oh-my-knowledge 0.0.1 → 0.18.0

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 (215) hide show
  1. package/LICENSE +21 -0
  2. package/README.md +537 -332
  3. package/dist/src/analysis/coverage-analyzer.d.ts +57 -0
  4. package/dist/src/analysis/coverage-analyzer.d.ts.map +1 -0
  5. package/dist/src/analysis/coverage-analyzer.js +262 -0
  6. package/dist/src/analysis/coverage-analyzer.js.map +1 -0
  7. package/dist/src/analysis/gap-analyzer.d.ts +114 -0
  8. package/dist/src/analysis/gap-analyzer.d.ts.map +1 -0
  9. package/dist/src/analysis/gap-analyzer.js +439 -0
  10. package/dist/src/analysis/gap-analyzer.js.map +1 -0
  11. package/dist/src/analysis/hedging-classifier.d.ts +34 -0
  12. package/dist/src/analysis/hedging-classifier.d.ts.map +1 -0
  13. package/dist/src/analysis/hedging-classifier.js +144 -0
  14. package/dist/src/analysis/hedging-classifier.js.map +1 -0
  15. package/dist/src/analysis/report-diagnostics.d.ts +9 -0
  16. package/dist/src/analysis/report-diagnostics.d.ts.map +1 -0
  17. package/dist/src/analysis/report-diagnostics.js +631 -0
  18. package/dist/src/analysis/report-diagnostics.js.map +1 -0
  19. package/dist/src/authoring/evolver.d.ts +64 -0
  20. package/dist/src/authoring/evolver.d.ts.map +1 -0
  21. package/dist/src/authoring/evolver.js +275 -0
  22. package/dist/src/authoring/evolver.js.map +1 -0
  23. package/dist/src/authoring/generator.d.ts +13 -0
  24. package/dist/src/authoring/generator.d.ts.map +1 -0
  25. package/dist/src/authoring/generator.js +60 -0
  26. package/dist/src/authoring/generator.js.map +1 -0
  27. package/dist/src/cli.d.ts +3 -0
  28. package/dist/src/cli.d.ts.map +1 -0
  29. package/dist/src/cli.js +1004 -0
  30. package/dist/src/cli.js.map +1 -0
  31. package/dist/src/eval-core/cache.d.ts +11 -0
  32. package/dist/src/eval-core/cache.d.ts.map +1 -0
  33. package/dist/src/eval-core/cache.js +53 -0
  34. package/dist/src/eval-core/cache.js.map +1 -0
  35. package/dist/src/eval-core/ci-gates.d.ts +17 -0
  36. package/dist/src/eval-core/ci-gates.d.ts.map +1 -0
  37. package/dist/src/eval-core/ci-gates.js +42 -0
  38. package/dist/src/eval-core/ci-gates.js.map +1 -0
  39. package/dist/src/eval-core/dependency-checker.d.ts +37 -0
  40. package/dist/src/eval-core/dependency-checker.d.ts.map +1 -0
  41. package/dist/src/eval-core/dependency-checker.js +280 -0
  42. package/dist/src/eval-core/dependency-checker.js.map +1 -0
  43. package/dist/src/eval-core/evaluation-execution.d.ts +26 -0
  44. package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -0
  45. package/dist/src/eval-core/evaluation-execution.js +196 -0
  46. package/dist/src/eval-core/evaluation-execution.js.map +1 -0
  47. package/dist/src/eval-core/evaluation-job.d.ts +48 -0
  48. package/dist/src/eval-core/evaluation-job.d.ts.map +1 -0
  49. package/dist/src/eval-core/evaluation-job.js +112 -0
  50. package/dist/src/eval-core/evaluation-job.js.map +1 -0
  51. package/dist/src/eval-core/evaluation-reporting.d.ts +28 -0
  52. package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -0
  53. package/dist/src/eval-core/evaluation-reporting.js +123 -0
  54. package/dist/src/eval-core/evaluation-reporting.js.map +1 -0
  55. package/dist/src/eval-core/execution-strategy.d.ts +11 -0
  56. package/dist/src/eval-core/execution-strategy.d.ts.map +1 -0
  57. package/dist/src/eval-core/execution-strategy.js +121 -0
  58. package/dist/src/eval-core/execution-strategy.js.map +1 -0
  59. package/dist/src/eval-core/fact-checker.d.ts +25 -0
  60. package/dist/src/eval-core/fact-checker.d.ts.map +1 -0
  61. package/dist/src/eval-core/fact-checker.js +66 -0
  62. package/dist/src/eval-core/fact-checker.js.map +1 -0
  63. package/dist/src/eval-core/schema.d.ts +30 -0
  64. package/dist/src/eval-core/schema.d.ts.map +1 -0
  65. package/dist/src/eval-core/schema.js +195 -0
  66. package/dist/src/eval-core/schema.js.map +1 -0
  67. package/dist/src/eval-core/statistics.d.ts +86 -0
  68. package/dist/src/eval-core/statistics.d.ts.map +1 -0
  69. package/dist/src/eval-core/statistics.js +210 -0
  70. package/dist/src/eval-core/statistics.js.map +1 -0
  71. package/dist/src/eval-core/task-planner.d.ts +4 -0
  72. package/dist/src/eval-core/task-planner.d.ts.map +1 -0
  73. package/dist/src/eval-core/task-planner.js +33 -0
  74. package/dist/src/eval-core/task-planner.js.map +1 -0
  75. package/dist/src/eval-workflows/each-evaluation-workflow.d.ts +133 -0
  76. package/dist/src/eval-workflows/each-evaluation-workflow.d.ts.map +1 -0
  77. package/dist/src/eval-workflows/each-evaluation-workflow.js +169 -0
  78. package/dist/src/eval-workflows/each-evaluation-workflow.js.map +1 -0
  79. package/dist/src/eval-workflows/evaluation-pipeline.d.ts +38 -0
  80. package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -0
  81. package/dist/src/eval-workflows/evaluation-pipeline.js +214 -0
  82. package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -0
  83. package/dist/src/eval-workflows/evaluation-preparation.d.ts +64 -0
  84. package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -0
  85. package/dist/src/eval-workflows/evaluation-preparation.js +90 -0
  86. package/dist/src/eval-workflows/evaluation-preparation.js.map +1 -0
  87. package/dist/src/eval-workflows/run-evaluation.d.ts +105 -0
  88. package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -0
  89. package/dist/src/eval-workflows/run-evaluation.js +264 -0
  90. package/dist/src/eval-workflows/run-evaluation.js.map +1 -0
  91. package/dist/src/executors/anthropic-api.d.ts +3 -0
  92. package/dist/src/executors/anthropic-api.d.ts.map +1 -0
  93. package/dist/src/executors/anthropic-api.js +43 -0
  94. package/dist/src/executors/anthropic-api.js.map +1 -0
  95. package/dist/src/executors/claude-cli.d.ts +3 -0
  96. package/dist/src/executors/claude-cli.d.ts.map +1 -0
  97. package/dist/src/executors/claude-cli.js +103 -0
  98. package/dist/src/executors/claude-cli.js.map +1 -0
  99. package/dist/src/executors/claude-sdk-trace.d.ts +10 -0
  100. package/dist/src/executors/claude-sdk-trace.d.ts.map +1 -0
  101. package/dist/src/executors/claude-sdk-trace.js +100 -0
  102. package/dist/src/executors/claude-sdk-trace.js.map +1 -0
  103. package/dist/src/executors/claude-sdk.d.ts +3 -0
