@plumloom/cli 0.1.0

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 (308) hide show
  1. package/LICENSE +202 -0
  2. package/NOTICE +2 -0
  3. package/README.md +71 -0
  4. package/dist/actions/configured-runs.d.ts +25 -0
  5. package/dist/actions/configured-runs.d.ts.map +1 -0
  6. package/dist/actions/configured-runs.js +35 -0
  7. package/dist/actions/configured-runs.js.map +1 -0
  8. package/dist/actions/context.d.ts +5 -0
  9. package/dist/actions/context.d.ts.map +1 -0
  10. package/dist/actions/context.js +2 -0
  11. package/dist/actions/context.js.map +1 -0
  12. package/dist/actions/doctor.d.ts +46 -0
  13. package/dist/actions/doctor.d.ts.map +1 -0
  14. package/dist/actions/doctor.js +130 -0
  15. package/dist/actions/doctor.js.map +1 -0
  16. package/dist/actions/evaluations.d.ts +87 -0
  17. package/dist/actions/evaluations.d.ts.map +1 -0
  18. package/dist/actions/evaluations.js +307 -0
  19. package/dist/actions/evaluations.js.map +1 -0
  20. package/dist/actions/gate.d.ts +63 -0
  21. package/dist/actions/gate.d.ts.map +1 -0
  22. package/dist/actions/gate.js +272 -0
  23. package/dist/actions/gate.js.map +1 -0
  24. package/dist/actions/identity.d.ts +4 -0
  25. package/dist/actions/identity.d.ts.map +1 -0
  26. package/dist/actions/identity.js +4 -0
  27. package/dist/actions/identity.js.map +1 -0
  28. package/dist/actions/models.d.ts +4 -0
  29. package/dist/actions/models.d.ts.map +1 -0
  30. package/dist/actions/models.js +4 -0
  31. package/dist/actions/models.js.map +1 -0
  32. package/dist/actions/quality-standards.d.ts +30 -0
  33. package/dist/actions/quality-standards.d.ts.map +1 -0
  34. package/dist/actions/quality-standards.js +86 -0
  35. package/dist/actions/quality-standards.js.map +1 -0
  36. package/dist/actions/results.d.ts +12 -0
  37. package/dist/actions/results.d.ts.map +1 -0
  38. package/dist/actions/results.js +65 -0
  39. package/dist/actions/results.js.map +1 -0
  40. package/dist/actions/runs.d.ts +22 -0
  41. package/dist/actions/runs.d.ts.map +1 -0
  42. package/dist/actions/runs.js +59 -0
  43. package/dist/actions/runs.js.map +1 -0
  44. package/dist/actions/suite-gate.d.ts +24 -0
  45. package/dist/actions/suite-gate.d.ts.map +1 -0
  46. package/dist/actions/suite-gate.js +40 -0
  47. package/dist/actions/suite-gate.js.map +1 -0
  48. package/dist/actions/suite.d.ts +49 -0
  49. package/dist/actions/suite.d.ts.map +1 -0
  50. package/dist/actions/suite.js +68 -0
  51. package/dist/actions/suite.js.map +1 -0
  52. package/dist/actions/validation.d.ts +3 -0
  53. package/dist/actions/validation.d.ts.map +1 -0
  54. package/dist/actions/validation.js +26 -0
  55. package/dist/actions/validation.js.map +1 -0
  56. package/dist/actions/workspaces.d.ts +13 -0
  57. package/dist/actions/workspaces.d.ts.map +1 -0
  58. package/dist/actions/workspaces.js +18 -0
  59. package/dist/actions/workspaces.js.map +1 -0
  60. package/dist/api/api.d.ts +237 -0
  61. package/dist/api/api.d.ts.map +1 -0
  62. package/dist/api/api.js +396 -0
  63. package/dist/api/api.js.map +1 -0
  64. package/dist/api/client.d.ts +38 -0
  65. package/dist/api/client.d.ts.map +1 -0
  66. package/dist/api/client.js +295 -0
  67. package/dist/api/client.js.map +1 -0
  68. package/dist/api/endpoints.d.ts +31 -0
  69. package/dist/api/endpoints.d.ts.map +1 -0
  70. package/dist/api/endpoints.js +32 -0
  71. package/dist/api/endpoints.js.map +1 -0
  72. package/dist/api/errors.d.ts +22 -0
  73. package/dist/api/errors.d.ts.map +1 -0
  74. package/dist/api/errors.js +84 -0
  75. package/dist/api/errors.js.map +1 -0
  76. package/dist/api/schemas.d.ts +169 -0
  77. package/dist/api/schemas.d.ts.map +1 -0
  78. package/dist/api/schemas.js +194 -0
  79. package/dist/api/schemas.js.map +1 -0
  80. package/dist/auth/credential-store.d.ts +9 -0
  81. package/dist/auth/credential-store.d.ts.map +1 -0
  82. package/dist/auth/credential-store.js +2 -0
  83. package/dist/auth/credential-store.js.map +1 -0
  84. package/dist/auth/credentials.d.ts +27 -0
  85. package/dist/auth/credentials.d.ts.map +1 -0
  86. package/dist/auth/credentials.js +81 -0
  87. package/dist/auth/credentials.js.map +1 -0
  88. package/dist/auth/keyring-store.d.ts +9 -0
  89. package/dist/auth/keyring-store.d.ts.map +1 -0
  90. package/dist/auth/keyring-store.js +57 -0
  91. package/dist/auth/keyring-store.js.map +1 -0
  92. package/dist/auth/login.d.ts +28 -0
  93. package/dist/auth/login.d.ts.map +1 -0
  94. package/dist/auth/login.js +27 -0
  95. package/dist/auth/login.js.map +1 -0
  96. package/dist/auth/password-prompt.d.ts +5 -0
  97. package/dist/auth/password-prompt.d.ts.map +1 -0
  98. package/dist/auth/password-prompt.js +10 -0
  99. package/dist/auth/password-prompt.js.map +1 -0
  100. package/dist/auth/redact.d.ts +3 -0
  101. package/dist/auth/redact.d.ts.map +1 -0
  102. package/dist/auth/redact.js +69 -0
  103. package/dist/auth/redact.js.map +1 -0
  104. package/dist/cli.d.ts +3 -0
  105. package/dist/cli.d.ts.map +1 -0
  106. package/dist/cli.js +4 -0
  107. package/dist/cli.js.map +1 -0
  108. package/dist/commands/executor.d.ts +36 -0
  109. package/dist/commands/executor.d.ts.map +1 -0
  110. package/dist/commands/executor.js +1218 -0
  111. package/dist/commands/executor.js.map +1 -0
  112. package/dist/commands/program.d.ts +21 -0
  113. package/dist/commands/program.d.ts.map +1 -0
  114. package/dist/commands/program.js +588 -0
  115. package/dist/commands/program.js.map +1 -0
  116. package/dist/commands/types.d.ts +139 -0
  117. package/dist/commands/types.d.ts.map +1 -0
  118. package/dist/commands/types.js +2 -0
  119. package/dist/commands/types.js.map +1 -0
  120. package/dist/config.d.ts +12 -0
  121. package/dist/config.d.ts.map +1 -0
  122. package/dist/config.js +57 -0
  123. package/dist/config.js.map +1 -0
  124. package/dist/configured-run/file-inputs.d.ts +8 -0
  125. package/dist/configured-run/file-inputs.d.ts.map +1 -0
  126. package/dist/configured-run/file-inputs.js +160 -0
  127. package/dist/configured-run/file-inputs.js.map +1 -0
  128. package/dist/configured-run/validation.d.ts +103 -0
  129. package/dist/configured-run/validation.d.ts.map +1 -0
  130. package/dist/configured-run/validation.js +327 -0
  131. package/dist/configured-run/validation.js.map +1 -0
  132. package/dist/domain/common.d.ts +6 -0
  133. package/dist/domain/common.d.ts.map +1 -0
  134. package/dist/domain/common.js +10 -0
  135. package/dist/domain/common.js.map +1 -0
  136. package/dist/domain/types.d.ts +117 -0
  137. package/dist/domain/types.d.ts.map +1 -0
  138. package/dist/domain/types.js +2 -0
  139. package/dist/domain/types.js.map +1 -0
  140. package/dist/errors/autoeval-error.d.ts +13 -0
  141. package/dist/errors/autoeval-error.d.ts.map +1 -0
  142. package/dist/errors/autoeval-error.js +36 -0
  143. package/dist/errors/autoeval-error.js.map +1 -0
  144. package/dist/errors/hints.d.ts +3 -0
  145. package/dist/errors/hints.d.ts.map +1 -0
  146. package/dist/errors/hints.js +43 -0
  147. package/dist/errors/hints.js.map +1 -0
  148. package/dist/gate/decision.d.ts +97 -0
  149. package/dist/gate/decision.d.ts.map +1 -0
  150. package/dist/gate/decision.js +436 -0
  151. package/dist/gate/decision.js.map +1 -0
  152. package/dist/gate/policy.d.ts +31 -0
  153. package/dist/gate/policy.d.ts.map +1 -0
  154. package/dist/gate/policy.js +26 -0
  155. package/dist/gate/policy.js.map +1 -0
  156. package/dist/guide/eval-creation-backend.d.ts +20 -0
  157. package/dist/guide/eval-creation-backend.d.ts.map +1 -0
  158. package/dist/guide/eval-creation-backend.js +16 -0
  159. package/dist/guide/eval-creation-backend.js.map +1 -0
  160. package/dist/guide/eval-guide.d.ts +24 -0
