evalkit-py 0.2.0.dev12__py3-none-any.whl

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 (371) hide show
  1. evalkit/__init__.py +0 -0
  2. evalkit/api/__init__.py +0 -0
  3. evalkit/api/access.py +74 -0
  4. evalkit/api/agents.py +159 -0
  5. evalkit/api/app.py +360 -0
  6. evalkit/api/auth.py +225 -0
  7. evalkit/api/cli_dist.py +10 -0
  8. evalkit/api/datasets.py +169 -0
  9. evalkit/api/deps.py +79 -0
  10. evalkit/api/download.py +15 -0
  11. evalkit/api/evaluators.py +121 -0
  12. evalkit/api/experiments.py +616 -0
  13. evalkit/api/fixtures.py +95 -0
  14. evalkit/api/harness_check.py +48 -0
  15. evalkit/api/imports.py +180 -0
  16. evalkit/api/limits.py +84 -0
  17. evalkit/api/llm_providers.py +120 -0
  18. evalkit/api/mcp/__init__.py +0 -0
  19. evalkit/api/mcp/auth.py +100 -0
  20. evalkit/api/mcp/descriptions.py +336 -0
  21. evalkit/api/mcp/models.py +259 -0
  22. evalkit/api/mcp/server.py +101 -0
  23. evalkit/api/mcp/tools.py +844 -0
  24. evalkit/api/meta.py +19 -0
  25. evalkit/api/otel.py +350 -0
  26. evalkit/api/pull.py +197 -0
  27. evalkit/api/review.py +271 -0
  28. evalkit/api/runs.py +555 -0
  29. evalkit/api/server_keys.py +44 -0
  30. evalkit/api/tokens.py +51 -0
  31. evalkit/api/trace_imports.py +173 -0
  32. evalkit/api/trace_sources.py +49 -0
  33. evalkit/api/traces.py +196 -0
  34. evalkit/api/users.py +72 -0
  35. evalkit/api/workspaces.py +131 -0
  36. evalkit/bundled/contracts/atif-accepted.json +5 -0
  37. evalkit/bundled/contracts/custom-evaluator-api-v1.json +243 -0
  38. evalkit/bundled/contracts/harness-api-v1.json +938 -0
  39. evalkit/bundled/contracts/span-mapping-v1.json +314 -0
  40. evalkit/bundled/docs/concepts.md +26 -0
  41. evalkit/bundled/docs/config.md +38 -0
  42. evalkit/bundled/docs/harness-api.md +44 -0
  43. evalkit/bundled/docs/install.md +60 -0
  44. evalkit/bundled/docs/login.md +36 -0
  45. evalkit/bundled/docs/mappings.md +136 -0
  46. evalkit/bundled/docs/mcp.md +44 -0
  47. evalkit/bundled/docs/quickstart.md +142 -0
  48. evalkit/bundled/docs/sdk-runs.md +171 -0
  49. evalkit/bundled/docs/tracing.md +171 -0
  50. evalkit/bundled/docs/try-locally.md +288 -0
  51. evalkit/bundled/docs/workspaces.md +45 -0
  52. evalkit/bundled/skills/authoring-eval-cases/SKILL.md +387 -0
  53. evalkit/bundled/skills/authoring-eval-cases/references/overlays.md +83 -0
  54. evalkit/bundled/skills/authoring-eval-cases/references/tool-checks.md +71 -0
  55. evalkit/bundled/skills/connect-traces/SKILL.md +296 -0
  56. evalkit/bundled/skills/connect-traces/references/gen_ai.json +42 -0
  57. evalkit/bundled/skills/connect-traces/references/openinference.json +35 -0
  58. evalkit/bundled/skills/connect-traces/references/span-mapping-schema.md +88 -0
  59. evalkit/bundled/skills/eval-driven-development/SKILL.md +458 -0
  60. evalkit/bundled/skills/eval-driven-development/references/analyze-run.md +97 -0
  61. evalkit/bundled/skills/eval-driven-development/references/diagnose.md +130 -0
  62. evalkit/bundled/skills/eval-driven-development/references/synthesize-cases.md +87 -0
  63. evalkit/bundled/skills/eval-driven-development/references/synthesize.md +165 -0
  64. evalkit/bundled/skills/evalkit-deployment/SKILL.md +332 -0
  65. evalkit/bundled/skills/evalkit-deployment/config.example.yaml +49 -0
  66. evalkit/bundled/skills/evalkit-deployment/docker-compose.yml +51 -0
  67. evalkit/bundled/skills/evalkit-deployment/references/config-and-deploy.md +333 -0
  68. evalkit/bundled/skills/evalkit-deployment/references/harness-api.md +139 -0
  69. evalkit/bundled/skills/evalkit-deployment/try/.env.example +11 -0
  70. evalkit/bundled/skills/evalkit-deployment/try/agent.py +131 -0
  71. evalkit/bundled/skills/evalkit-deployment/try/cases/acme/balance-sheet/cases.yaml +25 -0
  72. evalkit/bundled/skills/evalkit-deployment/try/cases/acme/income-statement/cases.yaml +43 -0
  73. evalkit/bundled/skills/evalkit-deployment/try/config.yaml +17 -0
  74. evalkit/bundled/skills/evalkit-deployment/try/docker-compose.yml +82 -0
  75. evalkit/bundled/skills/evalkit-deployment/try/evalkit.yaml +4 -0
  76. evalkit/bundled/skills/evalkit-deployment/try/mapping.json +42 -0
  77. evalkit/bundled/skills/evalkit-deployment/try/run.py +33 -0
  78. evalkit/bundled/skills/evalkit-usage/SKILL.md +509 -0
  79. evalkit/bundled/skills/evalkit-usage/references/suite-fixtures.md +42 -0
  80. evalkit/bundled/ui/assets/KaTeX_AMS-Regular-BQhdFMY1.woff2 +0 -0
  81. evalkit/bundled/ui/assets/KaTeX_AMS-Regular-DMm9YOAa.woff +0 -0
  82. evalkit/bundled/ui/assets/KaTeX_AMS-Regular-DRggAlZN.ttf +0 -0
  83. evalkit/bundled/ui/assets/KaTeX_Caligraphic-Bold-ATXxdsX0.ttf +0 -0
  84. evalkit/bundled/ui/assets/KaTeX_Caligraphic-Bold-BEiXGLvX.woff +0 -0
  85. evalkit/bundled/ui/assets/KaTeX_Caligraphic-Bold-Dq_IR9rO.woff2 +0 -0
  86. evalkit/bundled/ui/assets/KaTeX_Caligraphic-Regular-CTRA-rTL.woff +0 -0
  87. evalkit/bundled/ui/assets/KaTeX_Caligraphic-Regular-Di6jR-x-.woff2 +0 -0
  88. evalkit/bundled/ui/assets/KaTeX_Caligraphic-Regular-wX97UBjC.ttf +0 -0
  89. evalkit/bundled/ui/assets/KaTeX_Fraktur-Bold-BdnERNNW.ttf +0 -0
  90. evalkit/bundled/ui/assets/KaTeX_Fraktur-Bold-BsDP51OF.woff +0 -0
  91. evalkit/bundled/ui/assets/KaTeX_Fraktur-Bold-CL6g_b3V.woff2 +0 -0
  92. evalkit/bundled/ui/assets/KaTeX_Fraktur-Regular-CB_wures.ttf +0 -0
  93. evalkit/bundled/ui/assets/KaTeX_Fraktur-Regular-CTYiF6lA.woff2 +0 -0
  94. evalkit/bundled/ui/assets/KaTeX_Fraktur-Regular-Dxdc4cR9.woff +0 -0
  95. evalkit/bundled/ui/assets/KaTeX_Main-Bold-Cx986IdX.woff2 +0 -0
  96. evalkit/bundled/ui/assets/KaTeX_Main-Bold-Jm3AIy58.woff +0 -0