  104. package/dist/src/executors/claude-sdk.d.ts.map +1 -0
  105. package/dist/src/executors/claude-sdk.js +160 -0
  106. package/dist/src/executors/claude-sdk.js.map +1 -0
  107. package/dist/src/executors/gemini.d.ts +3 -0
  108. package/dist/src/executors/gemini.d.ts.map +1 -0
  109. package/dist/src/executors/gemini.js +78 -0
  110. package/dist/src/executors/gemini.js.map +1 -0
  111. package/dist/src/executors/index.d.ts +7 -0
  112. package/dist/src/executors/index.d.ts.map +1 -0
  113. package/dist/src/executors/index.js +22 -0
  114. package/dist/src/executors/index.js.map +1 -0
  115. package/dist/src/executors/openai-api.d.ts +3 -0
  116. package/dist/src/executors/openai-api.d.ts.map +1 -0
  117. package/dist/src/executors/openai-api.js +40 -0
  118. package/dist/src/executors/openai-api.js.map +1 -0
  119. package/dist/src/executors/openai-cli.d.ts +3 -0
  120. package/dist/src/executors/openai-cli.d.ts.map +1 -0
  121. package/dist/src/executors/openai-cli.js +60 -0
  122. package/dist/src/executors/openai-cli.js.map +1 -0
  123. package/dist/src/executors/script.d.ts +3 -0
  124. package/dist/src/executors/script.d.ts.map +1 -0
  125. package/dist/src/executors/script.js +63 -0
  126. package/dist/src/executors/script.js.map +1 -0
  127. package/dist/src/executors/shared.d.ts +117 -0
  128. package/dist/src/executors/shared.d.ts.map +1 -0
  129. package/dist/src/executors/shared.js +49 -0
  130. package/dist/src/executors/shared.js.map +1 -0
  131. package/dist/src/grading/assertions.d.ts +18 -0
  132. package/dist/src/grading/assertions.d.ts.map +1 -0
  133. package/dist/src/grading/assertions.js +239 -0
  134. package/dist/src/grading/assertions.js.map +1 -0
  135. package/dist/src/grading/index.d.ts +26 -0
  136. package/dist/src/grading/index.d.ts.map +1 -0
  137. package/dist/src/grading/index.js +98 -0
  138. package/dist/src/grading/index.js.map +1 -0
  139. package/dist/src/grading/judge.d.ts +13 -0
  140. package/dist/src/grading/judge.d.ts.map +1 -0
  141. package/dist/src/grading/judge.js +98 -0
  142. package/dist/src/grading/judge.js.map +1 -0
  143. package/dist/src/grading/layered-scores.d.ts +13 -0
  144. package/dist/src/grading/layered-scores.d.ts.map +1 -0
  145. package/dist/src/grading/layered-scores.js +62 -0
  146. package/dist/src/grading/layered-scores.js.map +1 -0
  147. package/dist/src/inputs/eval-config.d.ts +13 -0
  148. package/dist/src/inputs/eval-config.d.ts.map +1 -0
  149. package/dist/src/inputs/eval-config.js +136 -0
  150. package/dist/src/inputs/eval-config.js.map +1 -0
  151. package/dist/src/inputs/load-samples.d.ts +16 -0
  152. package/dist/src/inputs/load-samples.d.ts.map +1 -0
  153. package/dist/src/inputs/load-samples.js +51 -0
  154. package/dist/src/inputs/load-samples.js.map +1 -0
  155. package/dist/src/inputs/mcp-resolver.d.ts +50 -0
  156. package/dist/src/inputs/mcp-resolver.d.ts.map +1 -0
  157. package/dist/src/inputs/mcp-resolver.js +307 -0
  158. package/dist/src/inputs/mcp-resolver.js.map +1 -0
  159. package/dist/src/inputs/skill-loader.d.ts +18 -0
  160. package/dist/src/inputs/skill-loader.d.ts.map +1 -0
  161. package/dist/src/inputs/skill-loader.js +222 -0
  162. package/dist/src/inputs/skill-loader.js.map +1 -0
  163. package/dist/src/inputs/url-fetcher.d.ts +19 -0
  164. package/dist/src/inputs/url-fetcher.d.ts.map +1 -0
  165. package/dist/src/inputs/url-fetcher.js +241 -0
  166. package/dist/src/inputs/url-fetcher.js.map +1 -0
  167. package/dist/src/observability/production-analyzer.d.ts +61 -0
  168. package/dist/src/observability/production-analyzer.d.ts.map +1 -0
  169. package/dist/src/observability/production-analyzer.js +183 -0
  170. package/dist/src/observability/production-analyzer.js.map +1 -0
  171. package/dist/src/observability/trace-adapter.d.ts +75 -0
  172. package/dist/src/observability/trace-adapter.d.ts.map +1 -0
  173. package/dist/src/observability/trace-adapter.js +341 -0
  174. package/dist/src/observability/trace-adapter.js.map +1 -0
  175. package/dist/src/renderer/html-renderer.d.ts +9 -0
  176. package/dist/src/renderer/html-renderer.d.ts.map +1 -0
  177. package/dist/src/renderer/html-renderer.js +338 -0
  178. package/dist/src/renderer/html-renderer.js.map +1 -0
  179. package/dist/src/renderer/layout.d.ts +13 -0
  180. package/dist/src/renderer/layout.d.ts.map +1 -0
  181. package/dist/src/renderer/layout.js +476 -0
  182. package/dist/src/renderer/layout.js.map +1 -0
  183. package/dist/src/renderer/skill-health-renderer.d.ts +20 -0
  184. package/dist/src/renderer/skill-health-renderer.d.ts.map +1 -0
  185. package/dist/src/renderer/skill-health-renderer.js +218 -0
  186. package/dist/src/renderer/skill-health-renderer.js.map +1 -0
  187. package/dist/src/renderer/summary.d.ts +21 -0
  188. package/dist/src/renderer/summary.d.ts.map +1 -0
  189. package/dist/src/renderer/summary.js +1072 -0
  190. package/dist/src/renderer/summary.js.map +1 -0
  191. package/dist/src/renderer/table.d.ts +3 -0
  192. package/dist/src/renderer/table.d.ts.map +1 -0
  193. package/dist/src/renderer/table.js +134 -0
  194. package/dist/src/renderer/table.js.map +1 -0
  195. package/dist/src/renderer/trends.d.ts +6 -0
  196. package/dist/src/renderer/trends.d.ts.map +1 -0
  197. package/dist/src/renderer/trends.js +127 -0
  198. package/dist/src/renderer/trends.js.map +1 -0
  199. package/dist/src/server/job-store.d.ts +4 -0
  200. package/dist/src/server/job-store.d.ts.map +1 -0
  201. package/dist/src/server/job-store.js +68 -0
  202. package/dist/src/server/job-store.js.map +1 -0
  203. package/dist/src/server/report-server.d.ts +16 -0
  204. package/dist/src/server/report-server.d.ts.map +1 -0
  205. package/dist/src/server/report-server.js +196 -0
  206. package/dist/src/server/report-server.js.map +1 -0
  207. package/dist/src/server/report-store.d.ts +44 -0
  208. package/dist/src/server/report-store.d.ts.map +1 -0
  209. package/dist/src/server/report-store.js +190 -0
  210. package/dist/src/server/report-store.js.map +1 -0
  211. package/dist/src/types.d.ts +553 -0
  212. package/dist/src/types.d.ts.map +1 -0
  213. package/dist/src/types.js +2 -0
  214. package/dist/src/types.js.map +1 -0
  215. package/package.json +44 -3
@@ -0,0 +1,210 @@
1
+ /**
2
+ * Statistical functions for evaluation analysis.