  161. package/dist/guide/eval-guide.d.ts.map +1 -0
  162. package/dist/guide/eval-guide.js +219 -0
  163. package/dist/guide/eval-guide.js.map +1 -0
  164. package/dist/mcp/context.d.ts +21 -0
  165. package/dist/mcp/context.d.ts.map +1 -0
  166. package/dist/mcp/context.js +19 -0
  167. package/dist/mcp/context.js.map +1 -0
  168. package/dist/mcp/schemas.d.ts +206 -0
  169. package/dist/mcp/schemas.d.ts.map +1 -0
  170. package/dist/mcp/schemas.js +80 -0
  171. package/dist/mcp/schemas.js.map +1 -0
  172. package/dist/mcp/server.d.ts +5 -0
  173. package/dist/mcp/server.d.ts.map +1 -0
  174. package/dist/mcp/server.js +10 -0
  175. package/dist/mcp/server.js.map +1 -0
  176. package/dist/mcp/tools.d.ts +42 -0
  177. package/dist/mcp/tools.d.ts.map +1 -0
  178. package/dist/mcp/tools.js +287 -0
  179. package/dist/mcp/tools.js.map +1 -0
  180. package/dist/mcp.d.ts +3 -0
  181. package/dist/mcp.d.ts.map +1 -0
  182. package/dist/mcp.js +32 -0
  183. package/dist/mcp.js.map +1 -0
  184. package/dist/output/big-digits.d.ts +7 -0
  185. package/dist/output/big-digits.d.ts.map +1 -0
  186. package/dist/output/big-digits.js +43 -0
  187. package/dist/output/big-digits.js.map +1 -0
  188. package/dist/output/clipboard.d.ts +29 -0
  189. package/dist/output/clipboard.d.ts.map +1 -0
  190. package/dist/output/clipboard.js +82 -0
  191. package/dist/output/clipboard.js.map +1 -0
  192. package/dist/output/colors.d.ts +21 -0
  193. package/dist/output/colors.d.ts.map +1 -0
  194. package/dist/output/colors.js +21 -0
  195. package/dist/output/colors.js.map +1 -0
  196. package/dist/output/digest.d.ts +165 -0
  197. package/dist/output/digest.d.ts.map +1 -0
  198. package/dist/output/digest.js +771 -0
  199. package/dist/output/digest.js.map +1 -0
  200. package/dist/output/human.d.ts +95 -0
  201. package/dist/output/human.d.ts.map +1 -0
  202. package/dist/output/human.js +643 -0
  203. package/dist/output/human.js.map +1 -0
  204. package/dist/output/json.d.ts +2 -0
  205. package/dist/output/json.d.ts.map +1 -0
  206. package/dist/output/json.js +5 -0
  207. package/dist/output/json.js.map +1 -0
  208. package/dist/output/layout.d.ts +18 -0
  209. package/dist/output/layout.d.ts.map +1 -0
  210. package/dist/output/layout.js +64 -0
  211. package/dist/output/layout.js.map +1 -0
  212. package/dist/output/markdown.d.ts +16 -0
  213. package/dist/output/markdown.d.ts.map +1 -0
  214. package/dist/output/markdown.js +38 -0
  215. package/dist/output/markdown.js.map +1 -0
  216. package/dist/output/progress.d.ts +29 -0
  217. package/dist/output/progress.d.ts.map +1 -0
  218. package/dist/output/progress.js +136 -0
  219. package/dist/output/progress.js.map +1 -0
  220. package/dist/output/safe-text.d.ts +11 -0
  221. package/dist/output/safe-text.d.ts.map +1 -0
  222. package/dist/output/safe-text.js +44 -0
  223. package/dist/output/safe-text.js.map +1 -0
  224. package/dist/output/scorecard.d.ts +23 -0
  225. package/dist/output/scorecard.d.ts.map +1 -0
  226. package/dist/output/scorecard.js +506 -0
  227. package/dist/output/scorecard.js.map +1 -0
  228. package/dist/output/section.d.ts +15 -0
  229. package/dist/output/section.d.ts.map +1 -0
  230. package/dist/output/section.js +44 -0
  231. package/dist/output/section.js.map +1 -0
  232. package/dist/output/spinner.d.ts +24 -0
  233. package/dist/output/spinner.d.ts.map +1 -0
  234. package/dist/output/spinner.js +34 -0
  235. package/dist/output/spinner.js.map +1 -0
  236. package/dist/output/splash.d.ts +37 -0
  237. package/dist/output/splash.d.ts.map +1 -0
  238. package/dist/output/splash.js +169 -0
  239. package/dist/output/splash.js.map +1 -0
  240. package/dist/output/table.d.ts +33 -0
  241. package/dist/output/table.d.ts.map +1 -0
  242. package/dist/output/table.js +105 -0
  243. package/dist/output/table.js.map +1 -0
  244. package/dist/output/trend.d.ts +19 -0
  245. package/dist/output/trend.d.ts.map +1 -0
  246. package/dist/output/trend.js +38 -0
  247. package/dist/output/trend.js.map +1 -0
  248. package/dist/output/writer.d.ts +14 -0
  249. package/dist/output/writer.d.ts.map +1 -0
  250. package/dist/output/writer.js +33 -0
  251. package/dist/output/writer.js.map +1 -0
  252. package/dist/polling/clock.d.ts +9 -0
  253. package/dist/polling/clock.d.ts.map +1 -0
  254. package/dist/polling/clock.js +23 -0
  255. package/dist/polling/clock.js.map +1 -0
  256. package/dist/polling/wait-for-result.d.ts +16 -0
  257. package/dist/polling/wait-for-result.d.ts.map +1 -0
  258. package/dist/polling/wait-for-result.js +29 -0
  259. package/dist/polling/wait-for-result.js.map +1 -0
  260. package/dist/polling/wait-for-run.d.ts +14 -0
  261. package/dist/polling/wait-for-run.d.ts.map +1 -0
  262. package/dist/polling/wait-for-run.js +36 -0
  263. package/dist/polling/wait-for-run.js.map +1 -0
  264. package/dist/quickstart/prompt.d.ts +18 -0
  265. package/dist/quickstart/prompt.d.ts.map +1 -0
  266. package/dist/quickstart/prompt.js +37 -0
  267. package/dist/quickstart/prompt.js.map +1 -0
  268. package/dist/quickstart/sample.d.ts +28 -0
  269. package/dist/quickstart/sample.d.ts.map +1 -0
  270. package/dist/quickstart/sample.js +235 -0
  271. package/dist/quickstart/sample.js.map +1 -0
  272. package/dist/quickstart/selection.d.ts +53 -0
  273. package/dist/quickstart/selection.d.ts.map +1 -0
  274. package/dist/quickstart/selection.js +97 -0
  275. package/dist/quickstart/selection.js.map +1 -0
  276. package/dist/run.d.ts +28 -0
  277. package/dist/run.d.ts.map +1 -0
  278. package/dist/run.js +134 -0
  279. package/dist/run.js.map +1 -0
  280. package/dist/runtime/action-context.d.ts +19 -0
  281. package/dist/runtime/action-context.d.ts.map +1 -0
  282. package/dist/runtime/action-context.js +40 -0
  283. package/dist/runtime/action-context.js.map +1 -0
  284. package/dist/suite/execution.d.ts +46 -0
  285. package/dist/suite/execution.d.ts.map +1 -0
  286. package/dist/suite/execution.js +97 -0
  287. package/dist/suite/execution.js.map +1 -0
  288. package/dist/suite/manifest.d.ts +16 -0
  289. package/dist/suite/manifest.d.ts.map +1 -0
  290. package/dist/suite/manifest.js +92 -0
  291. package/dist/suite/manifest.js.map +1 -0
  292. package/dist/suite/reporting.d.ts +35 -0
  293. package/dist/suite/reporting.d.ts.map +1 -0
  294. package/dist/suite/reporting.js +131 -0
  295. package/dist/suite/reporting.js.map +1 -0
  296. package/dist/trace/deepseek-harness.d.ts +31 -0
  297. package/dist/trace/deepseek-harness.d.ts.map +1 -0
  298. package/dist/trace/deepseek-harness.js +412 -0
  299. package/dist/trace/deepseek-harness.js.map +1 -0
  300. package/dist/trace/import.d.ts +9 -0
  301. package/dist/trace/import.d.ts.map +1 -0
  302. package/dist/trace/import.js +45 -0
  303. package/dist/trace/import.js.map +1 -0
  304. package/dist/version.d.ts +3 -0
  305. package/dist/version.d.ts.map +1 -0
  306. package/dist/version.js +3 -0
  307. package/dist/version.js.map +1 -0
  308. package/package.json +82 -0
@@ -0,0 +1,272 @@
1
+ import { z } from 'zod';
2
+ const modelPerformanceSchema = z
3
+ .object({
4
+ models: z
5
+ .array(z
6
+ .object({
7
+ display_name: z.string().optional(),
8
+ model_key: z.string().optional(),
9
+ is_primary: z.boolean().optional(),
10
+ scores: z
11
+ .object({
12
+ overall: z
13
+ .object({ mean: z.number().nullable(), has_data: z.boolean().optional() })
14
+ .loose()
15
+ .optional(),
16
+ })
17
+ .loose()
18
+ .optional(),
19
+ })
20
+ .loose())
21
+ .optional(),
22
+ })
23
+ .loose();
24
+ const scenarioComparisonSchema = z
25
+ .object({
26
+ evaluation_summary: z.object({ overall_mean: z.number().nullable() }).loose().optional(),
27
+ scenarios: z
28
+ .array(z
29
+ .object({
30
+ scenario_name: z.string().optional(),