  97. evalkit/bundled/ui/assets/KaTeX_Main-Bold-waoOVXN0.ttf +0 -0
  98. evalkit/bundled/ui/assets/KaTeX_Main-BoldItalic-DxDJ3AOS.woff2 +0 -0
  99. evalkit/bundled/ui/assets/KaTeX_Main-BoldItalic-DzxPMmG6.ttf +0 -0
  100. evalkit/bundled/ui/assets/KaTeX_Main-BoldItalic-SpSLRI95.woff +0 -0
  101. evalkit/bundled/ui/assets/KaTeX_Main-Italic-3WenGoN9.ttf +0 -0
  102. evalkit/bundled/ui/assets/KaTeX_Main-Italic-BMLOBm91.woff +0 -0
  103. evalkit/bundled/ui/assets/KaTeX_Main-Italic-NWA7e6Wa.woff2 +0 -0
  104. evalkit/bundled/ui/assets/KaTeX_Main-Regular-B22Nviop.woff2 +0 -0
  105. evalkit/bundled/ui/assets/KaTeX_Main-Regular-Dr94JaBh.woff +0 -0
  106. evalkit/bundled/ui/assets/KaTeX_Main-Regular-ypZvNtVU.ttf +0 -0
  107. evalkit/bundled/ui/assets/KaTeX_Math-BoldItalic-B3XSjfu4.ttf +0 -0
  108. evalkit/bundled/ui/assets/KaTeX_Math-BoldItalic-CZnvNsCZ.woff2 +0 -0
  109. evalkit/bundled/ui/assets/KaTeX_Math-BoldItalic-iY-2wyZ7.woff +0 -0
  110. evalkit/bundled/ui/assets/KaTeX_Math-Italic-DA0__PXp.woff +0 -0
  111. evalkit/bundled/ui/assets/KaTeX_Math-Italic-flOr_0UB.ttf +0 -0
  112. evalkit/bundled/ui/assets/KaTeX_Math-Italic-t53AETM-.woff2 +0 -0
  113. evalkit/bundled/ui/assets/KaTeX_SansSerif-Bold-CFMepnvq.ttf +0 -0
  114. evalkit/bundled/ui/assets/KaTeX_SansSerif-Bold-D1sUS0GD.woff2 +0 -0
  115. evalkit/bundled/ui/assets/KaTeX_SansSerif-Bold-DbIhKOiC.woff +0 -0
  116. evalkit/bundled/ui/assets/KaTeX_SansSerif-Italic-C3H0VqGB.woff2 +0 -0
  117. evalkit/bundled/ui/assets/KaTeX_SansSerif-Italic-DN2j7dab.woff +0 -0
  118. evalkit/bundled/ui/assets/KaTeX_SansSerif-Italic-YYjJ1zSn.ttf +0 -0
  119. evalkit/bundled/ui/assets/KaTeX_SansSerif-Regular-BNo7hRIc.ttf +0 -0
  120. evalkit/bundled/ui/assets/KaTeX_SansSerif-Regular-CS6fqUqJ.woff +0 -0
  121. evalkit/bundled/ui/assets/KaTeX_SansSerif-Regular-DDBCnlJ7.woff2 +0 -0
  122. evalkit/bundled/ui/assets/KaTeX_Script-Regular-C5JkGWo-.ttf +0 -0
  123. evalkit/bundled/ui/assets/KaTeX_Script-Regular-D3wIWfF6.woff2 +0 -0
  124. evalkit/bundled/ui/assets/KaTeX_Script-Regular-D5yQViql.woff +0 -0
  125. evalkit/bundled/ui/assets/KaTeX_Size1-Regular-C195tn64.woff +0 -0
  126. evalkit/bundled/ui/assets/KaTeX_Size1-Regular-Dbsnue_I.ttf +0 -0
  127. evalkit/bundled/ui/assets/KaTeX_Size1-Regular-mCD8mA8B.woff2 +0 -0
  128. evalkit/bundled/ui/assets/KaTeX_Size2-Regular-B7gKUWhC.ttf +0 -0
  129. evalkit/bundled/ui/assets/KaTeX_Size2-Regular-Dy4dx90m.woff2 +0 -0
  130. evalkit/bundled/ui/assets/KaTeX_Size2-Regular-oD1tc_U0.woff +0 -0
  131. evalkit/bundled/ui/assets/KaTeX_Size3-Regular-CTq5MqoE.woff +0 -0
  132. evalkit/bundled/ui/assets/KaTeX_Size3-Regular-DgpXs0kz.ttf +0 -0
  133. evalkit/bundled/ui/assets/KaTeX_Size4-Regular-BF-4gkZK.woff +0 -0
  134. evalkit/bundled/ui/assets/KaTeX_Size4-Regular-DWFBv043.ttf +0 -0
  135. evalkit/bundled/ui/assets/KaTeX_Size4-Regular-Dl5lxZxV.woff2 +0 -0
  136. evalkit/bundled/ui/assets/KaTeX_Typewriter-Regular-C0xS9mPB.woff +0 -0
  137. evalkit/bundled/ui/assets/KaTeX_Typewriter-Regular-CO6r4hn1.woff2 +0 -0
  138. evalkit/bundled/ui/assets/KaTeX_Typewriter-Regular-D3Ib7_Hf.ttf +0 -0
  139. evalkit/bundled/ui/assets/abap-08VXUWAP.js +1 -0
  140. evalkit/bundled/ui/assets/apex-BWPQTe0t.js +1 -0
  141. evalkit/bundled/ui/assets/azcli-Bc_sGQ0U.js +1 -0
  142. evalkit/bundled/ui/assets/bat-i0X4ZdIN.js +1 -0
  143. evalkit/bundled/ui/assets/bicep-B5-_aFwp.js +2 -0
  144. evalkit/bundled/ui/assets/cameligo-DMUM7wLl.js +1 -0
  145. evalkit/bundled/ui/assets/clojure-Cm7r79vr.js +1 -0
  146. evalkit/bundled/ui/assets/codicon-DF1abBS2.ttf +0 -0
  147. evalkit/bundled/ui/assets/coffee-Ba7i2nA0.js +1 -0
  148. evalkit/bundled/ui/assets/cpp-C7h46wYY.js +1 -0
  149. evalkit/bundled/ui/assets/csharp-BKxtCVv1.js +1 -0
  150. evalkit/bundled/ui/assets/csp-bTuwJoIa.js +1 -0
  151. evalkit/bundled/ui/assets/css-DIMkf-bt.js +3 -0
  152. evalkit/bundled/ui/assets/css.worker-DwfUZ3fB.js +96 -0
  153. evalkit/bundled/ui/assets/cssMode-C97l1pbB.js +1 -0
  154. evalkit/bundled/ui/assets/cypher-CVaqCwHa.js +1 -0
  155. evalkit/bundled/ui/assets/dart-onAF5SnQ.js +1 -0
  156. evalkit/bundled/ui/assets/dockerfile-DZFCIeNp.js +1 -0
  157. evalkit/bundled/ui/assets/ecl-94qehhXe.js +1 -0
  158. evalkit/bundled/ui/assets/editor-DjpsE8to.css +1 -0
  159. evalkit/bundled/ui/assets/editor.api-NTesIxw5.js +908 -0
  160. evalkit/bundled/ui/assets/editor.worker-CRz__AKW.js +33 -0
  161. evalkit/bundled/ui/assets/elixir-6RTg0lbw.js +1 -0
  162. evalkit/bundled/ui/assets/flow9-C5_-GSwl.js +1 -0
  163. evalkit/bundled/ui/assets/freemarker2-BWqEFagp.js +3 -0
  164. evalkit/bundled/ui/assets/fsharp-C8Ef5oNN.js +1 -0
  165. evalkit/bundled/ui/assets/go-C-y9NEjX.js +1 -0
  166. evalkit/bundled/ui/assets/graphql-fmXr3nnJ.js +1 -0
  167. evalkit/bundled/ui/assets/handlebars-DkIF-whN.js +1 -0
  168. evalkit/bundled/ui/assets/hcl-CpzslTdj.js +1 -0
  169. evalkit/bundled/ui/assets/html-C_rpeQXJ.js +1 -0
  170. evalkit/bundled/ui/assets/html.worker-D9ugpX4y.js +509 -0
  171. evalkit/bundled/ui/assets/htmlMode-B0YIq3DF.js +1 -0
  172. evalkit/bundled/ui/assets/index-Cw-jBH6M.js +736 -0
  173. evalkit/bundled/ui/assets/index-DKtoQZtm.css +1 -0
  174. evalkit/bundled/ui/assets/ini-sBoK_t0W.js +1 -0
  175. evalkit/bundled/ui/assets/java-BEtHBSE6.js +1 -0
  176. evalkit/bundled/ui/assets/javascript-D4EGhYA6.js +1 -0
  177. evalkit/bundled/ui/assets/jetbrains-mono-cyrillic-wght-normal-D73BlboJ.woff2 +0 -0
  178. evalkit/bundled/ui/assets/jetbrains-mono-greek-wght-normal-Bw9x6K1M.woff2 +0 -0
  179. evalkit/bundled/ui/assets/jetbrains-mono-latin-ext-wght-normal-DBQx-q_a.woff2 +0 -0
