elasticdash-sdk 0.2.0

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 (349) hide show
  1. package/LICENSE +21 -0
  2. package/README.md +775 -0
  3. package/dist/browser-ui.d.ts +43 -0
  4. package/dist/browser-ui.d.ts.map +1 -0
  5. package/dist/browser-ui.js +246 -0
  6. package/dist/browser-ui.js.map +1 -0
  7. package/dist/capture/event.d.ts +33 -0
  8. package/dist/capture/event.d.ts.map +1 -0
  9. package/dist/capture/event.js +2 -0
  10. package/dist/capture/event.js.map +1 -0
  11. package/dist/capture/index.d.ts +4 -0
  12. package/dist/capture/index.d.ts.map +1 -0
  13. package/dist/capture/index.js +4 -0
  14. package/dist/capture/index.js.map +1 -0
  15. package/dist/capture/recorder.d.ts +24 -0
  16. package/dist/capture/recorder.d.ts.map +1 -0
  17. package/dist/capture/recorder.js +46 -0
  18. package/dist/capture/recorder.js.map +1 -0
  19. package/dist/capture/replay.d.ts +20 -0
  20. package/dist/capture/replay.d.ts.map +1 -0
  21. package/dist/capture/replay.js +47 -0
  22. package/dist/capture/replay.js.map +1 -0
  23. package/dist/ci/api-client.d.ts +38 -0
  24. package/dist/ci/api-client.d.ts.map +1 -0
  25. package/dist/ci/api-client.js +96 -0
  26. package/dist/ci/api-client.js.map +1 -0
  27. package/dist/ci/benchmark.d.ts +33 -0
  28. package/dist/ci/benchmark.d.ts.map +1 -0
  29. package/dist/ci/benchmark.js +213 -0
  30. package/dist/ci/benchmark.js.map +1 -0
  31. package/dist/ci/ed-runner.d.ts +48 -0
  32. package/dist/ci/ed-runner.d.ts.map +1 -0
  33. package/dist/ci/ed-runner.js +260 -0
  34. package/dist/ci/ed-runner.js.map +1 -0
  35. package/dist/ci/executor.d.ts +13 -0
  36. package/dist/ci/executor.d.ts.map +1 -0
  37. package/dist/ci/executor.js +542 -0
  38. package/dist/ci/executor.js.map +1 -0
  39. package/dist/ci/git-info.d.ts +17 -0
  40. package/dist/ci/git-info.d.ts.map +1 -0
  41. package/dist/ci/git-info.js +102 -0
  42. package/dist/ci/git-info.js.map +1 -0
  43. package/dist/ci/index.d.ts +6 -0
  44. package/dist/ci/index.d.ts.map +1 -0
  45. package/dist/ci/index.js +4 -0
  46. package/dist/ci/index.js.map +1 -0
  47. package/dist/ci/measurement.d.ts +9 -0
  48. package/dist/ci/measurement.d.ts.map +1 -0
  49. package/dist/ci/measurement.js +15 -0
  50. package/dist/ci/measurement.js.map +1 -0
  51. package/dist/ci/replay.d.ts +31 -0
  52. package/dist/ci/replay.d.ts.map +1 -0
  53. package/dist/ci/replay.js +96 -0
  54. package/dist/ci/replay.js.map +1 -0
  55. package/dist/ci/reporters/default.d.ts +8 -0
  56. package/dist/ci/reporters/default.d.ts.map +1 -0
  57. package/dist/ci/reporters/default.js +46 -0
  58. package/dist/ci/reporters/default.js.map +1 -0
  59. package/dist/ci/reporters/index.d.ts +8 -0
  60. package/dist/ci/reporters/index.d.ts.map +1 -0
  61. package/dist/ci/reporters/index.js +14 -0
  62. package/dist/ci/reporters/index.js.map +1 -0
  63. package/dist/ci/reporters/json.d.ts +8 -0
  64. package/dist/ci/reporters/json.d.ts.map +1 -0
  65. package/dist/ci/reporters/json.js +14 -0
  66. package/dist/ci/reporters/json.js.map +1 -0
  67. package/dist/ci/reporters/junit.d.ts +8 -0
  68. package/dist/ci/reporters/junit.d.ts.map +1 -0
  69. package/dist/ci/reporters/junit.js +48 -0
  70. package/dist/ci/reporters/junit.js.map +1 -0
  71. package/dist/ci/runner.d.ts +3 -0