31
+ model_scores: z
32
+ .array(z
33
+ .object({
34
+ model_key: z.string().optional(),
35
+ score: z.object({ mean: z.number().nullable() }).loose().optional(),
36
+ })
37
+ .loose())
38
+ .optional(),
39
+ })
40
+ .loose())
41
+ .optional(),
42
+ })
43
+ .loose();
44
+ const sessionScorecardSchema = z
45
+ .object({
46
+ overall: z.number().nullable().optional(),
47
+ judge_agreement: z
48
+ .object({
49
+ pass: z.number().nullable().optional(),
50
+ agreed: z.number().nullable().optional(),
51
+ total: z.number().nullable().optional(),
52
+ })
53
+ .loose()
54
+ .optional(),
55
+ })
56
+ .loose();
57
+ const trajectorySchema = z.object({ trajectory_score: z.number().nullable().optional() }).loose();
58
+ function readAgreement(input) {
59
+ const agreed = input?.agreed ?? input?.pass;
60
+ const total = input?.total;
61
+ if (agreed === undefined || agreed === null || total === undefined || total === null) {
62
+ return undefined;
63
+ }
64
+ if (total === 0)
65
+ return undefined;
66
+ return { agreed, total, ratio: agreed / total };
67
+ }
68
+ function extractScenarioMetrics(results) {
69
+ const metrics = {};
70
+ const performance = modelPerformanceSchema.safeParse(results.modelPerformance);
71
+ if (performance.success) {
72
+ const models = performance.data.models ?? [];
73
+ const primary = models.find((model) => model.is_primary === true) ?? models[0];
74
+ const mean = primary?.scores?.overall?.mean;
75
+ if (typeof mean === 'number')
76
+ metrics.overall = mean;
77
+ }
78
+ const comparison = scenarioComparisonSchema.safeParse(results.scenarioComparison);
79
+ if (comparison.success) {
80
+ if (metrics.overall === undefined) {
81
+ const summaryMean = comparison.data.evaluation_summary?.overall_mean;
82
+ if (typeof summaryMean === 'number')
83
+ metrics.overall = summaryMean;
84
+ }
85
+ for (const scenario of comparison.data.scenarios ?? []) {
86
+ for (const modelScore of scenario.model_scores ?? []) {
87
+ const mean = modelScore.score?.mean;
88
+ if (typeof mean !== 'number')
89
+ continue;
90
+ if (metrics.scenarioMinimum !== undefined && metrics.scenarioMinimum.score <= mean) {
91
+ continue;
92
+ }
93
+ metrics.scenarioMinimum = {
94
+ score: mean,
95
+ ...(scenario.scenario_name ? { scenarioName: scenario.scenario_name } : {}),
96
+ ...(modelScore.model_key ? { modelKey: modelScore.model_key } : {}),
97
+ };
98
+ }
99
+ }
100
+ }
101
+ return metrics;
102
+ }
103
+ function extractSessionMetrics(scorecardInput, trajectoryInput) {
104
+ const metrics = {};
105
+ const parsed = sessionScorecardSchema.safeParse(scorecardInput);
106
+ if (parsed.success) {
107
+ const scorecard = parsed.data;
108
+ if (typeof scorecard.overall === 'number')
109
+ metrics.overall = scorecard.overall;
110
+ const agreement = readAgreement(scorecard.judge_agreement);
111
+ if (agreement)
112
+ metrics.judgeAgreement = agreement;
113
+ }
114
+ if (trajectoryInput !== undefined) {
115
+ const trajectory = trajectorySchema.safeParse(trajectoryInput);
116
+ if (trajectory.success && typeof trajectory.data.trajectory_score === 'number') {
117
+ metrics.trajectoryScore = trajectory.data.trajectory_score;
118
+ }
119
+ }
120
+ return metrics;
121
+ }
122
+ export function extractGateMetrics(results) {
123
+ if (results.contextType === 'scenario')
124
+ return extractScenarioMetrics(results);
125
+ if (results.contextType === 'conversation')
126
+ return extractSessionMetrics(results.conversation);
127
+ return extractSessionMetrics(results.agentTrace, results.trajectory);
128
+ }
129
+ function formatNumber(value) {
130
+ return Number.isInteger(value) ? String(value) : value.toFixed(3).replace(/0+$/, '');
131
+ }
132
+ const NOT_AVAILABLE = 'Not available';
133
+ const perMetricSchema = z
134
+ .object({
135
+ per_metric: z
136
+ .record(z.string(), z
137
+ .object({
138
+ score: z.number().nullable().optional(),
139
+ has_data: z.boolean().nullable().optional(),
140
+ })
141
+ .loose())
142
+ .optional(),
143
+ })
144
+ .loose();
145
+ function worstVerdict(checks) {
146
+ if (checks.some((check) => check.decision === 'FAIL'))
147
+ return 'FAIL';
148
+ if (checks.some((check) => check.decision === 'INCONCLUSIVE'))
149
+ return 'INCONCLUSIVE';
150
+ return 'PASS';
151
+ }
152
+ function sessionMetricChecks(payload, thresholds) {
153
+ const parsed = perMetricSchema.safeParse(payload);
154
+ const perMetric = parsed.success ? (parsed.data.per_metric ?? {}) : {};
155
+ return Object.entries(thresholds).map(([name, threshold]) => {
156
+ const cell = perMetric[name];
157
+ const base = {
158
+ metric: `metric:${name}`,
159
+ label: name,
160
+ threshold: `>= ${formatNumber(threshold)}`,
161
+ };
162
+ const inconclusive = (actual, reason) => ({
163
+ ...base,
164
+ actual,
165
+ decision: 'INCONCLUSIVE',
166
+ passed: false,
167
+ reason,
168
+ });
169
+ if (cell === undefined) {
170
+ return inconclusive(NOT_AVAILABLE, 'required metric is missing from the result payload');
171
+ }
172
+ if (cell.has_data === false) {
173
+ return inconclusive(NOT_AVAILABLE, 'metric reported has_data: false');
174
+ }
175
+ if (typeof cell.score !== 'number') {
176
+ return inconclusive(NOT_AVAILABLE, 'required score is unavailable');
177
+ }
178
+ const passed = cell.score >= threshold;
179
+ return {
180
+ ...base,
181
+ actual: formatNumber(cell.score),
182
+ decision: passed ? 'PASS' : 'FAIL',
183
+ passed,
184
+ ...(passed ? {} : { reason: 'score is below the required threshold' }),
185
+ };
186
+ });
187
+ }
188
+ /** Canonical Scenario gate semantics. */
189
+ function scenarioChecks(metrics, thresholds) {
190
+ const checks = [];
191
+ if (thresholds.minOverall !== undefined) {
192
+ const actual = metrics.overall;
193
+ const passed = actual !== undefined && actual >= thresholds.minOverall;
194
+ checks.push({
195
+ metric: 'overall',
196
+ label: 'Overall score',
197
+ threshold: `>= ${formatNumber(thresholds.minOverall)}`,
198
+ actual: actual === undefined ? NOT_AVAILABLE : formatNumber(actual),
199
+ decision: passed ? 'PASS' : 'FAIL',
200
+ passed,
201
+ });
202
+ }
203
+ if (thresholds.minScenario !== undefined) {
204
+ const lowest = metrics.scenarioMinimum;
205
+ const detail = lowest === undefined
206
+ ? undefined
207
+ : [lowest.scenarioName, lowest.modelKey].filter(Boolean).join(' · ');
208
+ const passed = lowest !== undefined && lowest.score >= thresholds.minScenario;
209
+ checks.push({
210
+ metric: 'scenario-minimum',
211
+ label: 'Lowest scenario score',
212
+ threshold: `>= ${formatNumber(thresholds.minScenario)}`,
213
+ actual: lowest === undefined ? NOT_AVAILABLE : formatNumber(lowest.score),
214
+ decision: passed ? 'PASS' : 'FAIL',
215
+ passed,
216
+ ...(detail ? { detail } : {}),
217
+ });
218
+ }
219
+ if (thresholds.minJudgeAgreement !== undefined) {
220
+ const agreement = metrics.judgeAgreement;
221
+ const passed = agreement !== undefined && agreement.ratio >= thresholds.minJudgeAgreement;
222
+ checks.push({
223
+ metric: 'judge-agreement',
224
+ label: 'Judge agreement',
225
+ threshold: `>= ${formatNumber(thresholds.minJudgeAgreement)}`,
226
+ actual: agreement === undefined
227
+ ? NOT_AVAILABLE
228
+ : `${formatNumber(agreement.ratio)} (${agreement.agreed}/${agreement.total})`,
229
+ decision: passed ? 'PASS' : 'FAIL',
230
+ passed,
231
+ });
232
+ }
233
+ return checks;
234
+ }
235
+ /**
236
+ * Compare evaluation results against release thresholds. Pure: it performs no
237
+ * I/O and never inspects credentials or raw upstream diagnostics. Scenario
238
+ * evaluations retain the canonical score and agreement checks. Conversation
239
+ * and agent-trace evaluations gate only on configured `per_metric` scores;
240
+ * outcome-judge fields never contribute evidence.
