAAiB,EACzB,OAAO,EAAE,gBAAgB,EACzB,MAAM,EAAE,MAAM,GACf,OAAO,CAAC,eAAe,CAAC;IAUrB,OAAO,CACT,aAAa,EAAE,iBAAiB,EAChC,cAAc,EAAE,iBAAiB,EACjC,KAAK,EAAE,cAAc,EAAE,EACvB,MAAM,EAAE,MAAM,GACf,OAAO,CAAC,gBAAgB,CAAC;IAiD5B,YAAY,CAAC,SAAS,EAAE,eAAe,GAAG,MAAM;IAqChD,sBAAsB,CAAC,UAAU,EAAE,gBAAgB,GAAG,MAAM;CAwE/D;AAKD,eAAO,MAAM,cAAc,EAAE,cAAc,EAwD1C,CAAC"}
@@ -0,0 +1,359 @@
1
+ import { getBenchmarkSuite } from './BenchmarkSuites.js';
2
+ import { SemanticJudge } from './SemanticJudge.js';
3
+ export class Evaluator {
4
+ semanticJudge;
5
+ constructor(options) {
6
+ if (options?.llm && options?.enableSemanticJudge !== false) {
7
+ this.semanticJudge = new SemanticJudge(options.llm);
8
+ }
9
+ }
10
+ async evaluate(genome, tasks, userId) {
11
+ const results = [];
12
+ for (const task of tasks) {
13
+ const result = await this.evaluateTask(genome, task, userId);
14
+ results.push(result);
15
+ }
16
+ const successful = results.filter(r => r.success).length;
17
+ const successRate = (successful / results.length) * 100;
18
+ const avgTokens = results.reduce((sum, r) => sum + r.tokensUsed, 0) / results.length;
19
+ const avgTime = results.reduce((sum, r) => sum + r.responseTime, 0) / results.length;
20
+ const avgQuality = results.reduce((sum, r) => sum + r.qualityScore, 0) / results.length;
21
+ return {
22
+ totalTasks: results.length,
23
+ successfulTasks: successful,
24
+ successRate,
25
+ avgTokensPerTask: Math.round(avgTokens),
26
+ avgResponseTime: Math.round(avgTime),
27
+ avgQualityScore: Math.round(avgQuality * 100) / 100,
28
+ results,
29
+ timestamp: new Date(),
30
+ };
31
+ }
32
+ async evaluateTask(genome, task, userId) {
33
+ const startTime = Date.now();
34
+ try {
35
+ const response = await genome.chat(task.userMessage, { userId });
36
+ const endTime = Date.now();
37
+ const responseTime = endTime - startTime;
38
+ const tokensUsed = Math.ceil((task.userMessage.length + response.length) / 4);
39
+ const basicChecks = this.checkSuccess(response, task.expectedOutcome);
40
+ let semanticChecks = { success: true };
41
+ if ('semanticChecks' in task && task.semanticChecks) {
42
+ if (this.semanticJudge) {
43
+ const judgment = await this.semanticJudge.judge(task, response);
44
+ semanticChecks = {
45
+ success: judgment.passed && judgment.confidence >= 0.7,
46
+ failureReason: judgment.passed ? undefined : judgment.reasoning,
47
+ };
48
+ }
49
+ else {
50
+ semanticChecks = this.checkSemanticValidation(response, task.semanticChecks);
51
+ }
52
+ }
53
+ const success = basicChecks.success && semanticChecks.success;
54
+ const failureReason = basicChecks.failureReason || semanticChecks.failureReason;
55
+ const qualityScore = this.calculateQuality(response, task);
56
+ return {
57
+ taskId: task.id,
58
+ taskName: task.name,
59
+ success,
60
+ response,
61
+ tokensUsed,
62
+ responseTime,
63
+ qualityScore,
64
+ failureReason,
65
+ };
66
+ }
67
+ catch (error) {
68
+ const endTime = Date.now();
69
+ return {
70
+ taskId: task.id,
71
+ taskName: task.name,
72
+ success: false,
73
+ response: '',
74
+ tokensUsed: 0,
75
+ responseTime: endTime - startTime,
76
+ qualityScore: 0,
77
+ failureReason: error instanceof Error ? error.message : 'Unknown error',
78
+ };
79
+ }
80
+ }
81
+ checkSuccess(response, expected) {
82
+ if (expected.keywords) {
83
+ for (const keyword of expected.keywords) {
84
+ if (!response.toLowerCase().includes(keyword.toLowerCase())) {
85
+ return {
86
+ success: false,
87
+ failureReason: `Missing keyword: "${keyword}"`,
88
+ };
89
+ }
90
+ }
91
+ }
92
+ if (expected.minLength && response.length < expected.minLength) {
93
+ return {
94
+ success: false,
95
+ failureReason: `Response too short (${response.length} < ${expected.minLength})`,