  180. evalkit/bundled/ui/assets/jetbrains-mono-latin-wght-normal-B9CIFXIH.woff2 +0 -0
  181. evalkit/bundled/ui/assets/jetbrains-mono-vietnamese-wght-normal-Bt-aOZkq.woff2 +0 -0
  182. evalkit/bundled/ui/assets/json.worker-CV9fqkUp.js +65 -0
  183. evalkit/bundled/ui/assets/jsonMode-pQjSfIH6.js +7 -0
  184. evalkit/bundled/ui/assets/julia-Bri6UV-V.js +1 -0
  185. evalkit/bundled/ui/assets/kotlin-BOotOW0E.js +1 -0
  186. evalkit/bundled/ui/assets/less-B9JPFI3C.js +2 -0
  187. evalkit/bundled/ui/assets/lexon-CfSJPG6W.js +1 -0
  188. evalkit/bundled/ui/assets/liquid-DbrvoYjD.js +1 -0
  189. evalkit/bundled/ui/assets/lspLanguageFeatures-D0B7cH6y.js +4 -0
  190. evalkit/bundled/ui/assets/lua-CsQS60Ue.js +1 -0
  191. evalkit/bundled/ui/assets/m3-D-oSqn_W.js +1 -0
  192. evalkit/bundled/ui/assets/markdown-Cimd5fb3.js +1 -0
  193. evalkit/bundled/ui/assets/mdx-B0HrujLR.js +1 -0
  194. evalkit/bundled/ui/assets/mips-CIPQ_RoX.js +1 -0
  195. evalkit/bundled/ui/assets/monacoEditor-DsoPncxW.js +68 -0
  196. evalkit/bundled/ui/assets/monacoEditor-n7clI0si.css +1 -0
  197. evalkit/bundled/ui/assets/msdax-DauUninz.js +1 -0
  198. evalkit/bundled/ui/assets/mysql-SOo6toE5.js +1 -0
  199. evalkit/bundled/ui/assets/objective-c-FvmIjYaQ.js +1 -0
  200. evalkit/bundled/ui/assets/pascal-DrH0SRf2.js +1 -0
  201. evalkit/bundled/ui/assets/pascaligo-D-ptJ9y-.js +1 -0
  202. evalkit/bundled/ui/assets/perl-oz_6vUea.js +1 -0
  203. evalkit/bundled/ui/assets/pgsql-DTj74zXo.js +1 -0
  204. evalkit/bundled/ui/assets/php-nr791fC2.js +1 -0
  205. evalkit/bundled/ui/assets/pla-CopQ2nXW.js +1 -0
  206. evalkit/bundled/ui/assets/postiats-43DmfD33.js +1 -0
  207. evalkit/bundled/ui/assets/powerquery-D3hlyOfw.js +1 -0
  208. evalkit/bundled/ui/assets/powershell-DmHpPYUd.js +1 -0
  209. evalkit/bundled/ui/assets/protobuf-C531GsRP.js +2 -0
  210. evalkit/bundled/ui/assets/pug-Z5eAx3Zn.js +1 -0
  211. evalkit/bundled/ui/assets/python-CieHVM9n.js +1 -0
  212. evalkit/bundled/ui/assets/qsharp-DkqhCAOL.js +1 -0
  213. evalkit/bundled/ui/assets/r-BwWrilGY.js +1 -0
  214. evalkit/bundled/ui/assets/razor-sLewSMyp.js +1 -0
  215. evalkit/bundled/ui/assets/redis-ClamHrr6.js +1 -0
  216. evalkit/bundled/ui/assets/redshift-DT7zqm-g.js +1 -0
  217. evalkit/bundled/ui/assets/restructuredtext-BYgofb2h.js +1 -0
  218. evalkit/bundled/ui/assets/ruby-DezsRK8O.js +1 -0
  219. evalkit/bundled/ui/assets/rust-DdL9SqIa.js +1 -0
  220. evalkit/bundled/ui/assets/sb-CcwsVR0C.js +1 -0
  221. evalkit/bundled/ui/assets/scala-DHpiXF5c.js +1 -0
  222. evalkit/bundled/ui/assets/scheme-BeGwcela.js +1 -0
  223. evalkit/bundled/ui/assets/scss-gp-XZpBa.js +3 -0
  224. evalkit/bundled/ui/assets/shell-CC2rA5mh.js +1 -0
  225. evalkit/bundled/ui/assets/solidity-BEEn4gHE.js +1 -0
  226. evalkit/bundled/ui/assets/sophia-CRfGWb83.js +1 -0
  227. evalkit/bundled/ui/assets/sparql-D_Lu-MrJ.js +1 -0
  228. evalkit/bundled/ui/assets/sql-NEE52Syq.js +1 -0
  229. evalkit/bundled/ui/assets/st-DbInun42.js +1 -0
  230. evalkit/bundled/ui/assets/swift-Bxkupp3x.js +1 -0
  231. evalkit/bundled/ui/assets/systemverilog-Bz4Y3fRF.js +1 -0
  232. evalkit/bundled/ui/assets/tcl-DISqw1ZD.js +1 -0
  233. evalkit/bundled/ui/assets/ts.worker-DIAQkoB3.js +67741 -0
  234. evalkit/bundled/ui/assets/tsMode-BcVQ1nu0.js +11 -0
  235. evalkit/bundled/ui/assets/twig-De2hgUGE.js +1 -0
  236. evalkit/bundled/ui/assets/typescript-CUvNWd9p.js +1 -0
  237. evalkit/bundled/ui/assets/typespec-B8J7ngcE.js +1 -0
  238. evalkit/bundled/ui/assets/vb-DV3o63ZY.js +1 -0
  239. evalkit/bundled/ui/assets/wgsl-DpFanUEy.js +298 -0
  240. evalkit/bundled/ui/assets/workers-CwpYQtzp.js +1 -0
  241. evalkit/bundled/ui/assets/xml-C1HnXMdF.js +1 -0
  242. evalkit/bundled/ui/assets/yaml-D7Ey3HQu.js +1 -0
  243. evalkit/bundled/ui/favicon.svg +6 -0
  244. evalkit/bundled/ui/index.html +14 -0
  245. evalkit/cli/__init__.py +0 -0
  246. evalkit/cli/admin_commands.py +590 -0
  247. evalkit/cli/client.py +67 -0
  248. evalkit/cli/init_server.py +165 -0
  249. evalkit/cli/local_run.py +195 -0
  250. evalkit/cli/main.py +1365 -0
  251. evalkit/cli/runs_commands.py +539 -0
  252. evalkit/cli/server_commands.py +130 -0
  253. evalkit/cli/skill_targets.py +43 -0
  254. evalkit/cli/tokens_commands.py +52 -0
  255. evalkit/cli/traces_commands.py +332 -0
  256. evalkit/cli/update.py +85 -0
  257. evalkit/client/__init__.py +0 -0
  258. evalkit/client/agents.py +25 -0
  259. evalkit/client/api.py +193 -0
  260. evalkit/client/blob.py +104 -0
  261. evalkit/client/connection.py +88 -0
  262. evalkit/client/evaluators.py +11 -0
  263. evalkit/client/sessions.py +17 -0
  264. evalkit/client/store.py +139 -0
  265. evalkit/client/summary.py +46 -0
  266. evalkit/client/sync.py +198 -0
  267. evalkit/config.py +227 -0
  268. evalkit/contracts/__init__.py +0 -0
  269. evalkit/contracts/atif.py +115 -0
  270. evalkit/contracts/cases.py +456 -0
  271. evalkit/contracts/custom_evaluator.py +58 -0
  272. evalkit/contracts/enums.py +86 -0
  273. evalkit/contracts/esf.py +120 -0
  274. evalkit/contracts/harness.py +281 -0
  275. evalkit/contracts/notes.json +4 -0
  276. evalkit/contracts/notes.py +7 -0
  277. evalkit/contracts/settings.py +41 -0
  278. evalkit/core/__init__.py +0 -0
  279. evalkit/core/access.py +75 -0
  280. evalkit/core/agents.py +65 -0
  281. evalkit/core/blob.py +46 -0
  282. evalkit/core/capabilities.py +36 -0
  283. evalkit/core/costmap.py +90 -0
  284. evalkit/core/custom_eval.py +50 -0
  285. evalkit/core/data/cost_map.json +81546 -0
  286. evalkit/core/endpoints.py +31 -0
  287. evalkit/core/evaluators/__init__.py +29 -0
  288. evalkit/core/evaluators/base.py +121 -0
