@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 +1,21 @@
1
+ import chalk from 'chalk';
2
+ /**
3
+ * Semantic colour vocabulary for the line-based CLI.
4
+ *
5
+ * Colour carries meaning, never decoration: pass/positive is green, warning is
6
+ * yellow, failure or regression is red, identifiers and metadata are dim, and
7
+ * user input is cyan. Chalk disables itself for pipes, `NO_COLOR`, and
8
+ * non-TTY streams, so the same renderers stay pipe-safe.
9
+ */
10
+ export const paint = {
11
+ pass: (value) => chalk.green(value),
12
+ warn: (value) => chalk.yellow(value),
13
+ fail: (value) => chalk.red(value),
14
+ /** Identifiers and other reference metadata recede behind the content. */
15
+ meta: (value) => chalk.dim(value),
16
+ /** Echoed user input. */
17
+ input: (value) => chalk.cyan(value),
18
+ heading: (value) => chalk.bold(value),
19
+ header: (value) => chalk.dim.bold(value),
20
+ };
21
+ //# sourceMappingURL=colors.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"colors.js","sourceRoot":"","sources":["../../src/output/colors.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,OAAO,CAAC;AAE1B;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,KAAK,GAAG;IACnB,IAAI,EAAE,CAAC,KAAa,EAAU,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC;IACnD,IAAI,EAAE,CAAC,KAAa,EAAU,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC;IACpD,IAAI,EAAE,CAAC,KAAa,EAAU,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC;IACjD,0EAA0E;IAC1E,IAAI,EAAE,CAAC,KAAa,EAAU,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC;IACjD,yBAAyB;IACzB,KAAK,EAAE,CAAC,KAAa,EAAU,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC;IACnD,OAAO,EAAE,CAAC,KAAa,EAAU,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC;IACrD,MAAM,EAAE,CAAC,KAAa,EAAU,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC;CAChD,CAAC"}
@@ -0,0 +1,165 @@
1
+ import type { JsonObject } from '../domain/common.js';
2
+ import type { Evaluation, EvaluationResults, ScenarioResults } from '../domain/types.js';
3
+ /**
4
+ * Deterministic, bounded digests over already-validated Autoeval payloads.
5
+ *
6
+ * These helpers only read fields that are present and omit everything else, so
7
+ * a digest can never assert a detail the backend did not return. They perform
8
+ * no network access and no inference.
9
+ */
10
+ export declare const MAX_DIGEST_CHARS = 4000;
11
+ /**
12
+ * Summarises the validated evaluation configuration — the evaluation's inputs —
13
+ * for display and reasoning. The sections shown depend on the context type, so
14
+ * the consumer sees the material the judge actually scored.
15
+ */
16
+ export declare function summarizeEvaluationConfiguration(evaluation: Evaluation): string;
17
+ type ModelScore = {
18
+ name: string;
19
+ mean: number;
20
+ sampleSize?: number;
21
+ runs?: number;
22
+ isPrimary?: boolean;
23
+ };
24
+ /** Per-model overall means from the model-performance payload. */
25
+ export declare function scenarioModelScores(modelPerformance: JsonObject): ModelScore[];
26
+ /** One metric of one model, with the interval the payload reported. */
27
+ export type ScenarioMetricScore = {
28
+ metric: string;
29
+ mean: number;
30
+ ci95?: number;
31
+ ci95Lower?: number;
32
+ ci95Upper?: number;
33
+ sampleSize?: number;
34
+ insufficientSamples?: boolean;
35
+ };
36
+ export type ScenarioModelMetrics = {
37
+ model: string;
38
+ isPrimary?: boolean;
39
+ runs?: number;
40
+ metrics: ScenarioMetricScore[];
41
+ };
42
+ /**
43
+ * Per-metric means for each scored model. `overall` is an aggregate of the
44
+ * other metrics, so it is excluded here and reported separately as the headline
45
+ * score. Metrics the payload marks without data are dropped, never defaulted.
46
+ */
47
+ export declare function scenarioMetricScores(modelPerformance: JsonObject): ScenarioModelMetrics[];
48
+ export type ScenarioTestCaseScore = {
49
+ name: string;
50
+ index: number;
51
+ scores: {
52
+ model: string;
53
+ mean: number;
54
+ ci95?: number;
55
+ insufficientSamples?: boolean;
56
+ }[];
57
+ };
58
+ /**
59
+ * Per-test-case means in the order the backend indexed them, so repeated runs
60
+ * of the same evaluation always list the cases identically.