96
+ };
97
+ }
98
+ if (expected.maxLength && response.length > expected.maxLength) {
99
+ return {
100
+ success: false,
101
+ failureReason: `Response too long (${response.length} > ${expected.maxLength})`,
102
+ };
103
+ }
104
+ if (expected.successCriteria) {
105
+ try {
106
+ if (!expected.successCriteria(response)) {
107
+ return {
108
+ success: false,
109
+ failureReason: 'Custom success criteria not met',
110
+ };
111
+ }
112
+ }
113
+ catch (error) {
114
+ return {
115
+ success: false,
116
+ failureReason: `Success criteria error: ${error}`,
117
+ };
118
+ }
119
+ }
120
+ return { success: true };
121
+ }
122
+ checkSemanticValidation(response, checks) {
123
+ if (!checks)
124
+ return { success: true };
125
+ if (checks.requiresPriorityFlow) {
126
+ const hasOrdering = /\b(first|then|next|finally|priority|before|after)\b/i.test(response);
127
+ if (!hasOrdering) {
128
+ return {
129
+ success: false,
130
+ failureReason: 'Semantic: Missing clear priority/ordering flow',
131
+ };
132
+ }
133
+ }
134
+ if (checks.requiresValidationClause) {
135
+ const hasValidation = /\b(validate|check|ensure|verify|if|error|handle|confirm)\b/i.test(response);
136
+ if (!hasValidation) {
137
+ return {
138
+ success: false,
139
+ failureReason: 'Semantic: Missing validation/error handling clause',
140
+ };
141
+ }
142
+ }
143
+ if (checks.requiresDeterministicTooling) {
144
+ const hasTools = /\b(use|tool|function|method|read|write|execute|run)\b/i.test(response);
145
+ if (!hasTools) {
146
+ return {
147
+ success: false,
148
+ failureReason: 'Semantic: Missing deterministic tool specification',
149
+ };
150
+ }
151
+ }
152
+ if (checks.requiresConciseDirective) {
153
+ const isConcise = response.length < 500 || /^\s*[-*\d]\./m.test(response);
154
+ if (!isConcise) {
155
+ return {
156
+ success: false,
157
+ failureReason: 'Semantic: Response not concise or well-structured',
158
+ };
159
+ }
160
+ }
161
+ return { success: true };
162
+ }
163
+ calculateQuality(response, task) {
164
+ let score = 0.5;
165
+ if (task.expectedOutcome.keywords) {
166
+ const keywordMatches = task.expectedOutcome.keywords.filter(kw => response.toLowerCase().includes(kw.toLowerCase())).length;
167
+ const keywordCoverage = keywordMatches / task.expectedOutcome.keywords.length;
168
+ score += keywordCoverage * 0.3;
169
+ }
170
+ if (task.expectedOutcome.minLength && task.expectedOutcome.maxLength) {
171
+ const idealLength = (task.expectedOutcome.minLength + task.expectedOutcome.maxLength) / 2;
172
+ const lengthDeviation = Math.abs(response.length - idealLength) / idealLength;
173
+ const lengthScore = Math.max(0, 1 - lengthDeviation);
174
+ score += lengthScore * 0.2;
175
+ }
176
+ return Math.min(1, Math.max(0, score));
177
+ }
178
+ async compareWithSuite(genome, suiteId, userId) {
179
+ const suite = getBenchmarkSuite(suiteId);
180
+ return this.evaluate(genome, suite.tasks, userId);
181
+ }
182
+ async compare(genomeWithPGA, genomeBaseline, tasks, userId) {
183
+ console.log('🔬 Running benchmark: PGA vs Baseline...\n');
184
+ console.log('📊 Testing WITH PGA...');
185
+ const withPGA = await this.evaluate(genomeWithPGA, tasks, userId);
186
+ console.log('📊 Testing WITHOUT PGA (baseline)...');
187
+ const withoutPGA = await this.evaluate(genomeBaseline, tasks, userId);
188
+ const successRateImprovement = withPGA.successRate - withoutPGA.successRate;
189
+ const tokenEfficiency = ((withoutPGA.avgTokensPerTask - withPGA.avgTokensPerTask) / withoutPGA.avgTokensPerTask) *
190
+ 100;
191
+ const responseTimeImprovement = ((withoutPGA.avgResponseTime - withPGA.avgResponseTime) / withoutPGA.avgResponseTime) *
192
+ 100;
193