  72. package/dist/ci/runner.d.ts.map +1 -0
  73. package/dist/ci/runner.js +187 -0
  74. package/dist/ci/runner.js.map +1 -0
  75. package/dist/ci/test-discovery.d.ts +5 -0
  76. package/dist/ci/test-discovery.d.ts.map +1 -0
  77. package/dist/ci/test-discovery.js +11 -0
  78. package/dist/ci/test-discovery.js.map +1 -0
  79. package/dist/ci/test-loader.d.ts +19 -0
  80. package/dist/ci/test-loader.d.ts.map +1 -0
  81. package/dist/ci/test-loader.js +149 -0
  82. package/dist/ci/test-loader.js.map +1 -0
  83. package/dist/ci/test-registry.d.ts +42 -0
  84. package/dist/ci/test-registry.d.ts.map +1 -0
  85. package/dist/ci/test-registry.js +18 -0
  86. package/dist/ci/test-registry.js.map +1 -0
  87. package/dist/ci/trace-schema.d.ts +30 -0
  88. package/dist/ci/trace-schema.d.ts.map +1 -0
  89. package/dist/ci/trace-schema.js +66 -0
  90. package/dist/ci/trace-schema.js.map +1 -0
  91. package/dist/ci/trace-writer.d.ts +16 -0
  92. package/dist/ci/trace-writer.d.ts.map +1 -0
  93. package/dist/ci/trace-writer.js +108 -0
  94. package/dist/ci/trace-writer.js.map +1 -0
  95. package/dist/ci/types.d.ts +108 -0
  96. package/dist/ci/types.d.ts.map +1 -0
  97. package/dist/ci/types.js +3 -0
  98. package/dist/ci/types.js.map +1 -0
  99. package/dist/ci/upload-client.d.ts +74 -0
  100. package/dist/ci/upload-client.d.ts.map +1 -0
  101. package/dist/ci/upload-client.js +195 -0
  102. package/dist/ci/upload-client.js.map +1 -0
  103. package/dist/cli.d.ts +3 -0
  104. package/dist/cli.d.ts.map +1 -0
  105. package/dist/cli.js +716 -0
  106. package/dist/cli.js.map +1 -0
  107. package/dist/core/agent-state.d.ts +47 -0
  108. package/dist/core/agent-state.d.ts.map +1 -0
  109. package/dist/core/agent-state.js +137 -0
  110. package/dist/core/agent-state.js.map +1 -0
  111. package/dist/core/judge-utils.d.ts +22 -0
  112. package/dist/core/judge-utils.d.ts.map +1 -0
  113. package/dist/core/judge-utils.js +211 -0
  114. package/dist/core/judge-utils.js.map +1 -0
  115. package/dist/core/registry.d.ts +28 -0
  116. package/dist/core/registry.d.ts.map +1 -0
  117. package/dist/core/registry.js +52 -0
  118. package/dist/core/registry.js.map +1 -0
  119. package/dist/dashboard-server.d.ts +65 -0
  120. package/dist/dashboard-server.d.ts.map +1 -0
  121. package/dist/dashboard-server.js +3940 -0
  122. package/dist/dashboard-server.js.map +1 -0
  123. package/dist/execution/tool-runner.d.ts +26 -0
  124. package/dist/execution/tool-runner.d.ts.map +1 -0
  125. package/dist/execution/tool-runner.js +316 -0
  126. package/dist/execution/tool-runner.js.map +1 -0
  127. package/dist/html/dashboard.html +2218 -0
  128. package/dist/http.d.ts +14 -0
  129. package/dist/http.d.ts.map +1 -0
  130. package/dist/http.js +13 -0
  131. package/dist/http.js.map +1 -0
  132. package/dist/index.cjs +8102 -0
  133. package/dist/index.d.ts +61 -0
  134. package/dist/index.d.ts.map +1 -0
  135. package/dist/index.js +67 -0
  136. package/dist/index.js.map +1 -0
  137. package/dist/interceptors/ai-interceptor.d.ts +26 -0
  138. package/dist/interceptors/ai-interceptor.d.ts.map +1 -0
  139. package/dist/interceptors/ai-interceptor.js +756 -0
  140. package/dist/interceptors/ai-interceptor.js.map +1 -0
  141. package/dist/interceptors/db-auto.d.ts +8 -0
  142. package/dist/interceptors/db-auto.d.ts.map +1 -0