  289. evalkit/core/evaluators/gauges.py +84 -0
  290. evalkit/core/evaluators/presentation.py +166 -0
  291. evalkit/core/evaluators/response_time.py +21 -0
  292. evalkit/core/evaluators/rubrics.py +494 -0
  293. evalkit/core/evaluators/tool_trajectory.py +187 -0
  294. evalkit/core/evaluators/trajectory_rubrics.py +198 -0
  295. evalkit/core/fixtures.py +57 -0
  296. evalkit/core/harness.py +227 -0
  297. evalkit/core/harness_check.py +191 -0
  298. evalkit/core/importer.py +257 -0
  299. evalkit/core/judge.py +78 -0
  300. evalkit/core/lint.py +265 -0
  301. evalkit/core/overlays.py +76 -0
  302. evalkit/core/paths.py +25 -0
  303. evalkit/core/promotion.py +94 -0
  304. evalkit/core/pull.py +198 -0
  305. evalkit/core/python_eval.py +59 -0
  306. evalkit/core/remote_evaluators.py +32 -0
  307. evalkit/core/review.py +220 -0
  308. evalkit/core/run_digest.py +421 -0
  309. evalkit/core/runner.py +995 -0
  310. evalkit/core/scorecard.py +255 -0
  311. evalkit/core/traces/__init__.py +0 -0
  312. evalkit/core/traces/convert.py +336 -0
  313. evalkit/core/traces/importer.py +113 -0
  314. evalkit/core/traces/mapping.py +385 -0
  315. evalkit/core/traces/poll.py +38 -0
  316. evalkit/core/traces/raw.py +21 -0
  317. evalkit/core/traces/sessions.py +172 -0
  318. evalkit/core/traces/settle.py +14 -0
  319. evalkit/core/traces/sources/__init__.py +19 -0
  320. evalkit/core/traces/sources/base.py +25 -0
  321. evalkit/core/traces/sources/phoenix.py +121 -0
  322. evalkit/core/trajectory.py +248 -0
  323. evalkit/core/workspace.py +30 -0
  324. evalkit/credentials.py +119 -0
  325. evalkit/db/__init__.py +0 -0
  326. evalkit/db/agents.py +289 -0
  327. evalkit/db/alembic/env.py +90 -0
  328. evalkit/db/alembic/script.py.mako +28 -0
  329. evalkit/db/alembic/versions/0011_baseline.py +1002 -0
  330. evalkit/db/alembic/versions/0012_ticket_token_kind.py +27 -0
  331. evalkit/db/alembic/versions/0013_keys_and_llm_providers.py +67 -0
  332. evalkit/db/cases.py +608 -0
  333. evalkit/db/crypto.py +12 -0
  334. evalkit/db/evaluators.py +156 -0
  335. evalkit/db/experiments.py +363 -0
  336. evalkit/db/fixture_blobs.py +91 -0
  337. evalkit/db/keys.py +78 -0
  338. evalkit/db/llm_providers.py +176 -0
  339. evalkit/db/locks.py +6 -0
  340. evalkit/db/mappings.py +168 -0
  341. evalkit/db/models.py +567 -0
  342. evalkit/db/promotion.py +271 -0
  343. evalkit/db/rekey.py +46 -0
  344. evalkit/db/review.py +250 -0
  345. evalkit/db/review_state.py +50 -0
  346. evalkit/db/runs.py +533 -0
  347. evalkit/db/scorecard.py +300 -0
  348. evalkit/db/session.py +71 -0
  349. evalkit/db/sessions.py +90 -0
  350. evalkit/db/spans.py +360 -0
  351. evalkit/db/store.py +209 -0
  352. evalkit/db/tokens.py +100 -0
  353. evalkit/db/trace_imports.py +308 -0
  354. evalkit/db/trace_sources.py +131 -0
  355. evalkit/db/users.py +186 -0
  356. evalkit/db/workspaces.py +232 -0
  357. evalkit/paths.py +33 -0
  358. evalkit/sdk/__init__.py +29 -0
  359. evalkit/sdk/binding.py +69 -0
  360. evalkit/sdk/evaluators.py +56 -0
  361. evalkit/sdk/runner.py +957 -0
  362. evalkit/sdk/tracing.py +166 -0
  363. evalkit/store/__init__.py +0 -0
  364. evalkit/store/base.py +354 -0
  365. evalkit/store/blob.py +69 -0
  366. evalkit/version.py +44 -0
  367. evalkit_py-0.2.0.dev12.dist-info/METADATA +205 -0
  368. evalkit_py-0.2.0.dev12.dist-info/RECORD +371 -0
  369. evalkit_py-0.2.0.dev12.dist-info/WHEEL +4 -0
  370. evalkit_py-0.2.0.dev12.dist-info/entry_points.txt +2 -0
  371. evalkit_py-0.2.0.dev12.dist-info/licenses/LICENSE +661 -0
evalkit/db/models.py ADDED
@@ -0,0 +1,567 @@
1
+ from datetime import datetime
2
+ from enum import StrEnum
3
+
4
+ from sqlalchemy import (
5
+ BigInteger,
6
+ Boolean,
7
+ CheckConstraint,
8
+ DateTime,
9
+ Float,
10
+ ForeignKey,
11
+ ForeignKeyConstraint,
12
+ Index,
13
+ Integer,
14
+ LargeBinary,
15
+ String,
16
+ Text,
17
+ UniqueConstraint,
18
+ case,
19
+ func,
20
+ text,
21
+ )
22
+ from sqlalchemy.dialects.postgresql import ARRAY, JSONB
23
+ from sqlalchemy.ext.hybrid import hybrid_property
24
+ from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column
25
+ from sqlalchemy.sql import ColumnElement
26
+
27
+ from evalkit.contracts.enums import (
28
+ Bucket,
29
+ DatasetKind,
30
+ EvaluationSource,
31
+ EvaluatorKind,
32
+ ExperimentStatus,
33
+ MappingStatus,
34
+ RunStatus,
35
+ SyncSource,
36
+ TokenKind,
37
+ TraceImportStatus,
38
+ TraceProvider,
39
+ TrajectorySource,
40
+ WorkspaceRole,
41
+ )
42
+ from evalkit.core.review import CaseEventKind
43
+
44
+
45
+ class Base(DeclarativeBase):
46
+ pass
47
+
48
+
49
+ def check_in(column: str, enum: type[StrEnum], name: str) -> CheckConstraint:
50
+ values = ", ".join(f"'{v.value}'" for v in enum)
51
+ return CheckConstraint(f"{column} IN ({values})", name=name)
52
+
53
+
54
+ class AuditMixin:
55
+ created_at: Mapped[datetime] = mapped_column(
56
+ DateTime(timezone=True),
57
+ default=func.now(),
58
+ server_default=func.now(),
59
+ nullable=False,
60
+ )
61
+ updated_at: Mapped[datetime] = mapped_column(
62
+ DateTime(timezone=True),
63
+ default=func.now(),
64
+ server_default=func.now(),
65
+ onupdate=func.now(),
66
+ nullable=False,
67
+ )
68
+
69
+
70
+ class User(AuditMixin, Base):
71
+ __tablename__ = "users"
72
+
73
+ id: Mapped[int] = mapped_column(primary_key=True)
74
+ username: Mapped[str] = mapped_column(String(150), unique=True, index=True)
75
+ password_hash: Mapped[str] = mapped_column(String(200))
76
+ is_admin: Mapped[bool] = mapped_column(Boolean, default=False, server_default=text("false"))
77
+ archived_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