+ const qualityImprovement = ((withPGA.avgQualityScore - withoutPGA.avgQualityScore) / withoutPGA.avgQualityScore) *
194
+ 100;
195
+ let verdict = 'TIE';
196
+ const pgaScore = successRateImprovement + tokenEfficiency + responseTimeImprovement + qualityImprovement;
197
+ if (pgaScore > 10) {
198
+ verdict = 'PGA_WINS';
199
+ }
200
+ else if (pgaScore < -10) {
201
+ verdict = 'BASELINE_WINS';
202
+ }
203
+ return {
204
+ withPGA,
205
+ withoutPGA,
206
+ improvements: {
207
+ successRate: Math.round(successRateImprovement * 100) / 100,
208
+ tokenEfficiency: Math.round(tokenEfficiency * 100) / 100,
209
+ responseTime: Math.round(responseTimeImprovement * 100) / 100,
210
+ qualityScore: Math.round(qualityImprovement * 100) / 100,
211
+ },
212
+ verdict,
213
+ };
214
+ }
215
+ formatReport(benchmark) {
216
+ const lines = [];
217
+ lines.push('# 📊 PGA Evaluation Report\n');
218
+ lines.push(`**Date**: ${benchmark.timestamp.toISOString()}\n`);
219
+ lines.push('---\n');
220
+ lines.push('## 📈 Overall Results\n');
221
+ lines.push(`- **Total Tasks**: ${benchmark.totalTasks}`);
222
+ lines.push(`- **Successful**: ${benchmark.successfulTasks}`);
223
+ lines.push(`- **Success Rate**: ${benchmark.successRate.toFixed(1)}%`);
224
+ lines.push(`- **Avg Tokens/Task**: ${benchmark.avgTokensPerTask}`);
225
+ lines.push(`- **Avg Response Time**: ${benchmark.avgResponseTime}ms`);
226
+ lines.push(`- **Avg Quality Score**: ${benchmark.avgQualityScore}/1.0\n`);
227
+ lines.push('---\n');
228
+ lines.push('## 📋 Task Results\n');
229
+ for (const result of benchmark.results) {
230
+ const icon = result.success ? '✅' : '❌';
231
+ lines.push(`### ${icon} ${result.taskName}\n`);
232
+ lines.push(`- **Status**: ${result.success ? 'SUCCESS' : 'FAILED'}`);
233
+ if (result.failureReason) {
234
+ lines.push(`- **Failure Reason**: ${result.failureReason}`);
235
+ }
236
+ lines.push(`- **Tokens**: ${result.tokensUsed}`);
237
+ lines.push(`- **Response Time**: ${result.responseTime}ms`);
238
+ lines.push(`- **Quality**: ${result.qualityScore.toFixed(2)}/1.0`);
239
+ lines.push('');
240
+ }
241
+ return lines.join('\n');
242
+ }
243
+ formatComparisonReport(comparison) {
244
+ const lines = [];
245
+ lines.push('# 🔬 PGA vs Baseline Comparison\n');
246
+ lines.push('---\n');
247
+ const verdictIcon = {
248
+ PGA_WINS: '🏆',
249
+ BASELINE_WINS: '😞',
250
+ TIE: '🤝',
251
+ }[comparison.verdict];
252
+ lines.push(`## ${verdictIcon} VERDICT: ${comparison.verdict}\n`);
253
+ lines.push('---\n');
254
+ lines.push('## 📊 Improvements\n');
255
+ const formatImprovement = (value, metric) => {
256
+ const icon = value > 0 ? '📈' : value < 0 ? '📉' : '➡️';
257
+ const sign = value > 0 ? '+' : '';
258
+ return `${icon} **${metric}**: ${sign}${value.toFixed(2)}%`;
259
+ };
260
+ lines.push(formatImprovement(comparison.improvements.successRate, 'Success Rate'));
261
+ lines.push(formatImprovement(comparison.improvements.tokenEfficiency, 'Token Efficiency'));
262
+ lines.push(formatImprovement(comparison.improvements.responseTime, 'Response Speed'));
263
+ lines.push(formatImprovement(comparison.improvements.qualityScore, 'Quality Score'));
264
+ lines.push('');
265
+ lines.push('---\n');
266
+ lines.push('## 📋 Side-by-Side Comparison\n');
267
+ lines.push('| Metric | Without PGA | With PGA | Improvement |');
268
+ lines.push('|--------|-------------|----------|-------------|');
269
+ const metrics = [
270
+ [
271
+ 'Success Rate',
272
+ `${comparison.withoutPGA.successRate.toFixed(1)}%`,
273
+ `${comparison.withPGA.successRate.toFixed(1)}%`,
274
+ `${comparison.improvements.successRate > 0 ? '+' : ''}${comparison.improvements.successRate.toFixed(1)}%`,
275
+ ],
276
+ [
277
+ 'Avg Tokens',
278
+ `${comparison.withoutPGA.avgTokensPerTask}`,