3
+ *
4
+ * Assumptions:
5
+ * - All inputs are independent observations from normally distributed populations
6
+ * - Welch's t-test is used (does NOT assume equal variances)
7
+ * - Confidence intervals use the t-distribution for small samples (df ≤ 30)
8
+ * and normal approximation (z = 1.96) for larger samples
9
+ *
10
+ * Dependencies: simple-statistics (for mean, sampleStandardDeviation, sampleVariance)
11
+ */
12
+ import * as ss from 'simple-statistics';
13
+ /**
14
+ * Arithmetic mean. Returns 0 for empty/null input.
15
+ */
16
+ export function mean(arr) {
17
+ if (!arr || arr.length === 0)
18
+ return 0;
19
+ return ss.mean(arr);
20
+ }
21
+ /**
22
+ * Sample standard deviation (Bessel's correction: n-1 denominator).
23
+ * Returns 0 for arrays with fewer than 2 elements.
24
+ */
25
+ export function stddev(arr) {
26
+ if (!arr || arr.length < 2)
27
+ return 0;
28
+ return ss.sampleStandardDeviation(arr);
29
+ }
30
+ // t-distribution critical values for two-tailed tests at p = 0.05, 0.01, 0.001.
31
+ // Source: standard statistical tables. Used both for the significance gate and
32
+ // for bucketed p-value reporting (see pValueCategory below).
33
+ const T_CRITICAL_95 = {
34
+ 1: 12.706, 2: 4.303, 3: 3.182, 4: 2.776, 5: 2.571,
35
+ 6: 2.447, 7: 2.365, 8: 2.306, 9: 2.262, 10: 2.228,
36
+ 11: 2.201, 12: 2.179, 13: 2.160, 14: 2.145, 15: 2.131,
37
+ 16: 2.120, 17: 2.110, 18: 2.101, 19: 2.093, 20: 2.086,
38
+ 21: 2.080, 22: 2.074, 23: 2.069, 24: 2.064, 25: 2.060,
39
+ 26: 2.056, 27: 2.052, 28: 2.048, 29: 2.045, 30: 2.042,
40
+ };
41
+ const T_CRITICAL_99 = {
42
+ 1: 63.657, 2: 9.925, 3: 5.841, 4: 4.604, 5: 4.032,
43
+ 6: 3.707, 7: 3.499, 8: 3.355, 9: 3.250, 10: 3.169,
44
+ 11: 3.106, 12: 3.055, 13: 3.012, 14: 2.977, 15: 2.947,
45
+ 16: 2.921, 17: 2.898, 18: 2.878, 19: 2.861, 20: 2.845,
46
+ 21: 2.831, 22: 2.819, 23: 2.807, 24: 2.797, 25: 2.787,
47
+ 26: 2.779, 27: 2.771, 28: 2.763, 29: 2.756, 30: 2.750,
48
+ };
49
+ const T_CRITICAL_999 = {
50
+ 1: 636.619, 2: 31.599, 3: 12.924, 4: 8.610, 5: 6.869,
51
+ 6: 5.959, 7: 5.408, 8: 5.041, 9: 4.781, 10: 4.587,
52
+ 11: 4.437, 12: 4.318, 13: 4.221, 14: 4.140, 15: 4.073,
53
+ 16: 4.015, 17: 3.965, 18: 3.922, 19: 3.883, 20: 3.850,
54
+ 21: 3.819, 22: 3.792, 23: 3.768, 24: 3.745, 25: 3.725,
55
+ 26: 3.707, 27: 3.690, 28: 3.674, 29: 3.659, 30: 3.646,
56
+ };
57
+ /**
58
+ * Bucket a t-statistic into a p-value category by comparing |t| to standard
59
+ * critical-value tables at α = 0.001 / 0.01 / 0.05 (two-tailed).
60
+ *
61
+ * Returns one of four buckets instead of a continuous p value. This avoids
62
+ * implementing the regularized incomplete beta function for an exact CDF
63
+ * while still giving readers enough resolution to distinguish
64
+ * "barely significant" from "strongly significant".
65
+ *
66
+ * For df > 30 falls back to the normal-approximation critical values.
67
+ */
68
+ export function pValueCategory(tStatistic, df) {
69
+ if (!Number.isFinite(tStatistic) || df <= 0)
70
+ return '≥0.05';
71
+ const absT = Math.abs(tStatistic);
72
+ const dfKey = Math.min(Math.max(1, Math.floor(df)), 30);
73
+ const c999 = df > 30 ? 3.291 : (T_CRITICAL_999[dfKey] ?? 3.291);
74
+ const c99 = df > 30 ? 2.576 : (T_CRITICAL_99[dfKey] ?? 2.576);
75
+ const c95 = df > 30 ? 1.96 : (T_CRITICAL_95[dfKey] ?? 1.96);
76
+ if (absT >= c999)
77
+ return '<0.001';
78
+ if (absT >= c99)
79
+ return '<0.01';
80
+ if (absT >= c95)
81
+ return '<0.05';
82
+ return '≥0.05';
83
+ }
84
+ /**
85
+ * Get t-distribution critical value for given degrees of freedom (95% CI, two-tailed).
86
+ * Uses lookup table for df ≤ 30, normal approximation (1.96) for df > 30.
87
+ */
88
+ function tCritical(df) {
89
+ if (df <= 0)
90
+ return Infinity;
91
+ if (df <= 30)
92
+ return T_CRITICAL_95[df] || T_CRITICAL_95[30];
93
+ return 1.96;
94
+ }
95
+ /**
96
+ * Compute 95% confidence interval for the population mean.
97
+ *
98
+ * Method: t-interval (appropriate for small samples from normal populations)
99
+ * Formula: mean ± t(α/2, n-1) × (s / √n)
100
+ */
101
+ export function confidenceInterval(arr) {
102
+ const m = mean(arr);
103
+ const sd = stddev(arr);
104
+ if (!arr || arr.length < 2) {
105
+ return { mean: m, lower: m, upper: m, stddev: sd };
106
+ }
107
+ const n = arr.length;
108
+ const se = sd / Math.sqrt(n); // Standard error of the mean
109
+ const df = n - 1; // Degrees of freedom
110
+ const tc = tCritical(df); // Critical value
111
+ const margin = tc * se; // Margin of error
112
+ return {
113
+ mean: Number(m.toFixed(4)),
114
+ lower: Number((m - margin).toFixed(4)),
115
+ upper: Number((m + margin).toFixed(4)),
116
+ stddev: Number(sd.toFixed(4)),
117
+ };
118
+ }
119
+ /**
120
+ * Effect size for two independent samples.
121
+ *
122
+ * Cohen's d: (mean_a - mean_b) / pooled_stddev
123
+ * Hedges' g: J * d, where J = 1 - 3 / (4 * (n1 + n2) - 9)
124
+ *
125
+ * Hedges' g corrects Cohen's d's small-sample bias (Hedges 1981) and is
126
+ * preferred when n1 + n2 < 20. For n1 + n2 >= 20 the correction is < 5%
127
+ * and Cohen's d is the conventional choice.
128
+ *
129
+ * `primary` indicates which to emphasize in UI based on total sample size.
130
+ * `magnitude` uses standard thresholds (|effect| 0.2 / 0.5 / 0.8) applied
131
+ * to the primary metric.
132
+ *
133
+ * Returns effect 0 with primary 'none' for insufficient data or zero variance.