78
+
79
+
80
+ class Workspace(AuditMixin, Base):
81
+ __tablename__ = "workspaces"
82
+ __table_args__ = (UniqueConstraint("name", name="uq_workspaces_name"),)
83
+
84
+ id: Mapped[int] = mapped_column(primary_key=True)
85
+ name: Mapped[str] = mapped_column(String(150))
86
+ owner_user_id: Mapped[int | None] = mapped_column(ForeignKey("users.id"), default=None)
87
+ archived_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
88
+
89
+ @hybrid_property
90
+ def type(self) -> str:
91
+ return "shared" if self.owner_user_id is None else "personal"
92
+
93
+ @type.inplace.expression
94
+ @classmethod
95
+ def _type_expression(cls) -> ColumnElement[str]:
96
+ return case((cls.owner_user_id.is_(None), "shared"), else_="personal")
97
+
98
+
99
+ class WorkspacePromotion(AuditMixin, Base):
100
+ __tablename__ = "workspace_promotions"
101
+ __table_args__ = (UniqueConstraint("target_id", name="uq_workspace_promotions_target"),)
102
+
103
+ source_id: Mapped[int] = mapped_column(ForeignKey("workspaces.id"), primary_key=True)
104
+ target_id: Mapped[int] = mapped_column(ForeignKey("workspaces.id"))
105
+
106
+
107
+ class WorkspaceMember(AuditMixin, Base):
108
+ __tablename__ = "workspace_members"
109
+ __table_args__ = (check_in("role", WorkspaceRole, "ck_workspace_members_role"),)
110
+
111
+ workspace_id: Mapped[int] = mapped_column(ForeignKey("workspaces.id"), primary_key=True)
112
+ user_id: Mapped[int] = mapped_column(ForeignKey("users.id"), primary_key=True)
113
+ role: Mapped[str] = mapped_column(String(16), primary_key=True)
114
+ granted_by: Mapped[int] = mapped_column(ForeignKey("users.id"))
115
+
116
+
117
+ class AuthToken(AuditMixin, Base):
118
+ __tablename__ = "auth_tokens"
119
+ __table_args__ = (
120
+ CheckConstraint("kind <> 'agent' OR agent_id IS NOT NULL", name="ck_auth_tokens_agent"),
121
+ check_in("kind", TokenKind, "ck_auth_tokens_kind"),
122
+ )
123
+
124
+ id: Mapped[int] = mapped_column(primary_key=True)
125
+ user_id: Mapped[int] = mapped_column(ForeignKey("users.id", ondelete="CASCADE"), index=True)
126
+ token_hash: Mapped[str] = mapped_column(String(64), unique=True, index=True)
127
+ expires_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
128
+ kind: Mapped[str] = mapped_column(
129
+ String(8), default=TokenKind.USER, server_default=TokenKind.USER.value, nullable=False
130
+ )
131
+ agent_id: Mapped[int | None] = mapped_column(
132
+ ForeignKey("agents.id", ondelete="CASCADE"), nullable=True, default=None
133
+ )
134
+ last_used_at: Mapped[datetime | None] = mapped_column(
135
+ DateTime(timezone=True), nullable=True, default=None
136
+ )
137
+
138
+
139
+ class Agent(AuditMixin, Base):
140
+ __tablename__ = "agents"
141
+ __table_args__ = (
142
+ UniqueConstraint("workspace_id", "name", name="uq_agents_workspace_name"),
143
+ check_in("trajectory_source", TrajectorySource, "ck_agents_trajectory_source"),
144
+ )
145
+
146
+ id: Mapped[int] = mapped_column(primary_key=True)
147
+ workspace_id: Mapped[int] = mapped_column(ForeignKey("workspaces.id"))
148
+ name: Mapped[str] = mapped_column(String(150))
149
+ harness_url: Mapped[str | None] = mapped_column(String(500), default=None)
150
+ trajectory_source: Mapped[str] = mapped_column(
151
+ String(8), default=TrajectorySource.ATIF, server_default=TrajectorySource.ATIF.value
152
+ )
153
+ repo_url: Mapped[str | None] = mapped_column(String(500), default=None)
154
+ config: Mapped[dict] = mapped_column(JSONB, default=dict, server_default=text("'{}'"))
155
+ api_key_encrypted: Mapped[bytes | None] = mapped_column(LargeBinary, default=None)
156
+ created_by: Mapped[int | None] = mapped_column(ForeignKey("users.id"), default=None)
157
+ archived_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
158
+
159
+
160
+ class RemoteEvaluator(AuditMixin, Base):
161
+ __tablename__ = "remote_evaluators"
162
+ __table_args__ = (
163
+ UniqueConstraint("workspace_id", "name", name="uq_remote_evaluators_workspace_name"),
164
+ check_in("kind", EvaluatorKind, "ck_remote_evaluators_kind"),
165
+ )
166
+
167
+ id: Mapped[int] = mapped_column(primary_key=True)
168
+ workspace_id: Mapped[int] = mapped_column(ForeignKey("workspaces.id"))
169
+ name: Mapped[str] = mapped_column(String(150))
170
+ kind: Mapped[str] = mapped_column(String(8))
171
+ config: Mapped[dict] = mapped_column(JSONB)
172
+ api_key_encrypted: Mapped[bytes | None] = mapped_column(LargeBinary, default=None)
173
+ archived_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
174
+
175
+
176
+ class LlmProvider(AuditMixin, Base):
177
+ __tablename__ = "llm_providers"
178
+ __table_args__ = (
179
+ UniqueConstraint("name", name="uq_llm_providers_name"),
180
+ UniqueConstraint("base_url", name="uq_llm_providers_base_url"),
181
+ )
182
+
183
+ id: Mapped[int] = mapped_column(primary_key=True)
184
+ name: Mapped[str] = mapped_column(String(150))
185
+ base_url: Mapped[str] = mapped_column(String(500))
186
+ api_key_encrypted: Mapped[bytes | None] = mapped_column(LargeBinary, default=None)
187
+ archived_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
188
+
189
+
190
+ class ServerSettings(AuditMixin, Base):
191
+ __tablename__ = "server_settings"
192
+ __table_args__ = (CheckConstraint("id = 1", name="ck_server_settings_single_row"),)
193
+
194
+ id: Mapped[int] = mapped_column(primary_key=True)
195
+ default_judge: Mapped[dict | None] = mapped_column(JSONB, default=None)
196
+
197
+
198
+ class TraceSource(AuditMixin, Base):
199
+ __tablename__ = "trace_sources"
200
+ __table_args__ = (
201
+ UniqueConstraint("name", name="uq_trace_sources_name"),