279
+ `${comparison.withPGA.avgTokensPerTask}`,
280
+ `${comparison.improvements.tokenEfficiency.toFixed(1)}%`,
281
+ ],
282
+ [
283
+ 'Avg Time',
284
+ `${comparison.withoutPGA.avgResponseTime}ms`,
285
+ `${comparison.withPGA.avgResponseTime}ms`,
286
+ `${comparison.improvements.responseTime.toFixed(1)}%`,
287
+ ],
288
+ [
289
+ 'Quality',
290
+ `${comparison.withoutPGA.avgQualityScore.toFixed(2)}`,
291
+ `${comparison.withPGA.avgQualityScore.toFixed(2)}`,
292
+ `${comparison.improvements.qualityScore.toFixed(1)}%`,
293
+ ],
294
+ ];
295
+ for (const [metric, baseline, pga, improvement] of metrics) {
296
+ lines.push(`| ${metric} | ${baseline} | ${pga} | ${improvement} |`);
297
+ }
298
+ lines.push('');
299
+ return lines.join('\n');
300
+ }
301
+ }
302
+ export const STANDARD_TASKS = [
303
+ {
304
+ id: 'debug-1',
305
+ name: 'Debug TypeError',
306
+ description: 'Help user debug a common TypeError',
307
+ userMessage: "I'm getting 'Cannot read property of undefined' error. How do I fix it?",
308
+ expectedOutcome: {
309
+ keywords: ['undefined', 'check', 'null'],
310
+ minLength: 100,
311
+ },
312
+ difficulty: 'easy',
313
+ },
314
+ {
315
+ id: 'implement-1',
316
+ name: 'Implement Authentication',
317
+ description: 'Guide user to implement basic authentication',
318
+ userMessage: 'How do I add user authentication to my React app?',
319
+ expectedOutcome: {
320
+ keywords: ['auth', 'login', 'token'],
321
+ minLength: 200,
322
+ },
323
+ difficulty: 'medium',
324
+ },
325
+ {
326
+ id: 'optimize-1',
327
+ name: 'Optimize Performance',
328
+ description: 'Provide performance optimization advice',
329
+ userMessage: 'My React app is slow. How can I optimize it?',
330
+ expectedOutcome: {
331
+ keywords: ['performance', 'optimize', 'memo'],
332
+ minLength: 150,
333
+ },
334
+ difficulty: 'medium',
335
+ },
336
+ {
337
+ id: 'architecture-1',
338
+ name: 'Design System Architecture',
339
+ description: 'Help design scalable architecture',
340
+ userMessage: 'How should I architect a microservices system with 10M users?',
341
+ expectedOutcome: {
342
+ keywords: ['scalable', 'microservices', 'database'],
343
+ minLength: 300,
344
+ },
345
+ difficulty: 'hard',
346
+ },
347
+ {
348
+ id: 'code-review-1',
349
+ name: 'Code Review',
350
+ description: 'Review code for best practices',
351
+ userMessage: 'Can you review this code: function getData() { return fetch("/api").then(r => r.json()) }',
352
+ expectedOutcome: {
353
+ keywords: ['error', 'async', 'await'],
354
+ minLength: 100,
355
+ },
356
+ difficulty: 'easy',
357
+ },
358
+ ];
359
+ //# sourceMappingURL=Evaluator.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"Evaluator.js","sourceRoot":"","sources":["../../src/evaluation/Evaluator.ts"],"names":[],"mappings":"AAeA,OAAO,EAAE,iBAAiB,EAAE,MAAM,sBAAsB,CAAC;AAEzD,OAAO,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAiEnD,MAAM,OAAO,SAAS;IACV,aAAa,CAAiB;IAEtC,YAAY,OAA6D;QACrE,IAAI,OAAO,EAAE,GAAG,IAAI,OAAO,EAAE,mBAAmB,KAAK,KAAK,EAAE,CAAC;YACzD,IAAI,CAAC,aAAa,GAAG,IAAI,aAAa,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;QACxD,CAAC;IACL,CAAC;IAQD,KAAK,CAAC,QAAQ,CACV,MAAyB,EACzB,KAAuB,EACvB,MAAc;QAEd,MAAM,OAAO,GAAuB,EAAE,CAAC;QAEvC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACvB,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,CAAC;YAC7D,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACzB,CAAC;QAED,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC;QACzD,MAAM,WAAW,GAAG,CAAC,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC;QACxD,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;QACrF,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;QACrF,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