134
+ */
135
+ export function effectSize(a, b) {
136
+ const n1 = a?.length ?? 0;
137
+ const n2 = b?.length ?? 0;
138
+ if (n1 < 2 || n2 < 2) {
139
+ return { cohensD: 0, hedgesG: 0, primary: 'none', magnitude: 'none', pooledStddev: 0, n1, n2 };
140
+ }
141
+ const mA = ss.mean(a);
142
+ const mB = ss.mean(b);
143
+ const varA = ss.sampleVariance(a);
144
+ const varB = ss.sampleVariance(b);
145
+ // Pooled standard deviation (assumes similar variances; acceptable for
146
+ // effect-size reporting even when Welch's t-test is used for significance)
147
+ const pooledVar = ((n1 - 1) * varA + (n2 - 1) * varB) / (n1 + n2 - 2);
148
+ const pooledStddev = Math.sqrt(pooledVar);
149
+ if (pooledStddev === 0) {
150
+ return { cohensD: 0, hedgesG: 0, primary: 'none', magnitude: 'none', pooledStddev: 0, n1, n2 };
151
+ }
152
+ const d = (mA - mB) / pooledStddev;
153
+ // Hedges' correction factor J. Denominator guard: 4*(n1+n2)-9 > 0 iff n1+n2 > 2.25,
154
+ // which is already satisfied by the n1>=2, n2>=2 check above.
155
+ const J = 1 - 3 / (4 * (n1 + n2) - 9);
156
+ const g = J * d;
157
+ const totalN = n1 + n2;
158
+ const primary = totalN < 20 ? 'g' : 'd';
159
+ const ref = primary === 'g' ? g : d;
160
+ const abs = Math.abs(ref);
161
+ const magnitude = abs < 0.2 ? 'negligible' :
162
+ abs < 0.5 ? 'small' :
163
+ abs < 0.8 ? 'medium' : 'large';
164
+ return {
165
+ cohensD: Number(d.toFixed(4)),
166
+ hedgesG: Number(g.toFixed(4)),
167
+ primary,
168
+ magnitude,
169
+ pooledStddev: Number(pooledStddev.toFixed(4)),
170
+ n1,
171
+ n2,
172
+ };
173
+ }
174
+ export function tTest(a, b) {
175
+ // Minimum sample size check
176
+ if (!a || !b || a.length < 2 || b.length < 2) {
177
+ return { tStatistic: 0, df: 0, significant: false };
178
+ }
179
+ const mA = ss.mean(a);
180
+ const mB = ss.mean(b);
181
+ const varA = ss.sampleVariance(a);
182
+ const varB = ss.sampleVariance(b);
183
+ // Estimated variances of sample means
184
+ const seA = varA / a.length;
185
+ const seB = varB / b.length;
186
+ const seDiff = Math.sqrt(seA + seB);
187
+ // Both groups have zero variance — means are exact, can't compute t
188
+ if (seDiff === 0) {
189
+ return {
190
+ tStatistic: 0,
191
+ df: a.length + b.length - 2,
192
+ significant: mA !== mB, // Different constants are trivially significant
193
+ };
194
+ }
195
+ // t-statistic
196
+ const tStat = (mA - mB) / seDiff;
197
+ // Welch-Satterthwaite degrees of freedom
198
+ const dfNum = (seA + seB) ** 2;
199
+ const dfDen = (seA ** 2 / (a.length - 1)) + (seB ** 2 / (b.length - 1));
200
+ const df = Math.max(1, Math.floor(dfNum / dfDen));
201
+ // Significance at α = 0.05 (two-tailed)
202
+ const tc = tCritical(df);
203
+ const significant = Math.abs(tStat) > tc;
204
+ return {
205
+ tStatistic: Number(tStat.toFixed(4)),
206
+ df,
207
+ significant,
208
+ };
209
+ }
210
+ //# sourceMappingURL=statistics.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"statistics.js","sourceRoot":"","sources":["../../../src/eval-core/statistics.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,KAAK,EAAE,MAAM,mBAAmB,CAAC;AAExC;;GAEG;AACH,MAAM,UAAU,IAAI,CAAC,GAAa;IAChC,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACvC,OAAO,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;AACtB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,MAAM,CAAC,GAAa;IAClC,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,CAAC,CAAC;IACrC,OAAO,EAAE,CAAC,uBAAuB,CAAC,GAAG,CAAC,CAAC;AACzC,CAAC;AAED,gFAAgF;AAChF,+EAA+E;AAC/E,6DAA6D;AAC7D,MAAM,aAAa,GAA2B;IAC5C,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK;IACjD,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACjD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACrD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACrD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACrD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;CACtD,CAAC;AAEF,MAAM,aAAa,GAA2B;IAC5C,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK;IACjD,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACjD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACrD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACrD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACrD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;CACtD,CAAC;AAEF,MAAM,cAAc,GAA2B;IAC7C,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK;IACpD,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACjD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACrD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACrD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;IACrD,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK;CACtD,CAAC;AAIF;;;;;;;;;;GAUG;AACH,MAAM,UAAU,cAAc,CAAC,UAAkB,EAAE,EAAU;IAC3D,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,IAAI,EAAE,IAAI,CAAC;QAAE,OAAO,OAAO,CAAC;IAC5D,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;IAClC,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACxD,MAAM,IAAI,GAAG,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,CAAC;IAChE,MAAM,GAAG,GAAG,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,CAAC;IAC9D,MAAM,GAAG,GAAG,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,CAAC;IAC5D,IAAI,IAAI,IAAI,IAAI;QAAE,OAAO,QAAQ,CAAC;IAClC,IAAI,IAAI,IAAI,GAAG;QAAE,OAAO,OAAO,CAAC;IAChC,IAAI,IAAI,IAAI,GAAG;QAAE,OAAO,OAAO,CAAC;IAChC,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;;GAGG;AACH,SAAS,SAAS,CAAC,EAAU;IAC3B,IAAI,EAAE,IAAI,CAAC;QAAE,OAAO,QAAQ,CAAC;IAC7B,IAAI,EAAE,IAAI,EAAE;QAAE,OAAO,aAAa,CAAC,EAAE,CAAC,IAAI,aAAa,CAAC,EAAE,CAAC,CAAC;IAC5D,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,kBAAkB,CAAC,GAAa;IAC9C,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC;IACpB,MAAM,EAAE,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;IAEvB,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC3B,OAAO,EAAE,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;IACrD,CAAC;IAED,MAAM,CAAC,GAAG,GAAG,CAAC,MAAM,CAAC;IACrB,MAAM,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAQ,6BAA6B;IAClE,MAAM,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAqB,qBAAqB;IAC3D,MAAM,EAAE,GAAG,SAAS,CAAC,EAAE,CAAC,CAAC,CAAa,iBAAiB;IACvD,MAAM,MAAM,GAAG,EAAE,GAAG,EAAE,CAAC,CAAe,kBAAkB;IAExD,OAAO;QACL,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC1B,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACtC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACtC,MAAM,EAAE,MAAM,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;KAC9B,CAAC;AACJ,CAAC;AAqBD;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,UAAU,CAAC,CAAW,EAAE,CAAW;IACjD,MAAM,EAAE,GAAG,CAAC,EAAE,MAAM,IAAI,CAAC,CAAC;IAC1B,MAAM,EAAE,GAAG,CAAC,EAAE,MAAM,IAAI,CAAC,CAAC;IAE1B,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