  143. package/dist/interceptors/db-auto.js +217 -0
  144. package/dist/interceptors/db-auto.js.map +1 -0
  145. package/dist/interceptors/db.d.ts +23 -0
  146. package/dist/interceptors/db.d.ts.map +1 -0
  147. package/dist/interceptors/db.js +137 -0
  148. package/dist/interceptors/db.js.map +1 -0
  149. package/dist/interceptors/http.d.ts +28 -0
  150. package/dist/interceptors/http.d.ts.map +1 -0
  151. package/dist/interceptors/http.js +356 -0
  152. package/dist/interceptors/http.js.map +1 -0
  153. package/dist/interceptors/side-effects.d.ts +7 -0
  154. package/dist/interceptors/side-effects.d.ts.map +1 -0
  155. package/dist/interceptors/side-effects.js +72 -0
  156. package/dist/interceptors/side-effects.js.map +1 -0
  157. package/dist/interceptors/telemetry-push.d.ts +142 -0
  158. package/dist/interceptors/telemetry-push.d.ts.map +1 -0
  159. package/dist/interceptors/telemetry-push.js +463 -0
  160. package/dist/interceptors/telemetry-push.js.map +1 -0
  161. package/dist/interceptors/tool.d.ts +2 -0
  162. package/dist/interceptors/tool.d.ts.map +1 -0
  163. package/dist/interceptors/tool.js +274 -0
  164. package/dist/interceptors/tool.js.map +1 -0
  165. package/dist/interceptors/workflow-ai.d.ts +5 -0
  166. package/dist/interceptors/workflow-ai.d.ts.map +1 -0
  167. package/dist/interceptors/workflow-ai.js +382 -0
  168. package/dist/interceptors/workflow-ai.js.map +1 -0
  169. package/dist/internals/conditional-recorder.d.ts +21 -0
  170. package/dist/internals/conditional-recorder.d.ts.map +1 -0
  171. package/dist/internals/conditional-recorder.js +54 -0
  172. package/dist/internals/conditional-recorder.js.map +1 -0
  173. package/dist/internals/mock-resolver.d.ts +146 -0
  174. package/dist/internals/mock-resolver.d.ts.map +1 -0
  175. package/dist/internals/mock-resolver.js +427 -0
  176. package/dist/internals/mock-resolver.js.map +1 -0
  177. package/dist/matchers/index.d.ts +96 -0
  178. package/dist/matchers/index.d.ts.map +1 -0
  179. package/dist/matchers/index.js +668 -0
  180. package/dist/matchers/index.js.map +1 -0
  181. package/dist/observability.d.ts +82 -0
  182. package/dist/observability.d.ts.map +1 -0
  183. package/dist/observability.js +471 -0
  184. package/dist/observability.js.map +1 -0
  185. package/dist/portal-executor.d.ts +30 -0
  186. package/dist/portal-executor.d.ts.map +1 -0
  187. package/dist/portal-executor.js +324 -0
  188. package/dist/portal-executor.js.map +1 -0
  189. package/dist/portal-server.d.ts +3 -0
  190. package/dist/portal-server.d.ts.map +1 -0
  191. package/dist/portal-server.js +279 -0
  192. package/dist/portal-server.js.map +1 -0
  193. package/dist/proxy/llm-capture.d.ts +14 -0
  194. package/dist/proxy/llm-capture.d.ts.map +1 -0
  195. package/dist/proxy/llm-capture.js +264 -0
  196. package/dist/proxy/llm-capture.js.map +1 -0
  197. package/dist/reporter.d.ts +3 -0
  198. package/dist/reporter.d.ts.map +1 -0
  199. package/dist/reporter.js +72 -0
  200. package/dist/reporter.js.map +1 -0
  201. package/dist/runWorkflowSubprocess.d.ts +14 -0
  202. package/dist/runWorkflowSubprocess.d.ts.map +1 -0
  203. package/dist/runWorkflowSubprocess.js +66 -0
  204. package/dist/runWorkflowSubprocess.js.map +1 -0
  205. package/dist/runner.d.ts +16 -0
  206. package/dist/runner.d.ts.map +1 -0