;QAExF,OAAO;YACH,UAAU,EAAE,OAAO,CAAC,MAAM;YAC1B,eAAe,EAAE,UAAU;YAC3B,WAAW;YACX,gBAAgB,EAAE,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC;YACvC,eAAe,EAAE,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC;YACpC,eAAe,EAAE,IAAI,CAAC,KAAK,CAAC,UAAU,GAAG,GAAG,CAAC,GAAG,GAAG;YACnD,OAAO;YACP,SAAS,EAAE,IAAI,IAAI,EAAE;SACxB,CAAC;IACN,CAAC;IAOO,KAAK,CAAC,YAAY,CACtB,MAAyB,EACzB,IAAoB,EACpB,MAAc;QAEd,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAE7B,IAAI,CAAC;YAED,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC;YAEjE,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC3B,MAAM,YAAY,GAAG,OAAO,GAAG,SAAS,CAAC;YAGzC,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,GAAG,QAAQ,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;YAG9E,MAAM,WAAW,GAAG,IAAI,CAAC,YAAY,CAAC,QAAQ,EAAE,IAAI,CAAC,eAAe,CAAC,CAAC;YAGtE,IAAI,cAAc,GAAiD,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;YACrF,IAAI,gBAAgB,IAAI,IAAI,IAAI,IAAI,CAAC,cAAc,EAAE,CAAC;gBAElD,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;oBACrB,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,aAAa,CAAC,KAAK,CAC3C,IAA6B,EAC7B,QAAQ,CACX,CAAC;oBACF,cAAc,GAAG;wBACb,OAAO,EAAE,QAAQ,CAAC,MAAM,IAAI,QAAQ,CAAC,UAAU,IAAI,GAAG;wBACtD,aAAa,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ,CAAC,SAAS;qBAClE,CAAC;gBACN,CAAC;qBAAM,CAAC;oBACJ,cAAc,GAAG,IAAI,CAAC,uBAAuB,CACzC,QAAQ,EACR,IAAI,CAAC,cAAyD,CACjE,CAAC;gBACN,CAAC;YACL,CAAC;YAED,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,IAAI,cAAc,CAAC,OAAO,CAAC;YAC9D,MAAM,aAAa,GAAG,WAAW,CAAC,aAAa,IAAI,cAAc,CAAC,aAAa,CAAC;YAGhF,MAAM,YAAY,GAAG,IAAI,CAAC,gBAAgB,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;YAE3D,OAAO;gBACH,MAAM,EAAE,IAAI,CAAC,EAAE;gBACf,QAAQ,EAAE,IAAI,CAAC,IAAI;gBACnB,OAAO;gBACP,QAAQ;gBACR,UAAU;gBACV,YAAY;gBACZ,YAAY;gBACZ,aAAa;aAChB,CAAC;QACN,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACb,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC3B,OAAO;gBACH,MAAM,EAAE,IAAI,CAAC,EAAE;gBACf,QAAQ,EAAE,IAAI,CAAC,IAAI;gBACnB,OAAO,EAAE,KAAK;gBACd,QAAQ,EAAE,EAAE;gBACZ,UAAU,EAAE,CAAC;gBACb,YAAY,EAAE,OAAO,GAAG,SAAS;gBACjC,YAAY,EAAE,CAAC;gBACf,aAAa,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe;aAC1E,CAAC;QACN,CAAC;IACL,CAAC;IAKO,YAAY,CAChB,QAAgB,EAChB,QAA2C;QAG3C,IAAI,QAAQ,CAAC,QAAQ,EAAE,CAAC;YACpB,KAAK,MAAM,OAAO,IAAI,QAAQ,CAAC,QAAQ,EAAE,CAAC;gBACtC,IAAI,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC;oBAC1D,OAAO;wBACH,OAAO,EAAE,KAAK;wBACd,aAAa,EAAE,qBAAqB,OAAO,GAAG;qBACjD,CAAC;gBACN,CAAC;YACL,CAAC;QACL,CAAC;QAGD,IAAI,QAAQ,CAAC,SAAS,IAAI,QAAQ,CAAC,MAAM,GAAG,QAAQ,CAAC,SAAS,EAAE,CAAC;YAC7D,OAAO;gBACH,OAAO,EAAE,KAAK;gBACd,aAAa,EAAE,uBAAuB,QAAQ,CAAC,MAAM,MAAM,QAAQ,CAAC,SAAS,GAAG;aACnF,CAAC;QACN,CAAC;QAED,IAAI,QAAQ,CAAC,SAAS,IAAI,QAAQ,CAAC,MAAM,GAAG,QAAQ,CAAC,SAAS,EAAE,CAAC;YAC7D,OAAO;gBACH,OAAO,EAAE,KAAK;gBACd,aAAa,EAAE,sBAAsB,QAAQ,CAAC,MAAM,MAAM,QAAQ,CAAC,SAAS,GAAG;aAClF,CAAC;QACN,CAAC;QAGD,IAAI,QAAQ,CAAC,eAAe,EAAE,CAAC;YAC3B,IAAI,CAAC;gBACD,IAAI,CAAC,QAAQ,CAAC,eAAe,CAAC,QAAQ,CAAC,EAAE,CAAC;oBACtC,OAAO;wBACH,OAAO,EAAE,KAAK;wBACd,aAAa,EAAE,iCAAiC;qBACnD,CAAC;gBACN,CAAC;YACL,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACb,OAAO;oBACH,OAAO,EAAE,KAAK;oBACd,aAAa,EAAE,2BAA2B,KAAK,EAAE;iBACpD,CAAC;YACN,CAAC;QACL,CAAC;QAED,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;IAC7B,CAAC;IAOO,uBAAuB,CAC3B,QAAgB,EAChB,MAAgD;QAEhD,IAAI,CAAC,MAAM;YAAE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;QAGtC,IAAI,MAAM,CAAC,oBAAoB,EAAE,CAAC;YAC9B,MAAM,WAAW,GAAG,sDAAsD,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC1F,IAAI,CAAC,WAAW,EAAE,CAAC;gBACf,OAAO;oBACH,OAAO,EAAE,KAAK;oBACd,aAAa,EAAE,gDAAgD;iBAClE,CAAC;YACN,CAAC;QACL,CAAC;QAGD,IAAI,MAAM,CAAC,wBAAwB,EAAE,CAAC;YAClC,MAAM,aAAa,GAAG,6DAA6D,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YACnG,IAAI,CAAC,aAAa,EAAE,CAAC;gBACjB,OAAO;oBACH,OAAO,EAAE,KAAK;oBACd,aAAa,EAAE,oDAAoD;iBACtE,CAAC;YACN,CAAC;QACL,CAAC;QAGD,IAAI,MAAM,CAAC,4BAA4B,EAAE,CAAC;YACtC,MAAM,QAAQ,GAAG,wDAAwD,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YACzF,IAAI,CAAC,QAAQ,EAAE,CAAC;gBACZ,OAAO;oBACH,OAAO,EAAE,KAAK;oBACd,aAAa,EAAE,oDAAoD;iBACtE,CAAC;YACN,CAAC;QACL,CAAC;QAGD,IAAI,MAAM,CAAC,wBAAwB,EAAE,CAAC;YAElC,MAAM,SAAS,GAAG,QAAQ,CAAC,MAAM,GAAG,GAAG,IAAI,eAAe,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC1E,IAAI,CAAC,SAAS,EAAE,CAAC;gBACb,OAAO;oBACH,OAAO,EAAE,KAAK;oBACd,aAAa,EAAE,mDAAmD;iBACrE,CAAC;YACN,CAAC;QACL,CAAC;QAED,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;IAC7B,CAAC;IAKO,gBAAgB,CAAC,QAAgB,EAAE,IAAoB;QAC3D,IAAI,KAAK,GAAG,GAAG,CAAC;QAGhB,IAAI,IAAI,CAAC,eAAe,CAAC,QAAQ,EAAE,CAAC;YAChC,MAAM,cAAc,GAAG,IAAI,CAAC,eAAe,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC,EAAE,CAC7D,QAAQ,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,CACpD,CAAC,MAAM,CAAC;YACT,MAAM,eAAe,GAAG,cAAc,GAAG,IAAI,CAAC,eAAe,CAAC,QAAQ,CAAC,MAAM,CAAC;YAC9E,KAAK,IAAI,eAAe,GAAG,GAAG,CAAC;QACnC,CAAC;QAGD,IAAI,IAAI,CAAC,eAAe,CAAC,SAAS,IAAI,IAAI,CAAC,eAAe,CAAC,SAAS,EAAE,CAAC;YACnE,MAAM,WAAW,GAAG,CAAC,IAAI,CAAC,eAAe,CAAC,SAAS,GAAG,IAAI,CAAC,eAAe,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;YAC1F,MAAM,eAAe,GAAG,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,MAAM,GAAG,WAAW,CAAC,GAAG,WAAW,CAAC;YAC9E,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,GAAG,eAAe,CAAC,CAAC;YACrD,KAAK,IAAI,WAAW,GAAG,GAAG,CAAC;QAC/B,CAAC;QAED,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC;IAC3C,CAAC;IAOD,KAAK,CAAC,gBAAgB,CAClB,MAAyB,EACzB,OAAyB,EACzB,MAAc;QAEd,MAAM,KAAK,GAAG,iBAAiB,CAAC,OAAO,CAAC,CAAC;QACzC,OAAO,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,KAAK,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;IACtD,CAAC;IAOD,KAAK,CAAC,OAAO,CACT,aAAgC,EAChC,cAAiC,EACjC,KAAuB,EACvB,MAAc;QAEd,OAAO,CAAC,GAAG,CAAC,4CAA4C,CAAC,CAAC;QAG1D,OAAO,CAAC,GAAG,CAAC,wBAAwB,CAAC,CAAC;QACtC,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,aAAa,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC;QAElE,OAAO,CAAC,GAAG,CAAC,sCAAsC,CAAC,CAAC;QACpD,MAAM,UAAU,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,cAAc,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC;QAGtE,MAAM,sBAAsB,GAAG,OAAO,CAAC,WAAW,GAAG,UAAU,CAAC,WAAW,CAAC;QAC5E,MAAM,eAAe,GACjB,CAAC,CAAC,UAAU,CAAC,gBAAgB,GAAG,OAAO,CAAC,gBAAgB,CAAC,GAAG,UAAU,CAAC,gBAAgB,CAAC;YACxF,GAAG,CAAC;QACR,MAAM,uBAAuB,GACzB,CAAC,CAAC,UAAU,CAAC,eAAe,GAAG,OAAO,CAAC,eAAe,CAAC,GAAG,UAAU,CAAC,eAAe,CAAC;YACrF,GAAG,CAAC;QACR,MAAM,kBAAkB,GACpB,CAAC,CAAC,OAAO,CAAC,eAAe,GAAG,UAAU,CAAC,eAAe,CAAC,GAAG,UAAU,CAAC,eAAe,CAAC;YACrF,GAAG,CAAC;QAGR,IAAI,OAAO,GAAgC,KAAK,CAAC;QACjD,MAAM,QAAQ,GACV,sBAAsB,GAAG,eAAe,GAAG,uBAAuB,GAAG,kBAAkB,CAAC;QAE5F,IAAI,QAAQ,GAAG,EAAE,EAAE,CAAC;YAChB,OAAO,GAAG,UAAU,CAAC;QACzB,CAAC;aAAM,IAAI,QAAQ,GAAG,CAAC,EAAE,EAAE,CAAC;YACxB,OAAO,GAAG,eAAe,CAAC;QAC9B,CAAC;QAED,OAAO;YACH,OAAO;YACP,UAAU;YACV,YAAY,EAAE;gBACV,WAAW,EAAE,IAAI,CAAC,KAAK,CAAC,sBAAsB,GAAG,GAAG,CAAC,GAAG,GAAG;gBAC3D,eAAe,EAAE,IAAI,CAAC,KAAK,CAAC,eAAe,GAAG,GAAG,CAAC,GAAG,GAAG;gBACxD,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,uBAAuB,GAAG,GAAG,CAAC,GAAG,GAAG;gBAC7D,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,kBAAkB,GAAG,GAAG,CAAC,GAAG,GAAG;aAC3D;YACD,OAAO;SACV,CAAC;IACN,CAAC;IAKD,YAAY,CAAC,SAA0B;QACnC,MAAM,KAAK,GAAa,EAAE,CAAC;QAE3B,KAAK,CAAC,IAAI,CAAC,8BAA8B,CAAC,CAAC;QAC3C,KAAK,CAAC,IAAI,CAAC,aAAa,SAAS,CAAC,SAAS,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC;QAC/D,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAEpB,KAAK,