241
+ */
242
+ export function evaluateGate(results, thresholds) {
243
+ const metrics = extractGateMetrics(results);
244
+ if (results.contextType !== 'scenario') {
245
+ const payload = results.contextType === 'conversation' ? results.conversation : results.agentTrace;
246
+ const configured = thresholds.metrics ?? {};
247
+ const checks = sessionMetricChecks(payload, configured);
248
+ const advisory = Object.keys(configured).length === 0;
249
+ const decision = advisory ? 'INCONCLUSIVE' : worstVerdict(checks);
250
+ return {
251
+ contextType: results.contextType,
252
+ decision,
253
+ passed: decision === 'PASS',
254
+ advisory,
255
+ thresholds,
256
+ metrics,
257
+ checks,
258
+ };
259
+ }
260
+ const checks = scenarioChecks(metrics, thresholds);
261
+ const passed = checks.every((check) => check.passed);
262
+ return {
263
+ contextType: results.contextType,
264
+ decision: passed ? 'PASS' : 'FAIL',
265
+ passed,
266
+ advisory: checks.length === 0,
267
+ thresholds,
268
+ metrics,
269
+ checks,
270
+ };
271
+ }
272
+ //# sourceMappingURL=gate.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"gate.js","sourceRoot":"","sources":["../../src/actions/gate.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAqDxB,MAAM,sBAAsB,GAAG,CAAC;KAC7B,MAAM,CAAC;IACN,MAAM,EAAE,CAAC;SACN,KAAK,CACJ,CAAC;SACE,MAAM,CAAC;QACN,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;QACnC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;QAChC,UAAU,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;QAClC,MAAM,EAAE,CAAC;aACN,MAAM,CAAC;YACN,OAAO,EAAE,CAAC;iBACP,MAAM,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,EAAE,QAAQ,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC;iBACzE,KAAK,EAAE;iBACP,QAAQ,EAAE;SACd,CAAC;aACD,KAAK,EAAE;aACP,QAAQ,EAAE;KACd,CAAC;SACD,KAAK,EAAE,CACX;SACA,QAAQ,EAAE;CACd,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,MAAM,wBAAwB,GAAG,CAAC;KAC/B,MAAM,CAAC;IACN,kBAAkB,EAAE,CAAC,CAAC,MAAM,CAAC,EAAE,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE;IACxF,SAAS,EAAE,CAAC;SACT,KAAK,CACJ,CAAC;SACE,MAAM,CAAC;QACN,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;QACpC,YAAY,EAAE,CAAC;aACZ,KAAK,CACJ,CAAC;aACE,MAAM,CAAC;YACN,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;YAChC,KAAK,EAAE,CAAC,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE;SACpE,CAAC;aACD,KAAK,EAAE,CACX;aACA,QAAQ,EAAE;KACd,CAAC;SACD,KAAK,EAAE,CACX;SACA,QAAQ,EAAE;CACd,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,MAAM,sBAAsB,GAAG,CAAC;KAC7B,MAAM,CAAC;IACN,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACzC,eAAe,EAAE,CAAC;SACf,MAAM,CAAC;QACN,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;QACtC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;QACxC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;KACxC,CAAC;SACD,KAAK,EAAE;SACP,QAAQ,EAAE;CACd,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,MAAM,gBAAgB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC;AAElG,SAAS,aAAa,CACpB,KAMa;IAEb,MAAM,MAAM,GAAG,KAAK,EAAE,MAAM,IAAI,KAAK,EAAE,IAAI,CAAC;IAC5C,MAAM,KAAK,GAAG,KAAK,EAAE,KAAK,CAAC;IAC3B,IAAI,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,IAAI,IAAI,KAAK,KAAK,SAAS,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;QACrF,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,IAAI,KAAK,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAClC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,GAAG,KAAK,EAAE,CAAC;AAClD,CAAC;AAED,SAAS,sBAAsB,CAAC,OAAgE;IAC9F,MAAM,OAAO,GAAgB,EAAE,CAAC;IAChC,MAAM,WAAW,GAAG,sBAAsB,CAAC,SAAS,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;IAC/E,IAAI,WAAW,CAAC,OAAO,EAAE,CAAC;QACxB,MAAM,MAAM,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,IAAI,EAAE,CAAC;QAC7C,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,UAAU,KAAK,IAAI,CAAC,IAAI,MAAM,CAAC,CAAC,CAAC,CAAC;QAC/E,MAAM,IAAI,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,IAAI,CAAC;QAC5C,IAAI,OAAO,IAAI,KAAK,QAAQ;YAAE,OAAO,CAAC,OAAO,GAAG,IAAI,CAAC;IACvD,CAAC;IAED,MAAM,UAAU,GAAG,wBAAwB,CAAC,SAAS,CAAC,OAAO,CAAC,kBAAkB,CAAC,CAAC;IAClF,IAAI,UAAU,CAAC,OAAO,EAAE,CAAC;QACvB,IAAI,OAAO,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;YAClC,MAAM,WAAW,GAAG,UAAU,CAAC,IAAI,CAAC,kBAAkB,EAAE,YAAY,CAAC;YACrE,IAAI,OAAO,WAAW,KAAK,QAAQ;gBAAE,OAAO,CAAC,OAAO,GAAG,WAAW,CAAC;QACrE,CAAC;QACD,KAAK,MAAM,QAAQ,IAAI,UAAU,CAAC,IAAI,CAAC,SAAS,IAAI,EAAE,EAAE,CAAC;YACvD,KAAK,MAAM,UAAU,IAAI,QAAQ,CAAC,YAAY,IAAI,EAAE,EAAE,CAAC;gBACrD,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,EAAE,IAAI,CAAC;gBACpC,IAAI,OAAO,IAAI,KAAK,QAAQ;oBAAE,SAAS;gBACvC,IAAI,OAAO,CAAC,eAAe,KAAK,SAAS,IAAI,OAAO,CAAC,eAAe,CAAC,KAAK,IAAI,IAAI,EAAE,CAAC;oBACnF,SAAS;gBACX,CAAC;gBACD,OAAO,CAAC,eAAe,GAAG;oBACxB,KAAK,EAAE,IAAI;oBACX,GAAG,CAAC,QAAQ,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,QAAQ,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;oBAC3E,GAAG,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,UAAU,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;iBACpE,CAAC;YACJ,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,qBAAqB,CAAC,cAAuB,EAAE,eAAyB;IAC/E,MAAM,OAAO,GAAgB,EAAE,CAAC;IAChC,MAAM,MAAM,GAAG,sBAAsB,CAAC,SAAS,CAAC,cAAc,CAAC,CAAC;IAChE,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACnB,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,CAAC;QAC9B,IAAI,OAAO,SAAS,CAAC,OAAO,KAAK,QAAQ;YAAE,OAAO,CAAC,OAAO,GAAG,SAAS,CAAC,OAAO,CAAC;QAC/E,MAAM,SAAS,GAAG,aAAa,CAAC,SAAS,CAAC,eAAe,CAAC,CAAC;QAC3D,IAAI,SAAS;YAAE,OAAO,CAAC,cAAc,GAAG,SAAS,CAAC;IACpD,CAAC;IACD,IAAI,eAAe,KAAK,SAAS,EAAE,CAAC;QAClC,MAAM,UAAU,GAAG,gBAAgB,CAAC,SAAS,CAAC,eAAe,CAAC,CAAC;QAC/D,IAAI,UAAU,CAAC,OAAO,IAAI,OAAO,UAAU,CAAC,IAAI,CAAC,gBAAgB,KAAK,QAAQ,EAAE,CAAC;YAC/E,OAAO,CAAC,eAAe,GAAG,UAAU,CAAC,IAAI,CAAC,gBAAgB,CAAC;QAC7D,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,OAA0B;IAC3D,IAAI,OAAO,CAAC,WAAW,KAAK,UAAU;QAAE,OAAO,sBAAsB,CAAC,OAAO,CAAC,CAAC;IAC/E,IAAI,OAAO,CAAC,WAAW,KAAK,cAAc;QAAE,OAAO,qBAAqB,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC;IAC/F,OAAO,qBAAqB,CAAC,OAAO,CAAC,UAAU,EAAE,OAAO,CAAC,UAAU,CAAC,CAAC;AACvE,CAAC;AAED,SAAS,YAAY,CAAC,KAAa;IACjC,OAAO,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;AACvF,CAAC;AAED,MAAM,aAAa,GAAG,eAAe,CAAC;AAEtC,MAAM,eAAe,GAAG,CAAC;KACtB,MAAM,CAAC;IACN,UAAU,EAAE,CAAC;SACV,MAAM,CACL,CAAC,CAAC,MAAM,EAAE,EACV,CAAC;SACE,MAAM,CAAC;QACN,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;QACvC,QAAQ,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;KAC5C,CAAC;SACD,KAAK,EAAE,CACX;SACA,QAAQ,EAAE;CACd,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,SAAS,YAAY,CAAC,MAA4B;IAChD,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,MAAM,CAAC;QAAE,OAAO,MAAM,CAAC;IACrE,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,cAAc,CAAC;QAAE,OAAO,cAAc,CAAC;IACrF,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,SAAS,mBAAmB,CAC1B,OAAgB,EAChB,UAA4C;IAE5C,MAAM,MAAM,GAAG,eAAe,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;IAClD,MAAM,SAAS,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IAEvE,OAAO,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,SAAS,CAAC,EAAE,EAAE;QAC1D,MAAM,IAAI,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;QAC7B,MAAM,IAAI,GAAG;YACX,MAAM,EAAE,UAAU,IAAI,EAAE;YACxB,KAAK,EAAE,IAAI;YACX,SAAS,EAAE,MAAM,YAAY,CAAC,SAAS,CAAC,EAAE;SAC3C,CAAC;QACF,MAAM,YAAY,GAAG,CAAC,MAAc,EAAE,MAAc,EAAa,EAAE,CAAC,CAAC;YACnE,GAAG,IAAI;YACP,MAAM;YACN,QAAQ,EAAE,cAAc;YACxB,MAAM,EAAE,KAAK;YACb,MAAM;SACP,CAAC,CAAC;QAEH,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;YACvB,OAAO,YAAY,CAAC,aAAa,EAAE,oDAAoD,CAAC,CAAC;QAC3F,CAAC;QACD,IAAI,IAAI,CAAC,QAAQ,KAAK,KAAK,EAAE,CAAC;YAC5B,OAAO,YAAY,CAAC,aAAa,EAAE,iCAAiC,CAAC,CAAC;QACxE,CAAC;QACD,IAAI,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,EAAE,CAAC;YACnC,OAAO,YAAY,CAAC,aAAa,EAAE,+BAA