  207. package/dist/runner.js +138 -0
  208. package/dist/runner.js.map +1 -0
  209. package/dist/socket-connector.d.ts +22 -0
  210. package/dist/socket-connector.d.ts.map +1 -0
  211. package/dist/socket-connector.js +104 -0
  212. package/dist/socket-connector.js.map +1 -0
  213. package/dist/telemetry-batcher.d.ts +56 -0
  214. package/dist/telemetry-batcher.d.ts.map +1 -0
  215. package/dist/telemetry-batcher.js +143 -0
  216. package/dist/telemetry-batcher.js.map +1 -0
  217. package/dist/test-setup.d.ts +12 -0
  218. package/dist/test-setup.d.ts.map +1 -0
  219. package/dist/test-setup.js +13 -0
  220. package/dist/test-setup.js.map +1 -0
  221. package/dist/tool-registry.d.ts +31 -0
  222. package/dist/tool-registry.d.ts.map +1 -0
  223. package/dist/tool-registry.js +73 -0
  224. package/dist/tool-registry.js.map +1 -0
  225. package/dist/tool-runner-worker.d.ts +2 -0
  226. package/dist/tool-runner-worker.d.ts.map +1 -0
  227. package/dist/tool-runner-worker.js +215 -0
  228. package/dist/tool-runner-worker.js.map +1 -0
  229. package/dist/trace-adapter/context.d.ts +72 -0
  230. package/dist/trace-adapter/context.d.ts.map +1 -0
  231. package/dist/trace-adapter/context.js +80 -0
  232. package/dist/trace-adapter/context.js.map +1 -0
  233. package/dist/tracing.d.ts +2 -0
  234. package/dist/tracing.d.ts.map +1 -0
  235. package/dist/tracing.js +59 -0
  236. package/dist/tracing.js.map +1 -0
  237. package/dist/trigger-executor.d.ts +12 -0
  238. package/dist/trigger-executor.d.ts.map +1 -0
  239. package/dist/trigger-executor.js +130 -0
  240. package/dist/trigger-executor.js.map +1 -0
  241. package/dist/types/portal.d.ts +76 -0
  242. package/dist/types/portal.d.ts.map +1 -0
  243. package/dist/types/portal.js +2 -0
  244. package/dist/types/portal.js.map +1 -0
  245. package/dist/utils/debug.d.ts +3 -0
  246. package/dist/utils/debug.d.ts.map +1 -0
  247. package/dist/utils/debug.js +8 -0
  248. package/dist/utils/debug.js.map +1 -0
  249. package/dist/utils/license-error.d.ts +23 -0
  250. package/dist/utils/license-error.d.ts.map +1 -0
  251. package/dist/utils/license-error.js +42 -0
  252. package/dist/utils/license-error.js.map +1 -0
  253. package/dist/utils/redact.d.ts +7 -0
  254. package/dist/utils/redact.d.ts.map +1 -0
  255. package/dist/utils/redact.js +26 -0
  256. package/dist/utils/redact.js.map +1 -0
  257. package/dist/workflow-runner-worker.d.ts +2 -0
  258. package/dist/workflow-runner-worker.d.ts.map +1 -0
  259. package/dist/workflow-runner-worker.js +329 -0
  260. package/dist/workflow-runner-worker.js.map +1 -0
  261. package/dist/workflow-runner.d.ts +14 -0
  262. package/dist/workflow-runner.d.ts.map +1 -0
  263. package/dist/workflow-runner.js +34 -0
  264. package/dist/workflow-runner.js.map +1 -0
  265. package/docs/agent-coding-instructions.md +138 -0
  266. package/docs/agent-integration-guide.md +564 -0
  267. package/docs/agents.md +140 -0
  268. package/docs/dashboard.md +394 -0
  269. package/docs/deno.md +69 -0
  270. package/docs/instrumentation.md +424 -0
  271. package/docs/langfuse-trace-structure.md +145 -0
  272. package/docs/matchers.md +173 -0
  273. package/docs/observability_contract.md +192 -0
  274. package/docs/observability_mode.md +195 -0
  275. package/docs/quickstart.md +621 -0
  276. package/docs/security-compliance.md +566 -0