61
+ */
62
+ export declare function scenarioTestCaseScores(scenarioComparison: JsonObject): ScenarioTestCaseScore[];
63
+ export type ScenarioHeadline = {
64
+ mean?: number;
65
+ ci95?: number;
66
+ ci95Lower?: number;
67
+ ci95Upper?: number;
68
+ runs?: number;
69
+ testCases?: number;
70
+ primaryModel?: string;
71
+ };
72
+ /** Headline numbers for a scenario run, read straight from the payloads. */
73
+ export declare function scenarioHeadline(modelPerformance: JsonObject, scenarioComparison: JsonObject): ScenarioHeadline;
74
+ export type ScenarioConvergence = {
75
+ headline?: string;
76
+ consistencyTarget?: string;
77
+ observedVariation?: number;
78
+ autoStopRun?: number;
79
+ runsPlanned?: number;
80
+ judgesPerOutput?: number;
81
+ targetMet?: boolean;
82
+ };
83
+ /**
84
+ * The Evaluation Status card's own numbers. Nothing here is derived: when the
85
+ * backend does not report convergence detail the block is omitted entirely,
86
+ * because a variation figure recomputed from a standard deviation would be a
87
+ * different statistic wearing the same label.
88
+ */
89
+ export declare function scenarioConvergence(...payloads: readonly (JsonObject | undefined)[]): ScenarioConvergence;
90
+ /**
91
+ * The reliability figures the evaluation service reports for a multi-run scenario.
92
+ *
93
+ * `multiRun` is decided by the presence of `achievedConsistency`, and the
94
+ * stability verdict compares that reported figure against the reported target
95
+ * with the reported operator. Nothing here is recomputed from standard
96
+ * deviations, and the coarse `consistency_level` enum is deliberately ignored.
97
+ */
98
+ export type ScenarioReliability = {
99
+ multiRun: boolean;
100
+ achievedConsistency?: number;
101
+ targetConsistency?: number;
102
+ targetOperator?: string;
103
+ stable?: boolean;
104
+ runsCompleted?: number;
105
+ runsPlanned?: number;
106
+ stoppedAtRun?: number;
107
+ autoStopTriggered?: boolean;
108
+ maxRunsReached?: boolean;
109
+ };
110
+ /** Reliability detail read from the run status payload, or `multiRun: false`. */
111
+ export declare function scenarioReliability(status: JsonObject | undefined): ScenarioReliability;
112
+ export type ScenarioOutput = {
113
+ name: string;
114
+ index: number;
115
+ input?: string;
116
+ responses: {
117
+ model: string;
118
+ response?: string;
119
+ overallScore?: number;
120
+ runNumber?: number;
121
+ runCount?: number;
122
+ metrics: {
123
+ metric: string;
124
+ score: number;
125
+ }[];
126
+ tokens?: number;
127
+ cost?: number;
128
+ latencyMs?: number;
129
+ }[];
130
+ detailsUrl?: string;
131
+ };
132
+ /** Per-test-case model outputs, shown only when the user asks for them. */
133
+ export declare function scenarioOutputs(modelResponses: JsonObject): ScenarioOutput[];
134
+ /** Summarises validated scenario results for display, memory, and reasoning. */
135
+ export declare function summarizeScenarioResults(results: ScenarioResults): string;
136
+ export type MetricScore = {
137
+ name: string;
138
+ score: number;
139
+ };
140
+ /** Per-metric scores from a conversation or agent-trace session payload. */
141
+ export declare function sessionMetricScores(input: unknown): MetricScore[];
142
+ /** Per-dimension scores from a trajectory payload. */
143
+ export declare function trajectoryMetricScores(input: unknown): MetricScore[];
144
+ /** Every scored area of a result payload, whatever its context type. */
145
+ export declare function resultMetricScores(results: EvaluationResults): MetricScore[];
146
+ /**
147
+ * Deterministic ranking of the lowest-scoring areas, so a "what should I fix
148
+ * first" answer is ordered by measured evidence rather than model preference.
149
+ */
150
+ export declare function weakestAreasLine(results: EvaluationResults): string | undefined;
151
+ /**
152
+ * Builds the single interpretation prompt: the user's own question plus the
153
+ * digests gathered while answering it (evaluation inputs and result outputs).
154
+ * The question is kept so the answering model addresses it instead of being handed data
155
+ * with nothing to answer.
156
+ */
157
+ export declare function buildInterpretationMessage(question: string, digests: readonly string[], maxChars: number): string;
158
+ /**
159
+ * Deterministic answer used when no usable interpretation is returned.
160
+ * It only repeats headline facts already present in the digests, so the user
161
+ * still gets an answer instead of a failure message.