,GAAG,CAAC,EAAE,CAAC;QACrB,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,YAAY,EAAE,CAAC,EAAE,EAAE,EAAE,EAAE,EAAE,CAAC;IACjG,CAAC;IAED,MAAM,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACtB,MAAM,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACtB,MAAM,IAAI,GAAG,EAAE,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC;IAClC,MAAM,IAAI,GAAG,EAAE,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC;IAElC,uEAAuE;IACvE,2EAA2E;IAC3E,MAAM,SAAS,GAAG,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,GAAG,IAAI,GAAG,CAAC,EAAE,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;IACtE,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IAE1C,IAAI,YAAY,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,YAAY,EAAE,CAAC,EAAE,EAAE,EAAE,EAAE,EAAE,CAAC;IACjG,CAAC;IAED,MAAM,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,GAAG,YAAY,CAAC;IAEnC,oFAAoF;IACpF,8DAA8D;IAC9D,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACtC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAEhB,MAAM,MAAM,GAAG,EAAE,GAAG,EAAE,CAAC;IACvB,MAAM,OAAO,GAAc,MAAM,GAAG,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;IACnD,MAAM,GAAG,GAAG,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACpC,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAC1B,MAAM,SAAS,GACb,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC;QAC1B,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;YACrB,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC;IAEjC,OAAO;QACL,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC7B,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC7B,OAAO;QACP,SAAS;QACT,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC7C,EAAE;QACF,EAAE;KACH,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,KAAK,CAAC,CAAW,EAAE,CAAW;IAC5C,4BAA4B;IAC5B,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC7C,OAAO,EAAE,UAAU,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,WAAW,EAAE,KAAK,EAAE,CAAC;IACtD,CAAC;IAED,MAAM,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACtB,MAAM,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACtB,MAAM,IAAI,GAAG,EAAE,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC;IAClC,MAAM,IAAI,GAAG,EAAE,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC;IAElC,sCAAsC;IACtC,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5B,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5B,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,GAAG,GAAG,CAAC,CAAC;IAEpC,oEAAoE;IACpE,IAAI,MAAM,KAAK,CAAC,EAAE,CAAC;QACjB,OAAO;YACL,UAAU,EAAE,CAAC;YACb,EAAE,EAAE,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,MAAM,GAAG,CAAC;YAC3B,WAAW,EAAE,EAAE,KAAK,EAAE,EAAE,gDAAgD;SACzE,CAAC;IACJ,CAAC;IAED,cAAc;IACd,MAAM,KAAK,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,GAAG,MAAM,CAAC;IAEjC,yCAAyC;IACzC,MAAM,KAAK,GAAG,CAAC,GAAG,GAAG,GAAG,CAAC,IAAI,CAAC,CAAC;IAC/B,MAAM,KAAK,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC;IACxE,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC;IAElD,wCAAwC;IACxC,MAAM,EAAE,GAAG,SAAS,CAAC,EAAE,CAAC,CAAC;IACzB,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC;IAEzC,OAAO;QACL,UAAU,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACpC,EAAE;QACF,WAAW;KACZ,CAAC;AACJ,CAAC"}
@@ -0,0 +1,4 @@
1
+ import type { Artifact, Sample, Task } from '../types.js';
2
+ export declare function buildTasks(samples: Sample[], variants: string[], skills: Record<string, string | null>): Task[];
3
+ export declare function buildTasksFromArtifacts(samples: Sample[], artifacts: Artifact[]): Task[];
4
+ //# sourceMappingURL=task-planner.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"task-planner.d.ts","sourceRoot":"","sources":["../../../src/eval-core/task-planner.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,aAAa,CAAC;AAE1D,wBAAgB,UAAU,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC,GAAG,IAAI,EAAE,CAQ/G;AAED,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,QAAQ,EAAE,GAAG,IAAI,EAAE,CAyBxF"}
@@ -0,0 +1,33 @@
1
+ export function buildTasks(samples, variants, skills) {
2
+ const artifacts = variants.map((variant) => ({
3
+ name: variant,
4
+ kind: variant === 'baseline' ? 'baseline' : 'skill',
5
+ source: variant === 'baseline' ? 'baseline' : 'custom',
6
+ content: skills[variant] || null,
7
+ }));
8
+ return buildTasksFromArtifacts(samples, artifacts);
9
+ }
10
+ export function buildTasksFromArtifacts(samples, artifacts) {
11
+ const tasks = [];
12
+ for (const sample of samples) {
13
+ for (const artifact of artifacts) {
14
+ const userPrompt = sample.context
15
+ ? `${sample.prompt}\n\n\`\`\`\n${sample.context}\n\`\`\``
16
+ : sample.prompt;
17
+ tasks.push({
18
+ sample_id: sample.sample_id,
19
+ variant: artifact.name,
20
+ artifact,
21
+ prompt: userPrompt,
22
+ rubric: sample.rubric || null,
23
+ assertions: sample.assertions || null,
24
+ dimensions: sample.dimensions || null,
25
+ artifactContent: artifact.content,
26
+ cwd: artifact.cwd || sample.cwd || null,
27
+ _sample: sample,
28
+ });
29
+ }
30
+ }
31
+ return tasks;
32
+ }
33
+ //# sourceMappingURL=task-planner.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"task-planner.js","sourceRoot":"","sources":["../../../src/eval-core/task-planner.ts"],"names":[],"mappings":"AAEA,MAAM,UAAU,UAAU,CAAC,OAAiB,EAAE,QAAkB,EAAE,MAAqC;IACrG,MAAM,SAAS,GAAe,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACvD,IAAI,EAAE,OAAO;QACb,IAAI,EAAE,OAAO,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,OAAO;QACnD,MAAM,EAAE,OAAO,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ;QACtD,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,IAAI,IAAI;KACjC,CAAC,CAAC,CAAC;IACJ,OAAO,uBAAuB,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;AACrD,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,OAAiB,EAAE,SAAqB;IAC9E,MAAM,KAAK,GAAW,EAAE,CAAC;IAEzB,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YACjC,MAAM,UAAU,GAAG,MAAM,CAAC,OAAO;gBAC/B,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,eAAe,MAAM,CAAC,OAAO,UAAU;gBACzD,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC;YAElB,KAAK,CAAC,IAAI,CAAC;gBACT,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,OAAO,EAAE,QAAQ,CAAC,IAAI;gBACtB,QAAQ;gBACR,MAAM,EAAE,UAAU;gBAClB,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;gBAC7B,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;gBACrC,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;gBACrC,eAAe,EAAE,QAAQ,CAAC,OAAO;gBACjC,GAAG,EAAE,QAAQ,CAAC,GAAG,IAAI,MAAM,CAAC,GAAG,IAAI,IAAI;gBACvC,OAAO,EAAE,MAAM;aAChB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC"}
@@ -0,0 +1,133 @@
1
+ import type { Artifact, JobStore, ProgressCallback, Report, VariantResult, VariantSummary } from '../types.js';
2