  277. package/docs/test-writing-guidelines.md +444 -0
  278. package/docs/tools.md +165 -0
  279. package/docs/workflow-modes.md +253 -0
  280. package/package.json +76 -0
  281. package/src/browser-ui.ts +281 -0
  282. package/src/capture/event.ts +30 -0
  283. package/src/capture/index.ts +3 -0
  284. package/src/capture/recorder.ts +62 -0
  285. package/src/capture/replay.ts +55 -0
  286. package/src/ci/api-client.ts +136 -0
  287. package/src/ci/benchmark.ts +257 -0
  288. package/src/ci/ed-runner.ts +351 -0
  289. package/src/ci/executor.ts +671 -0
  290. package/src/ci/git-info.ts +127 -0
  291. package/src/ci/index.ts +5 -0
  292. package/src/ci/measurement.ts +25 -0
  293. package/src/ci/replay.ts +127 -0
  294. package/src/ci/reporters/default.ts +50 -0
  295. package/src/ci/reporters/index.ts +21 -0
  296. package/src/ci/reporters/json.ts +18 -0
  297. package/src/ci/reporters/junit.ts +61 -0
  298. package/src/ci/runner.ts +208 -0
  299. package/src/ci/test-discovery.ts +16 -0
  300. package/src/ci/test-loader.ts +187 -0
  301. package/src/ci/test-registry.ts +62 -0
  302. package/src/ci/trace-schema.ts +96 -0
  303. package/src/ci/trace-writer.ts +107 -0
  304. package/src/ci/types.ts +115 -0
  305. package/src/ci/upload-client.ts +300 -0
  306. package/src/cli.ts +811 -0
  307. package/src/core/agent-state.ts +162 -0
  308. package/src/core/judge-utils.ts +232 -0
  309. package/src/core/registry.ts +92 -0
  310. package/src/dashboard-server.ts +2047 -0
  311. package/src/execution/tool-runner.ts +352 -0
  312. package/src/html/dashboard.html +2218 -0
  313. package/src/http.ts +13 -0
  314. package/src/index.ts +138 -0
  315. package/src/interceptors/ai-interceptor.ts +798 -0
  316. package/src/interceptors/db-auto.ts +243 -0
  317. package/src/interceptors/db.ts +156 -0
  318. package/src/interceptors/http.ts +393 -0
  319. package/src/interceptors/side-effects.ts +83 -0
  320. package/src/interceptors/telemetry-push.ts +537 -0
  321. package/src/interceptors/tool.ts +287 -0
  322. package/src/interceptors/workflow-ai.ts +419 -0
  323. package/src/internals/conditional-recorder.ts +63 -0
  324. package/src/internals/mock-resolver.ts +492 -0
  325. package/src/matchers/index.ts +824 -0
  326. package/src/observability.ts +501 -0
  327. package/src/portal-executor.ts +355 -0
  328. package/src/portal-server.ts +304 -0
  329. package/src/proxy/llm-capture.ts +301 -0
  330. package/src/reporter.ts +81 -0
  331. package/src/runWorkflowSubprocess.ts +74 -0
  332. package/src/runner.ts +178 -0
  333. package/src/socket-connector.ts +117 -0
  334. package/src/telemetry-batcher.ts +191 -0
  335. package/src/test-setup.ts +16 -0
  336. package/src/tool-registry.ts +94 -0
  337. package/src/tool-runner-worker.ts +244 -0
  338. package/src/trace-adapter/context.ts +156 -0
  339. package/src/tracing.ts +62 -0
  340. package/src/trigger-executor.ts +171 -0
  341. package/src/types/agent.d.ts +63 -0
  342. package/src/types/expect.d.ts +81 -0
  343. package/src/types/modules.d.ts +2 -0
  344. package/src/types/portal.ts +69 -0
  345. package/src/utils/debug.ts +8 -0
  346. package/src/utils/license-error.ts +43 -0
  347. package/src/utils/redact.ts +25 -0
  348. package/src/workflow-runner-worker.ts +386 -0
  349. package/src/workflow-runner.ts +58 -0