202
+ check_in("provider", TraceProvider, "ck_trace_sources_provider"),
203
+ )
204
+
205
+ id: Mapped[int] = mapped_column(primary_key=True)
206
+ name: Mapped[str] = mapped_column(String(150))
207
+ provider: Mapped[str] = mapped_column(String(16))
208
+ config: Mapped[dict] = mapped_column(JSONB)
209
+ api_key_encrypted: Mapped[bytes | None] = mapped_column(LargeBinary, default=None)
210
+ archived_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
211
+
212
+
213
+ class TraceImport(AuditMixin, Base):
214
+ __tablename__ = "trace_imports"
215
+ __table_args__ = (
216
+ check_in("status", TraceImportStatus, "ck_trace_imports_status"),
217
+ CheckConstraint("since < until", name="ck_trace_imports_range"),
218
+ Index(
219
+ "uq_trace_imports_active",
220
+ "agent_id",
221
+ unique=True,
222
+ postgresql_where=text("status IN ('running', 'stopping')"),
223
+ ),
224
+ Index("ix_trace_imports_active", "status", "lease_expires_at"),
225
+ )
226
+
227
+ id: Mapped[int] = mapped_column(primary_key=True)
228
+ agent_id: Mapped[int] = mapped_column(ForeignKey("agents.id", ondelete="CASCADE"))
229
+ trace_source_id: Mapped[int] = mapped_column(ForeignKey("trace_sources.id"))
230
+ project: Mapped[str] = mapped_column(Text)
231
+ since: Mapped[datetime] = mapped_column(DateTime(timezone=True))
232
+ until: Mapped[datetime] = mapped_column(DateTime(timezone=True))
233
+ resume_from: Mapped[datetime] = mapped_column(DateTime(timezone=True))
234
+ resume_trace_ids: Mapped[list[str]] = mapped_column(ARRAY(Text), default=list)
235
+ status: Mapped[str] = mapped_column(String(16))
236
+ traces: Mapped[int] = mapped_column(Integer, default=0)
237
+ spans: Mapped[int] = mapped_column(BigInteger, default=0)
238
+ sample_session: Mapped[str | None] = mapped_column(String(255), default=None)
239
+ error: Mapped[str | None] = mapped_column(Text, default=None)
240
+ lease_holder: Mapped[str | None] = mapped_column(String(255), default=None)
241
+ lease_expires_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
242
+ created_by: Mapped[int] = mapped_column(ForeignKey("users.id"))
243
+ finished_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
244
+
245
+
246
+ class Span(Base):
247
+ __tablename__ = "spans"
248
+ __table_args__ = (
249
+ UniqueConstraint("agent_id", "trace_id", "span_id", name="uq_spans_agent_trace_span"),
250
+ Index("ix_spans_agent_session", "agent_id", "session_id"),
251
+ Index("ix_spans_agent_trace", "agent_id", "trace_id"),
252
+ )
253
+
254
+ id: Mapped[int] = mapped_column(BigInteger, primary_key=True)
255
+ agent_id: Mapped[int] = mapped_column(ForeignKey("agents.id", ondelete="CASCADE"))
256
+ trace_id: Mapped[str] = mapped_column(String(32))
257
+ span_id: Mapped[str] = mapped_column(String(16))
258
+ parent_span_id: Mapped[str | None] = mapped_column(String(16), nullable=True, default=None)
259
+ session_id: Mapped[str | None] = mapped_column(String(255), nullable=True, default=None)
260
+ name: Mapped[str] = mapped_column(Text)
261
+ kind: Mapped[str] = mapped_column(String(32))
262
+ start_time: Mapped[datetime] = mapped_column(DateTime(timezone=True))
263
+ end_time: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
264
+ status_code: Mapped[str] = mapped_column(String(8))
265
+ status_message: Mapped[str | None] = mapped_column(Text, nullable=True)
266
+ attributes: Mapped[dict] = mapped_column(JSONB)
267
+ events: Mapped[list] = mapped_column(JSONB)
268
+ resource: Mapped[dict] = mapped_column(JSONB)
269
+ received_at: Mapped[datetime] = mapped_column(
270
+ DateTime(timezone=True), default=func.now(), server_default=func.now(), nullable=False
271
+ )
272
+ external_trace_source_id: Mapped[int | None] = mapped_column(
273
+ ForeignKey("trace_sources.id"), default=None
274
+ )
275
+
276
+
277
+ class SpanMapping(AuditMixin, Base):
278
+ __tablename__ = "span_mappings"
279
+ __table_args__ = (
280
+ UniqueConstraint("agent_id", "version", name="uq_span_mappings_agent_version"),
281
+ check_in("status", MappingStatus, "ck_span_mappings_status"),
282
+ Index(
283
+ "uq_span_mappings_active",
284
+ "agent_id",
285
+ unique=True,
286
+ postgresql_where=text("status = 'active'"),
287
+ ),
288
+ Index(
289
+ "uq_span_mappings_draft",
290
+ "agent_id",
291
+ unique=True,
292
+ postgresql_where=text("status = 'draft'"),
293
+ ),
294
+ )
295
+
296
+ id: Mapped[int] = mapped_column(primary_key=True)
297
+ agent_id: Mapped[int] = mapped_column(ForeignKey("agents.id", ondelete="CASCADE"))
298
+ version: Mapped[int] = mapped_column(Integer)
299
+ status: Mapped[str] = mapped_column(String(8))
300
+ mapping: Mapped[dict] = mapped_column(JSONB)
301
+ created_by_id: Mapped[int | None] = mapped_column(ForeignKey("users.id"), default=None)
302
+ activated_at: Mapped[datetime | None] = mapped_column(
303
+ DateTime(timezone=True), nullable=True, default=None
304
+ )
305
+
306
+
307
+ class Case(AuditMixin, Base):
308
+ __tablename__ = "cases"
309
+ __table_args__ = (UniqueConstraint("case_id", name="uq_cases_case_id"),)
310
+
311
+ id: Mapped[int] = mapped_column(primary_key=True)
312
+ case_id: Mapped[str] = mapped_column(String(255))
313
+
314
+
315
+ class CaseVersion(AuditMixin, Base):
316
+ __tablename__ = "case_versions"
317
+ __table_args__ = (Index("ix_case_versions_tags", "tags", postgresql_using="gin"),)
318
+
319
+ case_ref: Mapped[int] = mapped_column(ForeignKey("cases.id"), primary_key=True)
320
+ content_hash: Mapped[str] = mapped_column(String(16), primary_key=True)
321
+ input: Mapped[dict] = mapped_column(JSONB)
322
+ expected: Mapped[dict | None] = mapped_column(JSONB, default=None)