+B,CAAC,CAAC;QACtE,CAAC;QACD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,IAAI,SAAS,CAAC;QACvC,OAAO;YACL,GAAG,IAAI;YACP,MAAM,EAAE,YAAY,CAAC,IAAI,CAAC,KAAK,CAAC;YAChC,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;YAClC,MAAM;YACN,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,uCAAuC,EAAE,CAAC;SACvE,CAAC;IACJ,CAAC,CAAC,CAAC;AACL,CAAC;AAED,yCAAyC;AACzC,SAAS,cAAc,CAAC,OAAoB,EAAE,UAA0B;IACtE,MAAM,MAAM,GAAgB,EAAE,CAAC;IAE/B,IAAI,UAAU,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;QAC/B,MAAM,MAAM,GAAG,MAAM,KAAK,SAAS,IAAI,MAAM,IAAI,UAAU,CAAC,UAAU,CAAC;QACvE,MAAM,CAAC,IAAI,CAAC;YACV,MAAM,EAAE,SAAS;YACjB,KAAK,EAAE,eAAe;YACtB,SAAS,EAAE,MAAM,YAAY,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE;YACtD,MAAM,EAAE,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,YAAY,CAAC,MAAM,CAAC;YACnE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;YAClC,MAAM;SACP,CAAC,CAAC;IACL,CAAC;IAED,IAAI,UAAU,CAAC,WAAW,KAAK,SAAS,EAAE,CAAC;QACzC,MAAM,MAAM,GAAG,OAAO,CAAC,eAAe,CAAC;QACvC,MAAM,MAAM,GACV,MAAM,KAAK,SAAS;YAClB,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,CAAC,MAAM,CAAC,YAAY,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACzE,MAAM,MAAM,GAAG,MAAM,KAAK,SAAS,IAAI,MAAM,CAAC,KAAK,IAAI,UAAU,CAAC,WAAW,CAAC;QAC9E,MAAM,CAAC,IAAI,CAAC;YACV,MAAM,EAAE,kBAAkB;YAC1B,KAAK,EAAE,uBAAuB;YAC9B,SAAS,EAAE,MAAM,YAAY,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE;YACvD,MAAM,EAAE,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC;YACzE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;YAClC,MAAM;YACN,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC9B,CAAC,CAAC;IACL,CAAC;IAED,IAAI,UAAU,CAAC,iBAAiB,KAAK,SAAS,EAAE,CAAC;QAC/C,MAAM,SAAS,GAAG,OAAO,CAAC,cAAc,CAAC;QACzC,MAAM,MAAM,GAAG,SAAS,KAAK,SAAS,IAAI,SAAS,CAAC,KAAK,IAAI,UAAU,CAAC,iBAAiB,CAAC;QAC1F,MAAM,CAAC,IAAI,CAAC;YACV,MAAM,EAAE,iBAAiB;YACzB,KAAK,EAAE,iBAAiB;YACxB,SAAS,EAAE,MAAM,YAAY,CAAC,UAAU,CAAC,iBAAiB,CAAC,EAAE;YAC7D,MAAM,EACJ,SAAS,KAAK,SAAS;gBACrB,CAAC,CAAC,aAAa;gBACf,CAAC,CAAC,GAAG,YAAY,CAAC,SAAS,CAAC,KAAK,CAAC,KAAK,SAAS,CAAC,MAAM,IAAI,SAAS,CAAC,KAAK,GAAG;YACjF,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;YAClC,MAAM;SACP,CAAC,CAAC;IACL,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,YAAY,CAAC,OAA0B,EAAE,UAA0B;IACjF,MAAM,OAAO,GAAG,kBAAkB,CAAC,OAAO,CAAC,CAAC;IAC5C,IAAI,OAAO,CAAC,WAAW,KAAK,UAAU,EAAE,CAAC;QACvC,MAAM,OAAO,GACX,OAAO,CAAC,WAAW,KAAK,cAAc,CAAC,CAAC,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC;QACrF,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,IAAI,EAAE,CAAC;QAC5C,MAAM,MAAM,GAAG,mBAAmB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;QACxD,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC;QACtD,MAAM,QAAQ,GAAG,QAAQ,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC;QAClE,OAAO;YACL,WAAW,EAAE,OAAO,CAAC,WAAW;YAChC,QAAQ;YACR,MAAM,EAAE,QAAQ,KAAK,MAAM;YAC3B,QAAQ;YACR,UAAU;YACV,OAAO;YACP,MAAM;SACP,CAAC;IACJ,CAAC;IAED,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;IACnD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IACrD,OAAO;QACL,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;QAClC,MAAM;QACN,QAAQ,EAAE,MAAM,CAAC,MAAM,KAAK,CAAC;QAC7B,UAAU;QACV,OAAO;QACP,MAAM;KACP,CAAC;AACJ,CAAC"}
@@ -0,0 +1,4 @@
1
+ import type { UserIdentity } from '../domain/types.js';
2
+ import type { ActionContext } from './context.js';
3
+ export declare function whoAmI(context: ActionContext, signal?: AbortSignal): Promise<UserIdentity>;
4
+ //# sourceMappingURL=identity.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"identity.d.ts","sourceRoot":"","sources":["../../src/actions/identity.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AACvD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAElD,wBAAgB,MAAM,CAAC,OAAO,EAAE,aAAa,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,YAAY,CAAC,CAE1F"}
@@ -0,0 +1,4 @@
1
+ export function whoAmI(context, signal) {
2
+ return context.api.getCurrentUser(signal);
3
+ }
4
+ //# sourceMappingURL=identity.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"identity.js","sourceRoot":"","sources":["../../src/actions/identity.ts"],"names":[],"mappings":"AAGA,MAAM,UAAU,MAAM,CAAC,OAAsB,EAAE,MAAoB;IACjE,OAAO,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC;AAC5C,CAAC"}
@@ -0,0 +1,4 @@
1
+ import type { SupportedModel } from '../domain/types.js';
2
+ import type { ActionContext } from './context.js';
3
+ export declare function getModels(context: ActionContext, signal?: AbortSignal): Promise<SupportedModel[]>;
4
+ //# sourceMappingURL=models.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"models.d.ts","sourceRoot":"","sources":["../../src/actions/models.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,oBAAoB,CAAC;AACzD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAElD,wBAAgB,SAAS,CAAC,OAAO,EAAE,aAAa,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,cAAc,EAAE,CAAC,CAEjG"}
@@ -0,0 +1,4 @@
1
+ export function getModels(context, signal) {
2
+ return context.api.getModels(signal);
3
+ }
4
+ //# sourceMappingURL=models.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"models.js","sourceRoot":"","sources":["../../src/actions/models.ts"],"names":[],"mappings":"AAGA,MAAM,UAAU,SAAS,CAAC,OAAsB,EAAE,MAAoB;IACpE,OAAO,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;AACvC,CAAC"}
@@ -0,0 +1,30 @@
1
+ import { z } from 'zod';
2
+ import type { ActionContext } from './context.js';
3
+ export declare const qualityStandardCreateSchema: z.ZodObject<{
4
+ name: z.ZodString;
5
+ judge_model: z.ZodUUID;
6
+ rubric: z.ZodString;
7
+ anchors: z.ZodArray<z.ZodObject<{
8
+ input: z.ZodString;
9
+ response: z.ZodString;
10
+ reference: z.ZodString;
11
+ score: z.ZodUnion<readonly [z.ZodLiteral<1>, z.ZodLiteral<5>]>;
12
+ reasoning: z.ZodString;
13
+ }, z.core.$strict>>;
14
+ }, z.core.$strict>;
15
+ export type QualityStandardCreateInput = z.infer<typeof qualityStandardCreateSchema>;
16
+ export type CreatedQualityStandard = {
17
+ qualityStandardId?: string;
18
+ raw: Record<string, unknown>;
19
+ };
20
+ export declare function createQualityStandard(context: ActionContext, input: QualityStandardCreateInput, signal?: AbortSignal): Promise<CreatedQualityStandard>;
21
+ export declare function assignQualityStandardToWorkspace(context: ActionContext, input: {
22
+ workspaceId: string;
23
+ qualityStandardId: string;
24
+ }, signal?: AbortSignal): Promise<Record<string, unknown>>;
25
+ export declare function updateQualityStandard(context: ActionContext, input: {
26
+ qualityStandardId: string;
27
+ } & QualityStandardCreateInput, signal?: AbortSignal): Promise<CreatedQualityStandard>;
28
+ export declare function getWorkspaceQualityStandard(context: ActionContext, workspaceIdInput: string, signal?: AbortSignal): Promise<Record<string, unknown> | null>;
29
+ export declare function getQualityStandard(context: ActionContext, qualityStandardIdInput: string, signal?: AbortSignal): Promise<Record<string, unknown>>;
30
+ //# sourceMappingURL=quality-standards.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"quality-standards.d.ts","sourceRoot":"","sources":["../../src/actions/quality-standards.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAalD,eAAO,MAAM,2BAA2B;;;;;;;;;;;kBAO7B,CAAC;AAEZ,MAAM,MAAM,0BAA0B,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,2BAA2B,CAAC,CAAC;AAErF,MAAM,MAAM,sBAAsB,GAAG;IACnC,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAC9B,CAAC;AAOF,wBAAsB,qBAAqB,CACzC,OAAO,EAAE,aAAa,EACtB,KAAK,EAAE,0BAA0B,EACjC,MAAM,CAAC,EAAE,WAAW,GACnB,OAAO,CAAC,sBAAsB,CAAC,CAajC;AAED,wBAAsB,gCAAgC,CACpD,OAAO,EAAE,aAAa,EACtB,KAAK,EAAE;IAAE,WAAW,EAAE,MAAM,CAAC;IAAC,iBAAiB,EAAE,MAAM,CAAA;CAAE,EACzD,MAAM,CAAC,EAAE,WAAW,GACnB,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAQlC;AAED,wBAAsB,qBAAqB,CACzC,OAAO,EAAE,aAAa,EACtB,KAAK,EAAE;IAAE,iBAAiB,EAAE,MAAM,CAAA;CAAE,GAAG,0BAA0B,EACjE,MAAM,CAAC,EAAE,WAAW,GACnB,OAAO,CAAC,sBAAsB,CAAC,CAyBjC;AAED,wBAAsB,2BAA2B,CAC/C,OAAO,EAAE,aAAa,EACtB,gBAAgB,EAAE,MAAM,EACxB,MAAM,CAAC,EAAE,WAAW,GACnB,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI,CAAC,CAGzC;AAED,wBAAsB,kBAAkB,CACtC,OAAO,EAAE,aAAa,EACtB,sBAAsB,EAAE,MAAM,EAC9B,MAAM,CAAC,EAAE,WAAW,GACnB,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAmBlC"}