@@ -0,0 +1,300 @@
1
+ import { randomUUID } from 'node:crypto'
2
+ import { mkdir, writeFile } from 'node:fs/promises'
3
+ import { join } from 'node:path'
4
+ import { getOriginalFetch } from '../interceptors/http.js'
5
+ import { notifyLicenseError } from '../utils/license-error.js'
6
+ import { detectGitInfo } from './git-info.js'
7
+ import type { EdTestRunResult, EdTestResult, EdSingleRunResult } from './ed-runner.js'
8
+ import type { BenchmarkResult } from './benchmark.js'
9
+
10
+ // ─── Upload payload types (aligned with POST /api/v1/test-runs/create) ──
11
+
12
+ export interface UploadPayload {
13
+ schema_version: 1
14
+ run: {
15
+ run_id: string
16
+ started_at: string
17
+ finished_at: string
18
+ sdk_version: string
19
+ git: {
20
+ commit_sha: string
21
+ branch: string
22
+ base_branch?: string
23
+ pr_number?: number
24
+ repo?: string
25
+ }
26
+ ci: {
27
+ provider?: string
28
+ run_url?: string
29
+ }
30
+ }
31
+ results: UploadTestResult[]
32
+ }
33
+
34
+ export interface UploadTestResult {
35
+ test_id: string
36
+ test_name: string
37
+ status: 'pass' | 'fail'
38
+ failure_reason?: string
39
+ trace_ref?: string
40
+ target?: { type: string; step_id: string }
41
+ duration_ms: number
42
+ reruns_count: number
43
+ input?: unknown
44
+ output?: unknown
45
+ metrics: UploadMetric[]
46
+ reruns?: UploadRerun[]
47
+ }
48
+
49
+ export interface UploadMetric {
50
+ name: 'duration_ms' | 'tokens_total' | 'tokens_input' | 'tokens_output' | 'output_contains' | 'output_not_contains' | 'llm_judge'
51
+ value: number
52
+ passed: boolean
53
+ benchmark?: { type: string; threshold: number }
54
+ /** Per-rerun measurements (when reruns > 1) */
55
+ measurements?: { rerun_index: number; value: number; passed: boolean }[]
56
+ }
57
+
58
+ export interface UploadRerun {
59
+ rerun_index: number
60
+ input?: unknown
61
+ output?: unknown
62
+ started_at?: string
63
+ finished_at?: string
64
+ }
65
+
66
+ // ─── Payload assembly ───────────────────────────────────────
67
+
68
+ export function buildUploadPayload(runResult: EdTestRunResult): UploadPayload {
69
+ const gitInfo = detectGitInfo()
70
+
71
+ return {
72
+ schema_version: 1,
73
+ run: {
74
+ run_id: runResult.runId,
75
+ started_at: runResult.startedAt,
76
+ finished_at: runResult.finishedAt,
77
+ sdk_version: runResult.sdkVersion,
78
+ git: {
79
+ commit_sha: gitInfo.commit || 'unknown',
80
+ branch: gitInfo.branch || 'unknown',
81
+ base_branch: gitInfo.baseBranch,
82
+ pr_number: gitInfo.prNumber,
83
+ repo: gitInfo.repo,
84
+ },
85
+ ci: {
86
+ provider: gitInfo.ciProvider,
87
+ run_url: gitInfo.ciRunUrl,
88
+ },
89
+ },
90
+ results: runResult.results.map(r => mapTestResult(r)),
91
+ }
92
+ }
93
+
94
+ function buildMetricsFromRun(run: EdSingleRunResult, benchmarkResult?: BenchmarkResult | null): UploadMetric[] {
95
+ const metrics: UploadMetric[] = []
96
+ const br = run.benchmarkResult ?? benchmarkResult
97
+
98
+ if (br) {
99
+ for (const m of br.metrics) {
100
+ metrics.push({
101
+ name: m.name,
102
+ value: m.value,
103
+ passed: m.passed,
104
+ benchmark: { type: 'max', threshold: m.threshold },
105
+ })
106
+ }
107
+ }
108
+
109
+ if (run.measurement) {
110
+ if (!metrics.some(m => m.name === 'duration_ms')) {
111
+ metrics.push({ name: 'duration_ms', value: run.measurement.duration_ms, passed: true })