323
+ evaluators: Mapped[dict] = mapped_column(JSONB)
324
+ tags: Mapped[list[str]] = mapped_column(ARRAY(Text), default=list)
325
+
326
+
327
+ class FixtureBlob(AuditMixin, Base):
328
+ __tablename__ = "fixture_blobs"
329
+
330
+ sha256: Mapped[str] = mapped_column(String(64), primary_key=True)
331
+ size: Mapped[int] = mapped_column(BigInteger)
332
+ uploaded_by: Mapped[int | None] = mapped_column(ForeignKey("users.id"), default=None)
333
+
334
+
335
+ class CaseVersionFile(AuditMixin, Base):
336
+ __tablename__ = "case_version_files"
337
+ __table_args__ = (
338
+ ForeignKeyConstraint(
339
+ ["case_ref", "content_hash"], ["case_versions.case_ref", "case_versions.content_hash"]
340
+ ),
341
+ )
342
+
343
+ case_ref: Mapped[int] = mapped_column(primary_key=True)
344
+ content_hash: Mapped[str] = mapped_column(String(16), primary_key=True)
345
+ key: Mapped[str] = mapped_column(String(255), primary_key=True)
346
+ sha256: Mapped[str] = mapped_column(ForeignKey("fixture_blobs.sha256"))
347
+ context: Mapped[str] = mapped_column(Text)
348
+ position: Mapped[int] = mapped_column(Integer)
349
+
350
+
351
+ class Dataset(AuditMixin, Base):
352
+ __tablename__ = "datasets"
353
+ __table_args__ = (
354
+ UniqueConstraint("workspace_id", "name", name="uq_datasets_workspace_name"),
355
+ check_in("kind", DatasetKind, "ck_datasets_kind"),
356
+ )
357
+
358
+ id: Mapped[int] = mapped_column(primary_key=True)
359
+ workspace_id: Mapped[int] = mapped_column(ForeignKey("workspaces.id"))
360
+ name: Mapped[str] = mapped_column(String(150))
361
+ kind: Mapped[str] = mapped_column(String(16))
362
+ archived_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
363
+
364
+
365
+ class DatasetSync(AuditMixin, Base):
366
+ __tablename__ = "dataset_syncs"
367
+ __table_args__ = (check_in("source", SyncSource, "ck_dataset_syncs_source"),)
368
+
369
+ id: Mapped[int] = mapped_column(primary_key=True)
370
+ dataset_id: Mapped[int] = mapped_column(ForeignKey("datasets.id"), index=True)
371
+ actor_id: Mapped[int] = mapped_column(ForeignKey("users.id"))
372
+ source: Mapped[str] = mapped_column(String(16))
373
+ detail: Mapped[dict] = mapped_column(JSONB)
374
+
375
+
376
+ class DatasetCase(AuditMixin, Base):
377
+ __tablename__ = "dataset_cases"
378
+ __table_args__ = (
379
+ ForeignKeyConstraint(
380
+ ["case_ref", "content_hash"], ["case_versions.case_ref", "case_versions.content_hash"]
381
+ ),
382
+ )
383
+
384
+ dataset_id: Mapped[int] = mapped_column(ForeignKey("datasets.id"), primary_key=True)
385
+ case_ref: Mapped[int] = mapped_column(ForeignKey("cases.id"), primary_key=True)
386
+ content_hash: Mapped[str] = mapped_column(String(16))
387
+ path_tags: Mapped[list[str]] = mapped_column(
388
+ ARRAY(Text), default=list, server_default=text("'{}'::text[]")
389
+ )
390
+ source: Mapped[str | None] = mapped_column(String(500), default=None)
391
+ removed_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
392
+
393
+
394
+ class CaseEvent(AuditMixin, Base):
395
+ __tablename__ = "case_events"
396
+ __table_args__ = (
397
+ ForeignKeyConstraint(
398
+ ["dataset_id", "case_ref"], ["dataset_cases.dataset_id", "dataset_cases.case_ref"]
399
+ ),
400
+ ForeignKeyConstraint(
401
+ ["case_ref", "content_hash"], ["case_versions.case_ref", "case_versions.content_hash"]
402
+ ),
403
+ check_in("kind", CaseEventKind, "ck_case_events_kind"),
404
+ Index("ix_case_events_dataset_case", "dataset_id", "case_ref", "content_hash"),
405
+ )
406
+
407
+ id: Mapped[int] = mapped_column(primary_key=True)
408
+ dataset_id: Mapped[int] = mapped_column(Integer)
409
+ case_ref: Mapped[int] = mapped_column(Integer)
410
+ content_hash: Mapped[str] = mapped_column(String(16))
411
+ kind: Mapped[str] = mapped_column(String(32))
412
+ sync_id: Mapped[int | None] = mapped_column(ForeignKey("dataset_syncs.id"), default=None)
413
+ actor_id: Mapped[int] = mapped_column(ForeignKey("users.id"))
414
+ target_id: Mapped[int | None] = mapped_column(ForeignKey("users.id"), default=None)
415
+
416
+
417
+ class CaseComment(AuditMixin, Base):
418
+ __tablename__ = "case_comments"
419
+ __table_args__ = (
420
+ ForeignKeyConstraint(
421
+ ["dataset_id", "case_ref"], ["dataset_cases.dataset_id", "dataset_cases.case_ref"]
422
+ ),
423
+ ForeignKeyConstraint(
424
+ ["case_ref", "content_hash"], ["case_versions.case_ref", "case_versions.content_hash"]
425
+ ),
426
+ Index("ix_case_comments_dataset_case", "dataset_id", "case_ref"),
427
+ )
428
+
429
+ id: Mapped[int] = mapped_column(primary_key=True)
430
+ dataset_id: Mapped[int] = mapped_column(Integer)
431
+ case_ref: Mapped[int] = mapped_column(Integer)
432
+ content_hash: Mapped[str] = mapped_column(String(16))
433
+ parent_id: Mapped[int | None] = mapped_column(
434
+ ForeignKey("case_comments.id", ondelete="CASCADE"), default=None
435
+ )
436
+ author_id: Mapped[int] = mapped_column(ForeignKey("users.id"))
437
+ body: Mapped[str] = mapped_column(Text)
438
+ resolved: Mapped[bool] = mapped_column(Boolean, default=False, server_default=text("false"))
439
+
440
+
441
+ class Overlay(AuditMixin, Base):
442
+ __tablename__ = "overlays"
443
+ __table_args__ = (
444
+ UniqueConstraint("source_dataset_id", "target_case", name="uq_overlays_source_target"),
445
+ )
446
+
447
+ id: Mapped[int] = mapped_column(primary_key=True)
448
+ source_dataset_id: Mapped[int] = mapped_column(ForeignKey("datasets.id"))
449
+ target_case: Mapped[int] = mapped_column(ForeignKey("cases.id"))
450
+ agents: Mapped[list[str]] = mapped_column(ARRAY(Text))
451
+ tags: Mapped[list[str]] = mapped_column(ARRAY(Text))
452
+ evaluators: Mapped[dict] = mapped_column(JSONB)