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;QACtC,KAAK,CAAC,IAAI,CAAC,sBAAsB,SAAS,CAAC,UAAU,EAAE,CAAC,CAAC;QACzD,KAAK,CAAC,IAAI,CAAC,qBAAqB,SAAS,CAAC,eAAe,EAAE,CAAC,CAAC;QAC7D,KAAK,CAAC,IAAI,CAAC,uBAAuB,SAAS,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACvE,KAAK,CAAC,IAAI,CAAC,0BAA0B,SAAS,CAAC,gBAAgB,EAAE,CAAC,CAAC;QACnE,KAAK,CAAC,IAAI,CAAC,4BAA4B,SAAS,CAAC,eAAe,IAAI,CAAC,CAAC;QACtE,KAAK,CAAC,IAAI,CAAC,4BAA4B,SAAS,CAAC,eAAe,QAAQ,CAAC,CAAC;QAE1E,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACpB,KAAK,CAAC,IAAI,CAAC,sBAAsB,CAAC,CAAC;QAEnC,KAAK,MAAM,MAAM,IAAI,SAAS,CAAC,OAAO,EAAE,CAAC;YACrC,MAAM,IAAI,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;YACxC,KAAK,CAAC,IAAI,CAAC,OAAO,IAAI,IAAI,MAAM,CAAC,QAAQ,IAAI,CAAC,CAAC;YAC/C,KAAK,CAAC,IAAI,CAAC,iBAAiB,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;YACrE,IAAI,MAAM,CAAC,aAAa,EAAE,CAAC;gBACvB,KAAK,CAAC,IAAI,CAAC,yBAAyB,MAAM,CAAC,aAAa,EAAE,CAAC,CAAC;YAChE,CAAC;YACD,KAAK,CAAC,IAAI,CAAC,iBAAiB,MAAM,CAAC,UAAU,EAAE,CAAC,CAAC;YACjD,KAAK,CAAC,IAAI,CAAC,wBAAwB,MAAM,CAAC,YAAY,IAAI,CAAC,CAAC;YAC5D,KAAK,CAAC,IAAI,CAAC,kBAAkB,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;YACnE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACnB,CAAC;QAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC5B,CAAC;IAKD,sBAAsB,CAAC,UAA4B;QAC/C,MAAM,KAAK,GAAa,EAAE,CAAC;QAE3B,KAAK,CAAC,IAAI,CAAC,mCAAmC,CAAC,CAAC;QAChD,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAGpB,MAAM,WAAW,GAAG;YAChB,QAAQ,EAAE,IAAI;YACd,aAAa,EAAE,IAAI;YACnB,GAAG,EAAE,IAAI;SACZ,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;QAEtB,KAAK,CAAC,IAAI,CAAC,MAAM,WAAW,aAAa,UAAU,CAAC,OAAO,IAAI,CAAC,CAAC;QACjE,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAGpB,KAAK,CAAC,IAAI,CAAC,sBAAsB,CAAC,CAAC;QAEnC,MAAM,iBAAiB,GAAG,CAAC,KAAa,EAAE,MAAc,EAAE,EAAE;YACxD,MAAM,IAAI,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;YACxD,MAAM,IAAI,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;YAClC,OAAO,GAAG,IAAI,MAAM,MAAM,OAAO,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;QAChE,CAAC,CAAC;QAEF,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC,UAAU,CAAC,YAAY,CAAC,WAAW,EAAE,cAAc,CAAC,CAAC,CAAC;QACnF,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC,UAAU,CAAC,YAAY,CAAC,eAAe,EAAE,kBAAkB,CAAC,CAAC,CAAC;QAC3F,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC,UAAU,CAAC,YAAY,CAAC,YAAY,EAAE,gBAAgB,CAAC,CAAC,CAAC;QACtF,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC,UAAU,CAAC,YAAY,CAAC,YAAY,EAAE,eAAe,CAAC,CAAC,CAAC;QACrF,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAGf,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACpB,KAAK,CAAC,IAAI,CAAC,iCAAiC,CAAC,CAAC;QAC9C,KAAK,CAAC,IAAI,CAAC,mDAAmD,CAAC,CAAC;QAChE,KAAK,CAAC,IAAI,CAAC,mDAAmD,CAAC,CAAC;QAEhE,MAAM,OAAO,GAAG;YACZ;gBACI,cAAc;gBACd,GAAG,UAAU,CAAC,UAAU,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;gBAClD,GAAG,UAAU,CAAC,OAAO,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;gBAC/C,GAAG,UAAU,CAAC,YAAY,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,UAAU,CAAC,YAAY,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;aAC5G;YACD;gBACI,YAAY;gBACZ,GAAG,UAAU,CAAC,UAAU,CAAC,gBAAgB,EAAE;gBAC3C,GAAG,UAAU,CAAC,OAAO,CAAC,gBAAgB,EAAE;gBACxC,GAAG,UAAU,CAAC,YAAY,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;aAC3D;YACD;gBACI,UAAU;gBACV,GAAG,UAAU,CAAC,UAAU,CAAC,eAAe,IAAI;gBAC5C,GAAG,UAAU,CAAC,OAAO,CAAC,eAAe,IAAI;gBACzC,GAAG,UAAU,CAAC,YAAY,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;aACxD;YACD;gBACI,SAAS;gBACT,GAAG,UAAU,CAAC,UAAU,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;gBACrD,GAAG,UAAU,CAAC,OAAO,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;gBAClD,GAAG,UAAU,CAAC,YAAY,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