@@ -0,0 +1,86 @@
1
+ import { z } from 'zod';
2
+ import { AutoevalError } from '../errors/autoeval-error.js';
3
+ import { requireUuid } from './validation.js';
4
+ const qualityStandardAnchorSchema = z
5
+ .object({
6
+ input: z.string().min(1),
7
+ response: z.string().min(1),
8
+ reference: z.string().min(1),
9
+ score: z.union([z.literal(1), z.literal(5)]),
10
+ reasoning: z.string().min(1),
11
+ })
12
+ .strict();
13
+ export const qualityStandardCreateSchema = z
14
+ .object({
15
+ name: z.string().trim().min(1),
16
+ judge_model: z.uuid(),
17
+ rubric: z.string().min(1),
18
+ anchors: z.array(qualityStandardAnchorSchema).min(1).max(5),
19
+ })
20
+ .strict();
21
+ function extractQualityStandardId(response) {
22
+ const idCandidate = response.quality_standard_id ?? response.qualityStandardId ?? response.id;
23
+ return typeof idCandidate === 'string' ? idCandidate : undefined;
24
+ }
25
+ export async function createQualityStandard(context, input, signal) {
26
+ const validated = qualityStandardCreateSchema.parse(input);
27
+ const response = await context.api.createQualityStandard(validated, signal);
28
+ const qualityStandardId = extractQualityStandardId(response);
29
+ if (qualityStandardId !== undefined) {
30
+ requireUuid(qualityStandardId, 'Quality standard ID');
31
+ }
32
+ return {
33
+ ...(qualityStandardId !== undefined ? { qualityStandardId } : {}),
34
+ raw: response,
35
+ };
36
+ }
37
+ export async function assignQualityStandardToWorkspace(context, input, signal) {
38
+ const workspaceId = requireUuid(input.workspaceId, 'Workspace ID');
39
+ const qualityStandardId = requireUuid(input.qualityStandardId, 'Quality standard ID');
40
+ return context.api.assignQualityStandardToWorkspace(workspaceId, { quality_standard_id: qualityStandardId }, signal);
41
+ }
42
+ export async function updateQualityStandard(context, input, signal) {
43
+ const qualityStandardId = requireUuid(input.qualityStandardId, 'Quality standard ID');
44
+ const validated = qualityStandardCreateSchema.parse({
45
+ name: input.name,
46
+ judge_model: input.judge_model,
47
+ rubric: input.rubric,
48
+ anchors: input.anchors,
49
+ });
50
+ const response = await context.api.updateQualityStandard(qualityStandardId, validated, signal);
51
+ const responseQualityStandardId = extractQualityStandardId(response);
52
+ if (responseQualityStandardId !== undefined) {
53
+ requireUuid(responseQualityStandardId, 'Quality standard ID');
54
+ if (responseQualityStandardId !== qualityStandardId) {
55
+ throw new AutoevalError('Autoeval API returned a mismatched quality standard identifier.', {
56
+ kind: 'upstream',
57
+ code: 'MISMATCHED_QUALITY_STANDARD_ID',
58
+ });
59
+ }
60
+ }
61
+ return {
62
+ qualityStandardId,
63
+ raw: response,
64
+ };
65
+ }
66
+ export async function getWorkspaceQualityStandard(context, workspaceIdInput, signal) {
67
+ const workspaceId = requireUuid(workspaceIdInput, 'Workspace ID');
68
+ return context.api.getWorkspaceQualityStandard(workspaceId, signal);
69
+ }
70
+ export async function getQualityStandard(context, qualityStandardIdInput, signal) {
71
+ const qualityStandardId = requireUuid(qualityStandardIdInput, 'Quality standard ID');
72
+ const response = await context.api.getQualityStandard(qualityStandardId, signal);
73
+ const responseId = typeof response.quality_standard_id === 'string'
74
+ ? response.quality_standard_id
75
+ : typeof response.id === 'string'
76
+ ? response.id
77
+ : undefined;
78
+ if (responseId !== undefined && responseId !== qualityStandardId) {
79
+ throw new AutoevalError('Autoeval API returned a mismatched quality standard identifier.', {
80
+ kind: 'upstream',
81
+ code: 'MISMATCHED_QUALITY_STANDARD_ID',
82
+ });
83
+ }
84
+ return response;
85
+ }
86
+ //# sourceMappingURL=quality-standards.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"quality-standards.js","sourceRoot":"","sources":["../../src/actions/quality-standards.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,aAAa,EAAE,MAAM,6BAA6B,CAAC;AAE5D,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AAE9C,MAAM,2BAA2B,GAAG,CAAC;KAClC,MAAM,CAAC;IACN,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACxB,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC3B,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC5B,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;IAC5C,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;CAC7B,CAAC;KACD,MAAM,EAAE,CAAC;AAEZ,MAAM,CAAC,MAAM,2BAA2B,GAAG,CAAC;KACzC,MAAM,CAAC;IACN,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC9B,WAAW,EAAE,CAAC,CAAC,IAAI,EAAE;IACrB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACzB,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,2BAA2B,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;CAC5D,CAAC;KACD,MAAM,EAAE,CAAC;AASZ,SAAS,wBAAwB,CAAC,QAAiC;IACjE,MAAM,WAAW,GAAG,QAAQ,CAAC,mBAAmB,IAAI,QAAQ,CAAC,iBAAiB,IAAI,QAAQ,CAAC,EAAE,CAAC;IAC9F,OAAO,OAAO,WAAW,KAAK,QAAQ,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;AACnE,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,qBAAqB,CACzC,OAAsB,EACtB,KAAiC,EACjC,MAAoB;IAEpB,MAAM,SAAS,GAAG,2BAA2B,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IAC3D,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;IAC5E,MAAM,iBAAiB,GAAG,wBAAwB,CAAC,QAAQ,CAAC,CAAC;IAE7D,IAAI,iBAAiB,KAAK,SAAS,EAAE,CAAC;QACpC,WAAW,CAAC,iBAAiB,EAAE,qBAAqB,CAAC,CAAC;IACxD,CAAC;IAED,OAAO;QACL,GAAG,CAAC,iBAAiB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACjE,GAAG,EAAE,QAAQ;KACd,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gCAAgC,CACpD,OAAsB,EACtB,KAAyD,EACzD,MAAoB;IAEpB,MAAM,WAAW,GAAG,WAAW,CAAC,KAAK,CAAC,WAAW,EAAE,cAAc,CAAC,CAAC;IACnE,MAAM,iBAAiB,GAAG,WAAW,CAAC,KAAK,CAAC,iBAAiB,EAAE,qBAAqB,CAAC,CAAC;IACtF,OAAO,OAAO,CAAC,GAAG,CAAC,gCAAgC,CACjD,WAAW,EACX,EAAE,mBAAmB,EAAE,iBAAiB,EAAE,EAC1C,MAAM,CACP,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,qBAAqB,CACzC,OAAsB,EACtB,KAAiE,EACjE,MAAoB;IAEpB,MAAM,iBAAiB,GAAG,WAAW,CAAC,KAAK,CAAC,iBAAiB,EAAE,qBAAqB,CAAC,CAAC;IACtF,MAAM,SAAS,GAAG,2BAA2B,CAAC,KAAK,CAAC;QAClD,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,WAAW,EAAE,KAAK,CAAC,WAAW;QAC9B,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,OAAO,EAAE,KAAK,CAAC,OAAO;KACvB,CAAC,CAAC;IACH,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC,iBAAiB,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;IAC/F,MAAM,yBAAyB,GAAG,wBAAwB,CAAC,QAAQ,CAAC,CAAC;IAErE,IAAI,yBAAyB,KAAK,SAAS,EAAE,CAAC;QAC5C,WAAW,CAAC,yBAAyB,EAAE,qBAAqB,CAAC,CAAC;QAC9D,IAAI,yBAAyB,KAAK,iBAAiB,EAAE,CAAC;YACpD,MAAM,IAAI,aAAa,CAAC,iEAAiE,EAAE;gBACzF,IAAI,EAAE,UAAU;gBAChB,IAAI,EAAE,gCAAgC;aACvC,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,OAAO;QACL,iBAAiB;QACjB,GAAG,EAAE,QAAQ;KACd,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,2BAA2B,CAC/C,OAAsB,EACtB,gBAAwB,EACxB,MAAoB;IAEpB,MAAM,WAAW,GAAG,WAAW,CAAC,gBAAgB,EAAE,cAAc,CAAC,CAAC;IAClE,OAAO,OAAO,CAAC,GAAG,CAAC,2BAA2B,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC;AACtE,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,OAAsB,EACtB,sBAA8B,EAC9B,MAAoB;IAEpB,MAAM,iBAAiB,GAAG,WAAW,CAAC,sBAAsB,EAAE,qBAAqB,CAAC,CAAC;IACrF,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,kBAAkB,CAAC,iBAAiB,EAAE,MAAM,CAAC,CAAC;IAEjF,MAAM,UAAU,GACd,OAAO,QAAQ,CAAC,mBAAmB,KAAK,QAAQ;QAC9C,CAAC,CAAC,QAAQ,CAAC,mBAAmB;QAC9B,CAAC,CAAC,OAAO,QAAQ,CAAC,EAAE,KAAK,QAAQ;YAC/B,CAAC,CAAC,QAAQ,CAAC,EAAE;YACb,CAAC,CAAC,SAAS,CAAC;IAElB,IAAI,UAAU,KAAK,SAAS,IAAI,UAAU,KAAK,iBAAiB,EAAE,CAAC;QACjE,MAAM,IAAI,aAAa,CAAC,iEAAiE,EAAE;YACzF,IAAI,EAAE,UAAU;YAChB,IAAI,EAAE,gCAAgC;SACvC,CAAC,CAAC;IACL,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC"}