453
+ source_file: Mapped[str | None] = mapped_column(String(500), default=None)
454
+
455
+
456
+ class Experiment(AuditMixin, Base):
457
+ __tablename__ = "experiments"
458
+ __table_args__ = (
459
+ UniqueConstraint("dataset_id", "number", name="uq_experiments_dataset_number"),
460
+ check_in("status", ExperimentStatus, "ck_experiments_status"),
461
+ Index("ix_experiments_active", "status", "lease_expires_at"),
462
+ )
463
+
464
+ id: Mapped[int] = mapped_column(primary_key=True)
465
+ workspace_id: Mapped[int] = mapped_column(ForeignKey("workspaces.id"), index=True)
466
+ dataset_id: Mapped[int] = mapped_column(ForeignKey("datasets.id"), index=True)
467
+ number: Mapped[int] = mapped_column(Integer)
468
+ agent_id: Mapped[int] = mapped_column(ForeignKey("agents.id"), index=True)
469
+ created_by: Mapped[int] = mapped_column(ForeignKey("users.id"))
470
+ status: Mapped[str] = mapped_column(String(16))
471
+ status_error: Mapped[str | None] = mapped_column(Text, default=None)
472
+ lease_holder: Mapped[str | None] = mapped_column(String(255), default=None)
473
+ lease_expires_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
474
+ repetitions: Mapped[int] = mapped_column(Integer)
475
+ config: Mapped[dict] = mapped_column(JSONB)
476
+ started_at: Mapped[datetime] = mapped_column(
477
+ DateTime(timezone=True), default=func.now(), server_default=func.now()
478
+ )
479
+ ended_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
480
+
481
+
482
+ class ExperimentCase(AuditMixin, Base):
483
+ __tablename__ = "experiment_cases"
484
+ __table_args__ = (
485
+ ForeignKeyConstraint(
486
+ ["case_ref", "content_hash"], ["case_versions.case_ref", "case_versions.content_hash"]
487
+ ),
488
+ )
489
+
490
+ experiment_id: Mapped[int] = mapped_column(ForeignKey("experiments.id"), primary_key=True)
491
+ case_ref: Mapped[int] = mapped_column(ForeignKey("cases.id"), primary_key=True)
492
+ content_hash: Mapped[str] = mapped_column(String(16))
493
+ effective_evaluators: Mapped[dict] = mapped_column(JSONB)
494
+
495
+
496
+ class AgentRun(AuditMixin, Base):
497
+ __tablename__ = "agent_runs"
498
+ __table_args__ = (
499
+ UniqueConstraint("agent_id", "harness_run_id", name="uq_agent_runs_agent_harness_run"),
500
+ check_in("status", RunStatus, "ck_agent_runs_status"),
501
+ CheckConstraint(
502
+ "trajectory_status IS NULL OR trajectory_status IN "
503
+ "('valid', 'partial', 'invalid', 'unavailable')",
504
+ name="ck_agent_runs_trajectory_status",
505
+ ),
506
+ Index("ix_agent_runs_agent_started", "agent_id", "started_at"),
507
+ )
508
+
509
+ id: Mapped[int] = mapped_column(primary_key=True)
510
+ agent_id: Mapped[int] = mapped_column(ForeignKey("agents.id"))
511
+ harness_run_id: Mapped[str] = mapped_column(String(255))
512
+ status: Mapped[str] = mapped_column(String(16))
513
+ answer: Mapped[str | None] = mapped_column(Text, default=None)
514
+ output: Mapped[dict | None] = mapped_column(JSONB, default=None)
515
+ error: Mapped[str | None] = mapped_column(Text, default=None)
516
+ started_at: Mapped[datetime] = mapped_column(DateTime(timezone=True))
517
+ ended_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), default=None)
518
+ trajectory_key: Mapped[str | None] = mapped_column(String(500), default=None)
519
+ trajectory_status: Mapped[str | None] = mapped_column(String(16), default=None)
520
+ metrics: Mapped[dict] = mapped_column(JSONB, default=dict)
521
+ title: Mapped[str | None] = mapped_column(String(200), default=None)
522
+ created_by: Mapped[int | None] = mapped_column(ForeignKey("users.id"), default=None)
523
+
524
+
525
+ class ExperimentRun(AuditMixin, Base):
526
+ __tablename__ = "experiment_runs"
527
+ __table_args__ = (
528
+ UniqueConstraint(
529
+ "experiment_id", "case_ref", "repetition", name="uq_experiment_runs_case_repetition"
530
+ ),
531
+ UniqueConstraint("agent_run_id", name="uq_experiment_runs_agent_run"),
532
+ ForeignKeyConstraint(
533
+ ["experiment_id", "case_ref"],
534
+ ["experiment_cases.experiment_id", "experiment_cases.case_ref"],
535
+ ),
536
+ )
537
+
538
+ id: Mapped[int] = mapped_column(primary_key=True)
539
+ experiment_id: Mapped[int] = mapped_column(Integer)
540
+ case_ref: Mapped[int] = mapped_column(Integer)
541
+ repetition: Mapped[int] = mapped_column(Integer)
542
+ agent_run_id: Mapped[int] = mapped_column(ForeignKey("agent_runs.id"))
543
+
544
+
545
+ class Evaluation(AuditMixin, Base):
546
+ __tablename__ = "evaluations"
547
+ __table_args__ = (
548
+ UniqueConstraint("experiment_run_id", "name", name="uq_evaluations_run_name"),
549
+ check_in("kind", EvaluatorKind, "ck_evaluations_kind"),
550
+ check_in("source", EvaluationSource, "ck_evaluations_source"),
551
+ CheckConstraint(
552
+ f"bucket IS NULL OR bucket IN ({', '.join(repr(b.value) for b in Bucket)})",
553
+ name="ck_evaluations_bucket",
554
+ ),
555
+ )
556
+
557
+ id: Mapped[int] = mapped_column(primary_key=True)
558
+ experiment_run_id: Mapped[int] = mapped_column(ForeignKey("experiment_runs.id"), index=True)
559
+ producer: Mapped[str] = mapped_column(String(150))
560
+ name: Mapped[str] = mapped_column(String(150))
561
+ kind: Mapped[str] = mapped_column(String(8))
562
+ source: Mapped[str] = mapped_column(String(8))
563
+ bucket: Mapped[str | None] = mapped_column(String(32), default=None)
564
+ value: Mapped[float | None] = mapped_column(Float, default=None)
565
+ label: Mapped[str | None] = mapped_column(String(150), default=None)
566
+ explanation: Mapped[str | None] = mapped_column(Text, default=None)
567
+ error: Mapped[str | None] = mapped_column(Text, default=None)