+ interface RunSingleEvaluationOptions {
3
+ samplesPath: string;
4
+ skillDir: string;
5
+ artifacts: Artifact[];
6
+ model: string;
7
+ judgeModel: string;
8
+ outputDir: null;
9
+ noJudge: boolean;
10
+ concurrency: number;
11
+ timeoutMs?: number;
12
+ executorName: string;
13
+ judgeExecutorName?: string;
14
+ jobStore: null;
15
+ persistJob: false;
16
+ onProgress: ProgressCallback | null;
17
+ skipPreflight: boolean;
18
+ mcpConfig?: string;
19
+ verbose: boolean;
20
+ }
21
+ export interface EachSkillResult {
22
+ name: string;
23
+ artifactHash: string;
24
+ samplesPath: string;
25
+ sampleCount: number;
26
+ summary: Record<string, VariantSummary>;
27
+ results: Array<{
28
+ sample_id: string;
29
+ variants: {
30
+ baseline: VariantResult;
31
+ skill: VariantResult;
32
+ };
33
+ }>;
34
+ }
35
+ export declare function buildEachReport({ skillDir, skillEntries, skillResults, model, judgeModel, noJudge, executorName, judgeExecutorName, project, owner, tags, dryRun, concurrency, timeoutMs, totalCostUSD, }: {
36
+ skillDir: string;
37
+ skillEntries: Array<{
38
+ name: string;
39
+ skillPath: string;
40
+ samplesPath: string;
41
+ }>;
42
+ skillResults: EachSkillResult[];
43
+ model: string;
44
+ judgeModel: string;
45
+ noJudge: boolean;
46
+ executorName: string;
47
+ judgeExecutorName?: string;
48
+ project?: string;
49
+ owner?: string;
50
+ tags?: string[];
51
+ dryRun: boolean;
52
+ concurrency: number;
53
+ timeoutMs?: number;
54
+ totalCostUSD: number;
55
+ }): {
56
+ report: {
57
+ id: string;
58
+ each: true;
59
+ meta: {
60
+ variants: string[];
61
+ model: string;
62
+ judgeModel: string | null;
63
+ executor: string;
64
+ sampleCount: number;
65
+ taskCount: number;
66
+ totalCostUSD: number;
67
+ timestamp: string;
68
+ cliVersion: string;
69
+ nodeVersion: string;
70
+ artifactHashes: {
71
+ [k: string]: string;
72
+ };
73
+ request: import("../types.js").EvaluationRequest;
74
+ run: import("../types.js").EvaluationRun;
75
+ job: import("../types.js").EvaluationJob;
76
+ };
77
+ summary: {};
78
+ results: never[];
79
+ overview: {
80
+ totalArtifacts: number;
81
+ totalSamples: number;
82
+ totalCostUSD: number;
83
+ artifacts: {
84
+ name: string;
85
+ baselineScore: number | null;
86
+ artifactScore: number | null;
87
+ improvement: string;
88
+ }[];
89
+ };
90
+ artifacts: EachSkillResult[];
91
+ };
92
+ job: import("../types.js").EvaluationJob;
93
+ };
94
+ export declare function executeEachEvaluationRuns({ skillDir, skillEntries, model, judgeModel, outputDir, project, owner, tags, noJudge, concurrency, timeoutMs, executorName, judgeExecutorName, jobStore, persistJob, onProgress, onSkillProgress, skipPreflight, mcpConfig, verbose, runSingleEvaluation, }: {
95
+ skillDir: string;
96
+ skillEntries: Array<{
97
+ name: string;
98
+ skillPath: string;
99
+ samplesPath: string;
100
+ }>;
101
+ model: string;
102
+ judgeModel: string;
103
+ outputDir?: string | null;
104
+ project?: string;
105
+ owner?: string;
106
+ tags?: string[];
107
+ noJudge?: boolean;
108
+ concurrency?: number;
109
+ timeoutMs?: number;
110
+ executorName: string;
111
+ judgeExecutorName?: string;
112
+ jobStore?: JobStore | null;
113
+ persistJob?: boolean;
114
+ onProgress?: ProgressCallback | null;
115
+ onSkillProgress?: ((info: {
116
+ phase: string;
117
+ skill: string;
118
+ current: number;
119
+ total: number;
120
+ }) => void) | null;
121
+ skipPreflight?: boolean;
122
+ mcpConfig?: string;
123
+ verbose?: boolean;
124
+ runSingleEvaluation: (options: RunSingleEvaluationOptions) => Promise<{
125
+ report: Report;
126
+ filePath: string | null;
127
+ }>;
128
+ }): Promise<{
129
+ report: Report;
130
+ filePath: string | null;
131
+ }>;
132
+ export {};
133
+ //# sourceMappingURL=each-evaluation-workflow.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"each-evaluation-workflow.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/each-evaluation-workflow.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,gBAAgB,EAAE,MAAM,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAE/G,UAAU,0BAA0B;IAClC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,IAAI,CAAC;IAChB,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,EAAE,IAAI,CAAC;IACf,UAAU,EAAE,KAAK,CAAC;IAClB,UAAU,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACpC,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;CAClB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACxC,OAAO,EAAE,KAAK,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;QAClB,QAAQ,EAAE;YACR,QAAQ,EAAE,aAAa,CAAC;YACxB,KAAK,EAAE,aAAa,CAAC;SACtB,CAAC;KACH,CAAC,CAAC;CACJ;AAsBD,wBAAgB,eAAe,CAAC,EAC9B,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,WAAW,EACX,SAAS,EACT,YAAY,GACb,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,YAAY,EAAE,eAAe,EAAE,CAAC;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,EAAE,OAAO,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;CACtB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAsEA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,WAAe,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,mBAAmB,GACpB,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;IAC5G,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,mBAAmB,EAAE,CAAC,OAAO,EAAE,0BAA0B,KAAK,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,CAAC,CAAC;CACpH,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CA+EvD"}
@@ -0,0 +1,169 @@
1
+ import { resolve } from 'node:path';
2
+ import { DEFAULT_OUTPUT_DIR, generateRunId, persistReport } from '../eval-core/evaluation-reporting.js';
3
+ import { buildEvaluationRequest, createEvaluationRun, createSucceededJob, finalizeEvaluationRun } from '../eval-core/evaluation-job.js';
4
+ import { createFileJobStore, DEFAULT_JOBS_DIR } from '../server/job-store.js';
5
+ import { resolveArtifacts } from '../inputs/skill-loader.js';
6
+ function buildEachOverview(skillResults, totalCostUSD) {
7
+ return {
8
+ totalArtifacts: skillResults.length,
9
+ totalSamples: skillResults.reduce((sum, skill) => sum + skill.sampleCount, 0),
10
+ totalCostUSD: Number(totalCostUSD.toFixed(6)),
11
+ artifacts: skillResults.map((skill) => {
12
+ const baselineSummary = skill.summary.baseline;
13
+ const artifactSummary = skill.summary.skill;
14
+ const baselineScore = baselineSummary?.avgCompositeScore ?? baselineSummary?.avgLlmScore ?? null;
15
+ const artifactScore = artifactSummary?.avgCompositeScore ?? artifactSummary?.avgLlmScore ?? null;
16
+ if (typeof baselineScore !== 'number' || typeof artifactScore !== 'number' || baselineScore <= 0) {
17
+ return { name: skill.name, baselineScore, artifactScore, improvement: '-' };
18
+ }
19
+ const delta = ((artifactScore - baselineScore) / baselineScore * 100).toFixed(0);
20
+ const improvement = artifactScore >= baselineScore ? `+${delta}%` : `${delta}%`;
21
+ return { name: skill.name, baselineScore, artifactScore, improvement };
22
+ }),
23
+ };
24
+ }
25
+ export function buildEachReport({ skillDir, skillEntries, skillResults, model, judgeModel, noJudge, executorName, judgeExecutorName, project, owner, tags, dryRun, concurrency, timeoutMs, totalCostUSD, }) {
26
+ const runId = generateRunId(['each']);
27
+ const request = buildEvaluationRequest({
28
+ samplesPath: '',
29
+ skillDir,
30
+ artifacts: skillEntries.map((entry) => ({
31
+ name: entry.name,
32
+ kind: 'skill',
33
+ source: 'file-path',
34
+ content: null,
35
+ locator: entry.skillPath,
36
+ })),
37
+ model,
38
+ judgeModel: noJudge ? null : judgeModel,
39
+ executor: executorName,
40
+ judgeExecutor: judgeExecutorName || executorName,
41
+ noJudge,
42
+ concurrency,
43
+ timeoutMs,
44
+ noCache: false,
45
+ dryRun,
46
+ blind: false,
47
+ project,
48
+ owner,
49
+ tags,
50
+ });
51
+ const createdAt = new Date().toISOString();
52
+ const { run: initialRun, startedAt } = createEvaluationRun(runId, createdAt);
53
+ const finishedAt = new Date().toISOString();
54
+ const run = finalizeEvaluationRun(initialRun, finishedAt);
55
+ const job = createSucceededJob({
56
+ jobId: `job-${runId}`,
57
+ runId,
58
+ reportId: runId,
59
+ request,
60
+ createdAt,
61
+ startedAt,
62
+ finishedAt,
63
+ });
64
+ const totalSampleCount = skillResults.reduce((sum, skill) => sum + skill.sampleCount, 0);
65
+ return {
66
+ report: {
67
+ id: runId,
68
+ each: true,
69
+ meta: {
70
+ variants: skillEntries.map((entry) => entry.name),
71
+ model,
72
+ judgeModel: noJudge ? null : judgeModel,
73
+ executor: executorName,
74
+ sampleCount: totalSampleCount,
75
+ taskCount: totalSampleCount * 2,
76
+ totalCostUSD: Number(totalCostUSD.toFixed(6)),
77
+ timestamp: new Date().toISOString(),
78
+ cliVersion: '',
79
+ nodeVersion: process.version,
80
+ artifactHashes: Object.fromEntries(skillResults.map((skill) => [skill.name, skill.artifactHash || 'no-skill'])),
81
+ request,
82
+ run,
83
+ job,
84
+ },
85
+ summary: {},
86
+ results: [],
87
+ overview: buildEachOverview(skillResults, totalCostUSD),
88
+ artifacts: skillResults,
89
+ },
90
+ job,
91
+ };
92
+ }
93
+ export async function executeEachEvaluationRuns({ skillDir, skillEntries, model, judgeModel, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, noJudge = false, concurrency = 1, timeoutMs, executorName, judgeExecutorName, jobStore = null, persistJob = true, onProgress = null, onSkillProgress = null, skipPreflight = false, mcpConfig, verbose = false, runSingleEvaluation, }) {
94
+ const skillResults = [];
95
+ let totalCostUSD = 0;
96
+ for (let i = 0; i < skillEntries.length; i++) {
97
+ const entry = skillEntries[i];
98
+ onSkillProgress?.({ phase: 'start', skill: entry.name, current: i + 1, total: skillEntries.length });
99
+ // each mode 的实验结构固定为"baseline (control) vs skill (treatment)"。
100
+ // 显式在 artifact 上填 experimentRole,下游 buildVariantConfig 直接读取。
101
+ const skillArtifacts = resolveArtifacts(resolve(skillDir), ['baseline', entry.skillPath]).map((artifact) => {
102
+ if (artifact.name === entry.skillPath) {
103
+ return { ...artifact, name: 'skill', experimentRole: 'treatment' };
104
+ }
105
+ return { ...artifact, experimentRole: 'control' };
106
+ });
107
+ const { report } = await runSingleEvaluation({
108
+ samplesPath: entry.samplesPath,
109
+ skillDir,
110
+ artifacts: skillArtifacts,
111
+ model,
112
+ judgeModel,
113
+ outputDir: null,
114
+ noJudge,
115
+ concurrency,
116
+ timeoutMs,
117
+ executorName,
118
+ judgeExecutorName,
119
+ jobStore: null,
120
+ persistJob: false,
121
+ onProgress,
122
+ skipPreflight: skipPreflight || i > 0,
123
+ mcpConfig,
124
+ verbose,
125
+ });
126
+ skillResults.push({
127
+ name: entry.name,
128
+ artifactHash: report.meta.artifactHashes?.skill || '',
129
+ samplesPath: entry.samplesPath,
130
+ sampleCount: report.meta.sampleCount,
131
+ summary: {
132
+ baseline: report.summary.baseline || {},
133
+ skill: (report.summary.skill || {}),
134
+ },
135
+ results: report.results.map((result) => ({
136
+ sample_id: result.sample_id,
137
+ variants: {
138
+ baseline: result.variants.baseline || result.variants['baseline'],
139
+ skill: result.variants.skill,
140
+ },
141
+ })),
142
+ });
143
+ totalCostUSD += report.meta.totalCostUSD;
144
+ onSkillProgress?.({ phase: 'done', skill: entry.name, current: i + 1, total: skillEntries.length });
145
+ }
146
+ const { report: combinedReport, job } = buildEachReport({
147
+ skillDir,
148
+ skillEntries,
149
+ skillResults,
150
+ model,
151
+ judgeModel,
152
+ noJudge,
153
+ executorName,
154
+ judgeExecutorName,
155
+ project,
156
+ owner,
157
+ tags,
158
+ dryRun: false,
159
+ concurrency,
160
+ timeoutMs,
161
+ totalCostUSD,
162
+ });
163
+ const filePath = persistReport(combinedReport, outputDir);
164
+ const resolvedJobStore = persistJob ? (jobStore ?? createFileJobStore(DEFAULT_JOBS_DIR)) : null;
165
+ if (resolvedJobStore)
166
+ await resolvedJobStore.save(job.jobId, job);
167
+ return { report: combinedReport, filePath };
168
+ }
169
+ //# sourceMappingURL=each-evaluation-workflow.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"each-evaluation-workflow.js","sourceRoot":"","sources":["../../../src/eval-workflows/each-evaluation-workflow.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACxG,OAAO,EAAE,sBAAsB,EAAE,mBAAmB,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,MAAM,gCAAgC,CAAC;AACxI,OAAO,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC9E,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAsC7D,SAAS,iBAAiB,CAAC,YAA+B,EAAE,YAAoB;IAC9E,OAAO;QACL,cAAc,EAAE,YAAY,CAAC,MAAM;QACnC,YAAY,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC;QAC7E,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC7C,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;YACpC,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC;YAC/C,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;YAC5C,MAAM,aAAa,GAAG,eAAe,EAAE,iBAAiB,IAAI,eAAe,EAAE,WAAW,IAAI,IAAI,CAAC;YACjG,MAAM,aAAa,GAAG,eAAe,EAAE,iBAAiB,IAAI,eAAe,EAAE,WAAW,IAAI,IAAI,CAAC;YACjG,IAAI,OAAO,aAAa,KAAK,QAAQ,IAAI,OAAO,aAAa,KAAK,QAAQ,IAAI,aAAa,IAAI,CAAC,EAAE,CAAC;gBACjG,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,aAAa,EAAE,aAAa,EAAE,WAAW,EAAE,GAAG,EAAE,CAAC;YAC9E,CAAC;YACD,MAAM,KAAK,GAAG,CAAC,CAAC,aAAa,GAAG,aAAa,CAAC,GAAG,aAAa,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YACjF,MAAM,WAAW,GAAG,aAAa,IAAI,aAAa,CAAC,CAAC,CAAC,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC,GAAG,KAAK,GAAG,CAAC;YAChF,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