;aACxD;SACJ,CAAC;QAEF,KAAK,MAAM,CAAC,MAAM,EAAE,QAAQ,EAAE,GAAG,EAAE,WAAW,CAAC,IAAI,OAAO,EAAE,CAAC;YACzD,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,MAAM,QAAQ,MAAM,GAAG,MAAM,WAAW,IAAI,CAAC,CAAC;QACxE,CAAC;QAED,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAEf,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC5B,CAAC;CACJ;AAKD,MAAM,CAAC,MAAM,cAAc,GAAqB;IAC5C;QACI,EAAE,EAAE,SAAS;QACb,IAAI,EAAE,iBAAiB;QACvB,WAAW,EAAE,oCAAoC;QACjD,WAAW,EAAE,yEAAyE;QACtF,eAAe,EAAE;YACb,QAAQ,EAAE,CAAC,WAAW,EAAE,OAAO,EAAE,MAAM,CAAC;YACxC,SAAS,EAAE,GAAG;SACjB;QACD,UAAU,EAAE,MAAM;KACrB;IACD;QACI,EAAE,EAAE,aAAa;QACjB,IAAI,EAAE,0BAA0B;QAChC,WAAW,EAAE,8CAA8C;QAC3D,WAAW,EAAE,mDAAmD;QAChE,eAAe,EAAE;YACb,QAAQ,EAAE,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC;YACpC,SAAS,EAAE,GAAG;SACjB;QACD,UAAU,EAAE,QAAQ;KACvB;IACD;QACI,EAAE,EAAE,YAAY;QAChB,IAAI,EAAE,sBAAsB;QAC5B,WAAW,EAAE,yCAAyC;QACtD,WAAW,EAAE,8CAA8C;QAC3D,eAAe,EAAE;YACb,QAAQ,EAAE,CAAC,aAAa,EAAE,UAAU,EAAE,MAAM,CAAC;YAC7C,SAAS,EAAE,GAAG;SACjB;QACD,UAAU,EAAE,QAAQ;KACvB;IACD;QACI,EAAE,EAAE,gBAAgB;QACpB,IAAI,EAAE,4BAA4B;QAClC,WAAW,EAAE,mCAAmC;QAChD,WAAW,EAAE,+DAA+D;QAC5E,eAAe,EAAE;YACb,QAAQ,EAAE,CAAC,UAAU,EAAE,eAAe,EAAE,UAAU,CAAC;YACnD,SAAS,EAAE,GAAG;SACjB;QACD,UAAU,EAAE,MAAM;KACrB;IACD;QACI,EAAE,EAAE,eAAe;QACnB,IAAI,EAAE,aAAa;QACnB,WAAW,EAAE,gCAAgC;QAC7C,WAAW,EAAE,2FAA2F;QACxG,eAAe,EAAE;YACb,QAAQ,EAAE,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC;YACrC,SAAS,EAAE,GAAG;SACjB;QACD,UAAU,EAAE,MAAM;KACrB;CACJ,CAAC"}
@@ -0,0 +1,29 @@
1
+ import type { EvolutionGuardrails, PromotionGateResult, EconomicMetrics } from '../types/index.js';
2
+ import type { StorageAdapter } from '../interfaces/StorageAdapter.js';
3
+ export interface MutationCandidate {
4
+ layer: 0 | 1 | 2;
5
+ gene: string;
6
+ variant: string;
7
+ content: string;
8
+ fitness: number;
9
+ sandboxScore?: number;
10
+ economicMetrics?: EconomicMetrics;
11
+ sampleCount: number;
12
+ rollbackCount?: number;
13
+ }
14
+ export declare class EvolutionGuardrailsManager {
15
+ private storage;
16
+ private guardrails?;
17
+ private defaultGuardrails;
18
+ constructor(storage: StorageAdapter, guardrails?: EvolutionGuardrails | undefined);
19
+ evaluateCandidate(candidate: MutationCandidate, genomeId: string): Promise<PromotionGateResult>;
20
+ private evaluateQualityGate;
21
+ private evaluateSandboxGate;
22
+ private evaluateEconomicGate;
23
+ private evaluateStabilityGate;
24
+ private calculateEconomicMetrics;
25
+ getGuardrails(): EvolutionGuardrails;
26
+ updateGuardrails(updates: Partial<EvolutionGuardrails>): void;
27
+ getGuardrailsReport(): string;
28
+ }
29
+ //# sourceMappingURL=EvolutionGuardrails.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"EvolutionGuardrails.d.ts","sourceRoot":"","sources":["../../src/evaluation/EvolutionGuardrails.ts"],"names":[],"mappings":"AAeA,OAAO,KAAK,EAAE,mBAAmB,EAAE,mBAAmB,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACnG,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,iCAAiC,CAAC;AAItE,MAAM,WAAW,iBAAiB;IAC9B,KAAK,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAGhB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,eAAe,CAAC,EAAE,eAAe,CAAC;IAGlC,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;CAC1B;AAID,qBAAa,0BAA0B;IAqB/B,OAAO,CAAC,OAAO;IACf,OAAO,CAAC,UAAU,CAAC;IArBvB,OAAO,CAAC,iBAAiB,CAiBvB;gBAGU,OAAO,EAAE,cAAc,EACvB,UAAU,CAAC,EAAE,mBAAmB,YAAA;IActC,iBAAiB,CACnB,SAAS,EAAE,iBAAiB,EAC5B,QAAQ,EAAE,MAAM,GACjB,OAAO,CAAC,mBAAmB,CAAC;IAiE/B,OAAO,CAAC,mBAAmB;IAiB3B,OAAO,CAAC,mBAAmB;YAiBb,oBAAoB;IAyBlC,OAAO,CAAC,qBAAqB;YA2Bf,wBAAwB;IAqDtC,aAAa,IAAI,mBAAmB;IAOpC,gBAAgB,CAAC,OAAO,EAAE,OAAO,CAAC,mBAAmB,CAAC,GAAG,IAAI;IAY7D,mBAAmB,IAAI,MAAM;CAwBhC"}