112
+ }
113
+ if (run.measurement.tokens_input !== undefined && !metrics.some(m => m.name === 'tokens_input')) {
114
+ metrics.push({ name: 'tokens_input', value: run.measurement.tokens_input, passed: true })
115
+ }
116
+ if (run.measurement.tokens_output !== undefined && !metrics.some(m => m.name === 'tokens_output')) {
117
+ metrics.push({ name: 'tokens_output', value: run.measurement.tokens_output, passed: true })
118
+ }
119
+ if (run.measurement.tokens_total !== undefined && !metrics.some(m => m.name === 'tokens_total')) {
120
+ metrics.push({ name: 'tokens_total', value: run.measurement.tokens_total, passed: true })
121
+ }
122
+ }
123
+
124
+ return metrics
125
+ }
126
+
127
+ function mapTestResult(result: EdTestResult): UploadTestResult {
128
+ const singleRuns = result.singleRuns || []
129
+ const rerunsCount = singleRuns.length || 1
130
+
131
+ // Single run (or fallback) — flat metrics, no per-rerun data
132
+ if (rerunsCount <= 1) {
133
+ const run = singleRuns[0]
134
+ const metrics = run
135
+ ? buildMetricsFromRun(run, result.benchmarkResult)
136
+ : buildMetricsFromRun(result as any, result.benchmarkResult)
137
+
138
+ return {
139
+ test_id: result.testId,
140
+ test_name: result.testName,
141
+ status: result.status,
142
+ failure_reason: result.failureReason,
143
+ trace_ref: result.traceRef,
144
+ target: result.target,
145
+ duration_ms: result.durationMs,
146
+ reruns_count: 1,
147
+ input: result.input,
148
+ output: result.output,
149
+ metrics,
150
+ }
151
+ }
152
+
153
+ // Multiple runs — use measurements[] format with rerun_index
154
+ // Collect all unique metric names across all runs
155
+ const metricNameSet = new Set<string>()
156
+ const perRunMetrics: UploadMetric[][] = singleRuns.map(run => {
157
+ const m = buildMetricsFromRun(run, result.benchmarkResult)
158
+ m.forEach(metric => metricNameSet.add(metric.name))
159
+ return m
160
+ })
161
+
162
+ // Build metrics with measurements[] arrays
163
+ const metrics: UploadMetric[] = []
164
+ for (const name of metricNameSet) {
165
+ const measurements: { rerun_index: number; value: number; passed: boolean }[] = []
166
+ let benchmark: { type: string; threshold: number } | undefined
167
+
168
+ for (let i = 0; i < singleRuns.length; i++) {
169
+ const metric = perRunMetrics[i].find(m => m.name === name)
170
+ if (metric) {
171
+ measurements.push({ rerun_index: i, value: metric.value, passed: metric.passed })
172
+ if (!benchmark && metric.benchmark) benchmark = metric.benchmark
173
+ }
174
+ }
175
+
176
+ // Aggregate: use first run's value as the top-level value
177
+ const firstMeasurement = measurements[0]
178
+ metrics.push({
179
+ name: name as UploadMetric['name'],
180
+ value: firstMeasurement?.value ?? 0,
181
+ passed: measurements.every(m => m.passed),
182
+ benchmark,
183
+ measurements,
184
+ })
185
+ }
186
+
187
+ // Build reruns[] snapshots
188
+ const reruns: UploadRerun[] = singleRuns.map((run, i) => ({
189
+ rerun_index: i,
190
+ output: run.output,
191
+ started_at: run.startedAt,
192
+ finished_at: run.finishedAt,
193
+ }))
194
+
195
+ return {
196
+ test_id: result.testId,
197
+ test_name: result.testName,
198
+ status: result.status,
199
+ failure_reason: result.failureReason,
200
+ trace_ref: result.traceRef,
201
+ target: result.target,
202