162
+ */
163
+ export declare function fallbackInterpretation(digests: readonly string[]): string | undefined;
164
+ export {};
165
+ //# sourceMappingURL=digest.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"digest.d.ts","sourceRoot":"","sources":["../../src/output/digest.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,KAAK,EAAE,UAAU,EAAE,iBAAiB,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAC;AAGzF;;;;;;GAMG;AAEH,eAAO,MAAM,gBAAgB,OAAQ,CAAC;AAmOtC;;;;GAIG;AACH,wBAAgB,gCAAgC,CAAC,UAAU,EAAE,UAAU,GAAG,MAAM,CAoC/E;AAED,KAAK,UAAU,GAAG;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB,CAAC;AAMF,kEAAkE;AAClE,wBAAgB,mBAAmB,CAAC,gBAAgB,EAAE,UAAU,GAAG,UAAU,EAAE,CAoB9E;AAED,uEAAuE;AACvE,MAAM,MAAM,mBAAmB,GAAG;IAChC,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,mBAAmB,CAAC,EAAE,OAAO,CAAC;CAC/B,CAAC;AAEF,MAAM,MAAM,oBAAoB,GAAG;IACjC,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,mBAAmB,EAAE,CAAC;CAChC,CAAC;AAEF;;;;GAIG;AACH,wBAAgB,oBAAoB,CAAC,gBAAgB,EAAE,UAAU,GAAG,oBAAoB,EAAE,CAwCzF;AAED,MAAM,MAAM,qBAAqB,GAAG;IAClC,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAC;QAAC,mBAAmB,CAAC,EAAE,OAAO,CAAA;KAAE,EAAE,CAAC;CACzF,CAAC;AAEF;;;GAGG;AACH,wBAAgB,sBAAsB,CAAC,kBAAkB,EAAE,UAAU,GAAG,qBAAqB,EAAE,CA4B9F;AAED,MAAM,MAAM,gBAAgB,GAAG;IAC7B,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB,CAAC;AAEF,4EAA4E;AAC5E,wBAAgB,gBAAgB,CAC9B,gBAAgB,EAAE,UAAU,EAC5B,kBAAkB,EAAE,UAAU,GAC7B,gBAAgB,CAuBlB;AAED,MAAM,MAAM,mBAAmB,GAAG;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB,CAAC;AAIF;;;;;GAKG;AACH,wBAAgB,mBAAmB,CACjC,GAAG,QAAQ,EAAE,SAAS,CAAC,UAAU,GAAG,SAAS,CAAC,EAAE,GAC/C,mBAAmB,CAgCrB;AAED;;;;;;;GAOG;AACH,MAAM,MAAM,mBAAmB,GAAG;IAChC,QAAQ,EAAE,OAAO,CAAC;IAClB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,iBAAiB,CAAC,EAAE,OAAO,CAAC;IAC5B,cAAc,CAAC,EAAE,OAAO,CAAC;CAC1B,CAAC;AAWF,iFAAiF;AACjF,wBAAgB,mBAAmB,CAAC,MAAM,EAAE,UAAU,GAAG,SAAS,GAAG,mBAAmB,CA+CvF;AAED,MAAM,MAAM,cAAc,GAAG;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,SAAS,EAAE;QACT,KAAK,EAAE,MAAM,CAAC;QACd,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE;YAAE,MAAM,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,MAAM,CAAA;SAAE,EAAE,CAAC;QAC7C,MAAM,CAAC,EAAE,MAAM,CAAC;QAChB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,SAAS,CAAC,EAAE,MAAM,CAAC;KACpB,EAAE,CAAC;IACJ,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,CAAC;AAIF,2EAA2E;AAC3E,wBAAgB,eAAe,CAAC,cAAc,EAAE,UAAU,GAAG,cAAc,EAAE,CAoE5E;AA2HD,gFAAgF;AAChF,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,eAAe,GAAG,MAAM,CAkBzE;AAED,MAAM,MAAM,WAAW,GAAG;IACxB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;CACf,CAAC;AAIF,4EAA4E;AAC5E,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,OAAO,GAAG,WAAW,EAAE,CAUjE;AAED,sDAAsD;AACtD,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,OAAO,GAAG,WAAW,EAAE,CAUpE;AAED,wEAAwE;AACxE,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,iBAAiB,GAAG,WAAW,EAAE,CAY5E;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,iBAAiB,GAAG,MAAM,GAAG,SAAS,CAO/E;AAED;;;;;GAKG;AACH,wBAAgB,0BAA0B,CACxC,QAAQ,EAAE,MAAM,EAChB,OAAO,EAAE,SAAS,MAAM,EAAE,EAC1B,QAAQ,EAAE,MAAM,GACf,MAAM,CAWR;AAKD;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,OAAO,EAAE,SAAS,MAAM,EAAE,GAAG,MAAM,GAAG,SAAS,CAQrF"}