,aAAa,EAAE,aAAa,EAAE,WAAW,EAAE,CAAC;QACzE,CAAC,CAAC;KACH,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,EAC9B,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,WAAW,EACX,SAAS,EACT,YAAY,GAiBb;IACC,MAAM,KAAK,GAAG,aAAa,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;IACtC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW,EAAE,EAAE;QACf,QAAQ;QACR,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;YACtC,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,IAAI,EAAE,OAAO;YACb,MAAM,EAAE,WAAW;YACnB,OAAO,EAAE,IAAI;YACb,OAAO,EAAE,KAAK,CAAC,SAAS;SACzB,CAAC,CAAC;QACH,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB,IAAI,YAAY;QAChD,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO,EAAE,KAAK;QACd,MAAM;QACN,KAAK,EAAE,KAAK;QACZ,OAAO;QACP,KAAK;QACL,IAAI;KACL,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC7E,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAC1D,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,OAAO,KAAK,EAAE;QACrB,KAAK;QACL,QAAQ,EAAE,KAAK;QACf,OAAO;QACP,SAAS;QACT,SAAS;QACT,UAAU;KACX,CAAC,CAAC;IACH,MAAM,gBAAgB,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;IAEzF,OAAO;QACL,MAAM,EAAE;YACN,EAAE,EAAE,KAAK;YACT,IAAI,EAAE,IAAI;YACV,IAAI,EAAE;gBACJ,QAAQ,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC;gBACjD,KAAK;gBACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;gBACvC,QAAQ,EAAE,YAAY;gBACtB,WAAW,EAAE,gBAAgB;gBAC7B,SAAS,EAAE,gBAAgB,GAAG,CAAC;gBAC/B,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;gBAC7C,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;gBACnC,UAAU,EAAE,EAAE;gBACd,WAAW,EAAE,OAAO,CAAC,OAAO;gBAC5B,cAAc,EAAE,MAAM,CAAC,WAAW,CAChC,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,YAAY,IAAI,UAAU,CAAC,CAAC,CAC5E;gBACD,OAAO;gBACP,GAAG;gBACH,GAAG;aACJ;YACD,OAAO,EAAE,EAAE;YACX,OAAO,EAAE,EAAE;YACX,QAAQ,EAAE,iBAAiB,CAAC,YAAY,EAAE,YAAY,CAAC;YACvD,SAAS,EAAE,YAAY;SACP;QAClB,GAAG;KACJ,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAO,GAAG,KAAK,EACf,WAAW,GAAG,CAAC,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,eAAe,GAAG,IAAI,EACtB,aAAa,GAAG,KAAK,EACrB,SAAS,EACT,OAAO,GAAG,KAAK,EACf,mBAAmB,GAuBpB;IACC,MAAM,YAAY,GAAsB,EAAE,CAAC;IAC3C,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,YAAY,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7C,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC;QAC9B,eAAe,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC;QAErG,+DAA+D;QAC/D,6DAA6D;QAC7D,MAAM,cAAc,GAAG,gBAAgB,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC,UAAU,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE;YACzG,IAAI,QAAQ,CAAC,IAAI,KAAK,KAAK,CAAC,SAAS,EAAE,CAAC;gBACtC,OAAO,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,OAAO,EAAE,cAAc,EAAE,WAAoB,EAAE,CAAC;YAC9E,CAAC;YACD,OAAO,EAAE,GAAG,QAAQ,EAAE,cAAc,EAAE,SAAkB,EAAE,CAAC;QAC7D,CAAC,CAAC,CAAC;QACH,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,mBAAmB,CAAC;YAC3C,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,QAAQ;YACR,SAAS,EAAE,cAAc;YACzB,KAAK;YACL,UAAU;YACV,SAAS,EAAE,IAAI;YACf,OAAO;YACP,WAAW;YACX,SAAS;YACT,YAAY;YACZ,iBAAiB;YACjB,QAAQ,EAAE,IAAI;YACd,UAAU,EAAE,KAAK;YACjB,UAAU;YACV,aAAa,EAAE,aAAa,IAAI,CAAC,GAAG,CAAC;YACrC,SAAS;YACT,OAAO;SACR,CAAC,CAAC;QAEH,YAAY,CAAC,IAAI,CAAC;YAChB,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,YAAY,EAAE,MAAM,CAAC,IAAI,CAAC,cAAc,EAAE,KAAK,IAAI,EAAE;YACrD,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW;YACpC,OAAO,EAAE;gBACP,QAAQ,EAAE,MAAM,CAAC,OAAO,CAAC,QAAQ,IAAK,EAAqB;gBAC3D,KAAK,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,IAAI,EAAE,CAAmB;aACtD;YACD,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;gBACvC,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,QAAQ,EAAE;oBACR,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC;oBACjE,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,KAAK;iBAC7B;aACF,CAAC,CAAC;SACJ,CAAC,CAAC;QAEH,YAAY,IAAI,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC;QACzC,eAAe,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC;IACtG,CAAC;IAED,MAAM,EAAE,MAAM,EAAE,cAAc,EAAE,GAAG,EAAE,GAAG,eAAe,CAAC;QACtD,QAAQ;QACR,YAAY;QACZ,YAAY;QACZ,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM,EAAE,KAAK;QACb,WAAW;QACX,SAAS;QACT,YAAY;KACb,CAAC,CAAC;IACH,MAAM,QAAQ,GAAG,aAAa,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;IAC1D,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAClE,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,QAAQ,EAAE,CAAC;AAC9C,CAAC"}
@@ -0,0 +1,38 @@
1
+ import type { DependencyRequirements } from '../eval-core/dependency-checker.js';
2
+ import type { Artifact, ExecutorFn, JobStore, ProgressCallback, Report, Sample, Task, VariantResult } from '../types.js';
3
+ export interface EvaluationPipelineOptions {
4
+ samplesPath: string;
5
+ skillDir: string;
6
+ samples: Sample[];
7
+ tasks: Task[];
8
+ artifacts: Artifact[];
9
+ model: string;
10
+ judgeModel: string;
11
+ noJudge: boolean;
12
+ executorName: string;
13
+ judgeExecutorName: string;
14
+ executor: ExecutorFn;
15
+ judgeExecutor: ExecutorFn;
16
+ outputDir?: string | null;
17
+ project?: string;
18
+ owner?: string;
19
+ tags?: string[];
20
+ blind?: boolean;
21
+ concurrency?: number;
22
+ timeoutMs?: number;
23
+ noCache?: boolean;
24
+ jobStore?: JobStore | null;
25
+ persistJob?: boolean;
26
+ onProgress?: ProgressCallback | null;
27
+ skipPreflight?: boolean;
28
+ verbose?: boolean;
29
+ retry?: number;
30
+ existingResults?: Record<string, Record<string, VariantResult>>;
31
+ requires?: DependencyRequirements;
32
+ layeredStats?: boolean;
33
+ }
34
+ export declare function executeEvaluationPipeline({ samplesPath, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir, project, owner, tags, blind, concurrency, timeoutMs, noCache, jobStore, persistJob, onProgress, skipPreflight, verbose, retry, existingResults, requires, layeredStats, }: EvaluationPipelineOptions): Promise<{
35
+ report: Report;
36
+ filePath: string | null;
37
+ }>;
38
+ //# sourceMappingURL=evaluation-pipeline.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"evaluation-pipeline.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAgBjF,OAAO,KAAK,EACV,QAAQ,EAIR,UAAU,EACV,QAAQ,EACR,gBAAgB,EAChB,MAAM,EACN,MAAM,EACN,IAAI,EACJ,aAAa,EACd,MAAM,aAAa,CAAC;AAuLrB,MAAM,WAAW,yBAAyB;IACxC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,QAAQ,EAAE,UAAU,CAAC;IACrB,aAAa,EAAE,UAAU,CAAC;IAC1B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IAChE,QAAQ,CAAC,EAAE,sBAAsB,CAAC;IAClC,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,WAAW,EACX,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,aAAqB,EACrB,OAAe,EACf,KAAS,EACT,eAAe,EACf,QAAQ,EACR,YAAoB,GACrB,EAAE,yBAAyB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CA4FlF"}