+ duration_ms: result.durationMs,
203
+ reruns_count: rerunsCount,
204
+ input: result.input,
205
+ output: result.output,
206
+ metrics,
207
+ reruns,
208
+ }
209
+ }
210
+
211
+ // ─── Upload with retries ────────────────────────────────────
212
+
213
+ export interface UploadOptions {
214
+ serverUrl: string
215
+ apiKey: string
216
+ }
217
+
218
+ export interface UploadResponse {
219
+ // Backend applies snake2Camel conversion, so response arrives in camelCase
220
+ runId: string
221
+ status: 'ingested' | 'already_ingested'
222
+ resultsCount: number
223
+ }
224
+
225
+ export async function uploadResults(
226
+ payload: UploadPayload,
227
+ options: UploadOptions,
228
+ ): Promise<UploadResponse> {
229
+ const url = `${options.serverUrl.replace(/\/+$/, '')}/api/v1/test-runs/create`
230
+ const headers: Record<string, string> = {
231
+ 'Content-Type': 'application/json',
232
+ 'api-key': options.apiKey,
233
+ 'X-Correlation-ID': randomUUID(),
234
+ }
235
+
236
+ const body = JSON.stringify(payload)
237
+ console.log(`[elasticdash] Uploading to ${url}, api-key=${options.apiKey ? options.apiKey.slice(0, 10) + '...' : '(none)'}`)
238
+ const maxRetries = 3
239
+ const backoffMs = [1000, 2000, 4000]
240
+
241
+ let lastError: Error | undefined
242
+
243
+ for (let attempt = 0; attempt <= maxRetries; attempt++) {
244
+ try {
245
+ const res = await getOriginalFetch()(url, { method: 'POST', headers, body })
246
+
247
+ if (res.ok) {
248
+ const json = await res.json() as { success: boolean; result: UploadResponse }
249
+ return json.result
250
+ }
251
+
252
+ const status = res.status
253
+ const text = await res.text().catch(() => '')
254
+
255
+ // Don't retry 4xx — payload is malformed
256
+ if (status >= 400 && status < 500) {
257
+ if (status === 402) {
258
+ notifyLicenseError(status, 'ci-upload')
259
+ throw new Error(
260
+ `Upload failed with 402: no available license for your ElasticDash account. ` +
261
+ `Please go to https://app.elasticdash.com to buy a plan.`
262
+ )
263
+ }
264
+ throw new Error(`Upload failed with ${status}: ${text.substring(0, 200)}`)
265
+ }
266
+
267
+ // 5xx — retry
268
+ lastError = new Error(`Upload failed with ${status}: ${text.substring(0, 200)}`)
269
+ } catch (err) {
270
+ if (err instanceof Error && err.message.startsWith('Upload failed with 4')) {
271
+ throw err // Don't retry 4xx
272
+ }
273
+ lastError = err instanceof Error ? err : new Error(String(err))
274
+ }
275
+
276
+ // Wait before retry (skip on last attempt)
277
+ if (attempt < maxRetries) {
278
+ await new Promise(resolve => setTimeout(resolve, backoffMs[attempt]))
279
+ }
280
+ }
281
+
282
+ throw lastError ?? new Error('Upload failed after retries')
283
+ }
284
+
285
+ // ─── Persist failed uploads ─────────────────────────────────
286
+
287
+ export async function persistFailedUpload(
288
+ payload: UploadPayload,
289
+ error: string,
290
+ cwd?: string,
291
+ ): Promise<void> {
292
+ const dir = join(cwd ?? process.cwd(), '.ed_traces', 'failed_uploads')
293
+ try {
294
+ await mkdir(dir, { recursive: true })
295
+ const filepath = join(dir, `${payload.run.run_id}.json`)
296
+ await writeFile(filepath, JSON.stringify({ payload, error, timestamp: new Date().toISOString() }, null, 2), 'utf-8')
297
+ } catch {
298
+ // Best-effort — don't throw
299
+ }
300
+ }