@@ -0,0 +1,12 @@
1
+ import type { EvaluationResults } from '../domain/types.js';
2
+ import type { PollClock } from '../polling/clock.js';
3
+ import type { ActionContext } from './context.js';
4
+ export declare function getResults(context: ActionContext, input: {
5
+ evaluationId: string;
6
+ runId: string;
7
+ clock: PollClock;
8
+ resultReadyIntervalMs: number;
9
+ resultReadyTimeoutMs: number;
10
+ signal?: AbortSignal;
11
+ }): Promise<EvaluationResults>;
12
+ //# sourceMappingURL=results.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"results.d.ts","sourceRoot":"","sources":["../../src/actions/results.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AAC5D,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAGrD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAgClD,wBAAsB,UAAU,CAC9B,OAAO,EAAE,aAAa,EACtB,KAAK,EAAE;IACL,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,SAAS,CAAC;IACjB,qBAAqB,EAAE,MAAM,CAAC;IAC9B,oBAAoB,EAAE,MAAM,CAAC;IAC7B,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GACA,OAAO,CAAC,iBAAiB,CAAC,CA2D5B"}
@@ -0,0 +1,65 @@
1
+ import { waitForResult } from '../polling/wait-for-result.js';
2
+ import { ApiError } from '../api/errors.js';
3
+ import { resolveEvaluationVersion } from './evaluations.js';
4
+ import { requireUuid } from './validation.js';
5
+ function loadWhenReady(load, input, isReady) {
6
+ return waitForResult({ load, ...input, ...(isReady ? { isReady } : {}) });
7
+ }
8
+ function valueAt(payload, ...path) {
9
+ let current = payload;
10
+ for (const key of path) {
11
+ if (typeof current !== 'object' || current === null)
12
+ return undefined;
13
+ current = current[key];
14
+ }
15
+ return current;
16
+ }
17
+ function nonEmptyArrayAt(payload, ...path) {
18
+ const value = valueAt(payload, ...path);
19
+ return Array.isArray(value) && value.length > 0;
20
+ }
21
+ /** Zero while judging is still running, even though the request answers 200. */
22
+ function countedEvaluations(payload) {
23
+ const value = valueAt(payload, 'evaluation_summary', 'total_evaluations');
24
+ return typeof value === 'number' ? value : 0;
25
+ }
26
+ export async function getResults(context, input) {
27
+ const evaluationId = requireUuid(input.evaluationId, 'Evaluation ID');
28
+ const runId = requireUuid(input.runId, 'Run ID');
29
+ const version = await resolveEvaluationVersion(context, evaluationId, undefined, input.signal);
30
+ const readiness = {
31
+ clock: input.clock,
32
+ intervalMs: input.resultReadyIntervalMs,
33
+ timeoutMs: input.resultReadyTimeoutMs,
34
+ ...(input.signal ? { signal: input.signal } : {}),
35
+ };
36
+ if (version.contextType === 'scenario') {
37
+ const [modelPerformance, scenarioComparison, modelResponses] = await Promise.all([
38
+ // Scored models appear only once judging finishes; until then the
39
+ // endpoint answers 200 with an empty list.
40
+ loadWhenReady(() => context.api.getModelPerformance(evaluationId, runId, input.signal), readiness, (payload) => nonEmptyArrayAt(payload, 'models')),
41
+ loadWhenReady(() => context.api.getScenarioComparison(evaluationId, runId, input.signal), readiness, (payload) => countedEvaluations(payload) > 0),
42
+ loadWhenReady(() => context.api.getModelResponses(evaluationId, runId, input.signal), readiness),
43
+ ]);
44
+ return { contextType: 'scenario', modelPerformance, scenarioComparison, modelResponses };
45
+ }
46
+ if (version.contextType === 'conversation') {
47
+ const conversation = await loadWhenReady(() => context.api.getConversationResults(evaluationId, runId, input.signal), readiness);
48
+ return { contextType: 'conversation', conversation };
49
+ }
50
+ const agentTrace = await loadWhenReady(() => context.api.getAgentTraceResults(evaluationId, runId, input.signal), readiness);
51
+ let trajectory;
52
+ try {
53
+ trajectory = await loadWhenReady(() => context.api.getTrajectory(evaluationId, runId, input.signal), readiness);
54
+ }
55
+ catch (error) {
56
+ if (!(error instanceof ApiError) || error.status !== 404)
57
+ throw error;
58
+ }
59
+ return {
60
+ contextType: 'agent_trace',
61
+ agentTrace,
62
+ ...(trajectory ? { trajectory } : {}),
63
+ };
64
+ }
65
+ //# sourceMappingURL=results.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"results.js","sourceRoot":"","sources":["../../src/actions/results.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,aAAa,EAAE,MAAM,+BAA+B,CAAC;AAC9D,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAE5C,OAAO,EAAE,wBAAwB,EAAE,MAAM,kBAAkB,CAAC;AAC5D,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AAE9C,SAAS,aAAa,CACpB,IAA+B,EAC/B,KAAwF,EACxF,OAAwC;IAExC,OAAO,aAAa,CAAC,EAAE,IAAI,EAAE,GAAG,KAAK,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;AAC5E,CAAC;AAED,SAAS,OAAO,CAAC,OAAmB,EAAE,GAAG,IAAuB;IAC9D,IAAI,OAAO,GAAY,OAAO,CAAC;IAC/B,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI;YAAE,OAAO,SAAS,CAAC;QACtE,OAAO,GAAI,OAAmC,CAAC,GAAG,CAAC,CAAC;IACtD,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,eAAe,CAAC,OAAmB,EAAE,GAAG,IAAuB;IACtE,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,EAAE,GAAG,IAAI,CAAC,CAAC;IACxC,OAAO,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC;AAClD,CAAC;AAED,gFAAgF;AAChF,SAAS,kBAAkB,CAAC,OAAmB;IAC7C,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,EAAE,oBAAoB,EAAE,mBAAmB,CAAC,CAAC;IAC1E,OAAO,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AAC/C,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,UAAU,CAC9B,OAAsB,EACtB,KAOC;IAED,MAAM,YAAY,GAAG,WAAW,CAAC,KAAK,CAAC,YAAY,EAAE,eAAe,CAAC,CAAC;IACtE,MAAM,KAAK,GAAG,WAAW,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;IACjD,MAAM,OAAO,GAAG,MAAM,wBAAwB,CAAC,OAAO,EAAE,YAAY,EAAE,SAAS,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IAC/F,MAAM,SAAS,GAAG;QAChB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,UAAU,EAAE,KAAK,CAAC,qBAAqB;QACvC,SAAS,EAAE,KAAK,CAAC,oBAAoB;QACrC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClD,CAAC;IAEF,IAAI,OAAO,CAAC,WAAW,KAAK,UAAU,EAAE,CAAC;QACvC,MAAM,CAAC,gBAAgB,EAAE,kBAAkB,EAAE,cAAc,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;YAC/E,kEAAkE;YAClE,2CAA2C;YAC3C,aAAa,CACX,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,YAAY,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EACxE,SAAS,EACT,CAAC,OAAO,EAAE,EAAE,CAAC,eAAe,CAAC,OAAO,EAAE,QAAQ,CAAC,CAChD;YACD,aAAa,CACX,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC,YAAY,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EAC1E,SAAS,EACT,CAAC,OAAO,EAAE,EAAE,CAAC,kBAAkB,CAAC,OAAO,CAAC,GAAG,CAAC,CAC7C;YACD,aAAa,CACX,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,YAAY,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EACtE,SAAS,CACV;SACF,CAAC,CAAC;QACH,OAAO,EAAE,WAAW,EAAE,UAAU,EAAE,gBAAgB,EAAE,kBAAkB,EAAE,cAAc,EAAE,CAAC;IAC3F,CAAC;IAED,IAAI,OAAO,CAAC,WAAW,KAAK,cAAc,EAAE,CAAC;QAC3C,MAAM,YAAY,GAAG,MAAM,aAAa,CACtC,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,sBAAsB,CAAC,YAAY,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EAC3E,SAAS,CACV,CAAC;QACF,OAAO,EAAE,WAAW,EAAE,cAAc,EAAE,YAAY,EAAE,CAAC;IACvD,CAAC;IAED,MAAM,UAAU,GAAG,MAAM,aAAa,CACpC,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,YAAY,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EACzE,SAAS,CACV,CAAC;IACF,IAAI,UAAkC,CAAC;IACvC,IAAI,CAAC;QACH,UAAU,GAAG,MAAM,aAAa,CAC9B,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,YAAY,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EAClE,SAAS,CACV,CAAC;IACJ,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,CAAC,CAAC,KAAK,YAAY,QAAQ,CAAC,IAAI,KAAK,CAAC,MAAM,KAAK,GAAG;YAAE,MAAM,KAAK,CAAC;IACxE,CAAC;IACD,OAAO;QACL,WAAW,EAAE,aAAa;QAC1B,UAAU;QACV,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACtC,CAAC;AACJ,CAAC"}
@@ -0,0 +1,22 @@
1
+ import type { RunExecution, RunHandle, RunStatus } from '../domain/types.js';
2
+ import type { PollClock } from '../polling/clock.js';
3
+ import type { ActionContext } from './context.js';
4
+ export declare function getRunStatus(context: ActionContext, input: {
5
+ evaluationId: string;
6
+ runId: string;
7
+ }, signal?: AbortSignal): Promise<RunStatus>;
8
+ export declare function submitEvaluation(context: ActionContext, input: {
9
+ evaluationId: string;
10
+ signal?: AbortSignal;
11
+ idempotencyKeyFactory?: () => string;
12
+ }): Promise<RunHandle>;
13
+ export declare function runEvaluation(context: ActionContext, input: {
14
+ evaluationId: string;
15
+ clock: PollClock;
16
+ pollIntervalMs: number;
17
+ pollTimeoutMs: number;
18
+ signal?: AbortSignal;
19
+ onStatus?: (status: RunStatus) => void;
20
+ idempotencyKeyFactory?: () => string;
21
+ }): Promise<RunExecution>;
22
+ //# sourceMappingURL=runs.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"runs.d.ts","sourceRoot":"","sources":["../../src/actions/runs.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,YAAY,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AAE7E,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAErD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAIlD,wBAAgB,YAAY,CAC1B,OAAO,EAAE,aAAa,EACtB,KAAK,EAAE;IAAE,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,EAC9C,MAAM,CAAC,EAAE,WAAW,GACnB,OAAO,CAAC,SAAS,CAAC,CAMpB;AAED,wBAAsB,gBAAgB,CACpC,OAAO,EAAE,aAAa,EACtB,KAAK,EAAE;IACL,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,qBAAqB,CAAC,EAAE,MAAM,MAAM,CAAC;CACtC,GACA,OAAO,CAAC,SAAS,CAAC,CAgCpB;AAED,wBAAsB,aAAa,CACjC,OAAO,EAAE,aAAa,EACtB,KAAK,EAAE;IACL,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,SAAS,CAAC;IACjB,cAAc,EAAE,MAAM,CAAC;IACvB,aAAa,EAAE,MAAM,CAAC;IACtB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,QAAQ,CAAC,EAAE,CAAC,MAAM,EAAE,SAAS,KAAK,IAAI,CAAC;IACvC,qBAAqB,CAAC,EAAE,MAAM,MAAM,CAAC;CACtC,GACA,OAAO,CAAC,YAAY,CAAC,CA4BvB"}
@@ -0,0 +1,59 @@
1
+ import { randomUUID } from 'node:crypto';
2
+ import { AutoevalError } from '../errors/autoeval-error.js';
3
+ import { waitForRun } from '../polling/wait-for-run.js';
4
+ import { resolveEvaluationVersion } from './evaluations.js';
5
+ import { requireUuid } from './validation.js';
6
+ export function getRunStatus(context, input, signal) {
7
+ return context.api.getRunStatus(requireUuid(input.evaluationId, 'Evaluation ID'), requireUuid(input.runId, 'Run ID'), signal);
8
+ }
9
+ export async function submitEvaluation(context, input) {
10
+ const version = await resolveEvaluationVersion(context, input.evaluationId, undefined, input.signal);
11
+ const idempotencyKey = (input.idempotencyKeyFactory ?? randomUUID)();
12
+ const runResponse = await context.api.createRun({
13
+ evaluationId: version.evaluationId,
14
+ configVersionId: version.configVersionId,
15
+ idempotencyKey,
16
+ ...(input.signal ? { signal: input.signal } : {}),
17
+ });
18
+ if (runResponse.using.configVersionId !== version.configVersionId ||
19
+ runResponse.using.methodologyVersionId !== version.methodologyVersionId) {
20
+ throw new AutoevalError('Autoeval API created the run with unexpected version identifiers.', {
21
+ kind: 'upstream',
22
+ code: 'MISMATCHED_RUN_VERSION_ID',
23
+ });
24
+ }
25
+ return {
26
+ evaluationId: version.evaluationId,
27
+ runId: runResponse.runId,
28
+ status: runResponse.status,
29
+ configVersionId: runResponse.using.configVersionId,
30
+ methodologyVersionId: runResponse.using.methodologyVersionId,
31
+ };
32
+ }
33
+ export async function runEvaluation(context, input) {
34
+ const run = await submitEvaluation(context, {
35
+ evaluationId: input.evaluationId,
36
+ ...(input.signal ? { signal: input.signal } : {}),
37
+ ...(input.idempotencyKeyFactory ? { idempotencyKeyFactory: input.idempotencyKeyFactory } : {}),
38
+ });
39
+ const outcome = await waitForRun({
40
+ api: context.api,
41
+ clock: input.clock,
42
+ evaluationId: run.evaluationId,
43
+ runId: run.runId,
44
+ intervalMs: input.pollIntervalMs,
45
+ timeoutMs: input.pollTimeoutMs,
46
+ ...(input.signal ? { signal: input.signal } : {}),
47
+ ...(input.onStatus ? { onStatus: input.onStatus } : {}),
48
+ });
49
+ if (outcome.status.state !== 'COMPLETED') {
50
+ const failureCode = outcome.status.failureCode ? ` (${outcome.status.failureCode})` : '';
51
+ throw new AutoevalError(`Run ${run.runId} ended in state ${outcome.status.state}${failureCode}.`, {
52
+ kind: 'run_failed',
53
+ code: outcome.status.failureCode ?? 'RUN_FAILED',
54
+ cause: outcome.status,
55
+ });
56
+ }
57
+ return { run, outcome };
58
+ }
59
+ //# sourceMappingURL=runs.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"runs.js","sourceRoot":"","sources":["../../src/actions/runs.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAGzC,OAAO,EAAE,aAAa,EAAE,MAAM,6BAA6B,CAAC;AAE5D,OAAO,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAC;AAExD,OAAO,EAAE,wBAAwB,EAAE,MAAM,kBAAkB,CAAC;AAC5D,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AAE9C,MAAM,UAAU,YAAY,CAC1B,OAAsB,EACtB,KAA8C,EAC9C,MAAoB;IAEpB,OAAO,OAAO,CAAC,GAAG,CAAC,YAAY,CAC7B,WAAW,CAAC,KAAK,CAAC,YAAY,EAAE,eAAe,CAAC,EAChD,WAAW,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,EAClC,MAAM,CACP,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CACpC,OAAsB,EACtB,KAIC;IAED,MAAM,OAAO,GAAG,MAAM,wBAAwB,CAC5C,OAAO,EACP,KAAK,CAAC,YAAY,EAClB,SAAS,EACT,KAAK,CAAC,MAAM,CACb,CAAC;IACF,MAAM,cAAc,GAAG,CAAC,KAAK,CAAC,qBAAqB,IAAI,UAAU,CAAC,EAAE,CAAC;IACrE,MAAM,WAAW,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC;QAC9C,YAAY,EAAE,OAAO,CAAC,YAAY;QAClC,eAAe,EAAE,OAAO,CAAC,eAAe;QACxC,cAAc;QACd,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClD,CAAC,CAAC;IAEH,IACE,WAAW,CAAC,KAAK,CAAC,eAAe,KAAK,OAAO,CAAC,eAAe;QAC7D,WAAW,CAAC,KAAK,CAAC,oBAAoB,KAAK,OAAO,CAAC,oBAAoB,EACvE,CAAC;QACD,MAAM,IAAI,aAAa,CAAC,mEAAmE,EAAE;YAC3F,IAAI,EAAE,UAAU;YAChB,IAAI,EAAE,2BAA2B;SAClC,CAAC,CAAC;IACL,CAAC;IAED,OAAO;QACL,YAAY,EAAE,OAAO,CAAC,YAAY;QAClC,KAAK,EAAE,WAAW,CAAC,KAAK;QACxB,MAAM,EAAE,WAAW,CAAC,MAAM;QAC1B,eAAe,EAAE,WAAW,CAAC,KAAK,CAAC,eAAe;QAClD,oBAAoB,EAAE,WAAW,CAAC,KAAK,CAAC,oBAAoB;KAC7D,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,OAAsB,EACtB,KAQC;IAED,MAAM,GAAG,GAAG,MAAM,gBAAgB,CAAC,OAAO,EAAE;QAC1C,YAAY,EAAE,KAAK,CAAC,YAAY;QAChC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACjD,GAAG,CAAC,KAAK,CAAC,qBAAqB,CAAC,CAAC,CAAC,EAAE,qBAAqB,EAAE,KAAK,CAAC,qBAAqB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC/F,CAAC,CAAC;IACH,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC;QAC/B,GAAG,EAAE,OAAO,CAAC,GAAG;QAChB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,YAAY,EAAE,GAAG,CAAC,YAAY;QAC9B,KAAK,EAAE,GAAG,CAAC,KAAK;QAChB,UAAU,EAAE,KAAK,CAAC,cAAc;QAChC,SAAS,EAAE,KAAK,CAAC,aAAa;QAC9B,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACjD,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACxD,CAAC,CAAC;IACH,IAAI,OAAO,CAAC,MAAM,CAAC,KAAK,KAAK,WAAW,EAAE,CAAC;QACzC,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,KAAK,OAAO,CAAC,MAAM,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QACzF,MAAM,IAAI,aAAa,CACrB,OAAO,GAAG,CAAC,KAAK,mBAAmB,OAAO,CAAC,MAAM,CAAC,KAAK,GAAG,WAAW,GAAG,EACxE;YACE,IAAI,EAAE,YAAY;YAClB,IAAI,EAAE,OAAO,CAAC,MAAM,CAAC,WAAW,IAAI,YAAY;YAChD,KAAK,EAAE,OAAO,CAAC,MAAM;SACtB,CACF,CAAC;IACJ,CAAC;IACD,OAAO,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC;AAC1B,CAAC"}