@piagent/platform 1.6.0 → 1.7.1

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 (736) hide show
  1. package/CHANGELOG.md +104 -0
  2. package/README.md +24 -15
  3. package/README.vi.md +36 -2
  4. package/adapters/common/assurance-producers.json +109 -0
  5. package/adapters/common/structured-assurance-facts.mjs +153 -0
  6. package/adapters/node-typescript/contract-families.json +2811 -0
  7. package/adapters/node-typescript/contract-families.md +90 -0
  8. package/benchmarks/production-v1/grade.mjs +20 -0
  9. package/benchmarks/production-v1/spend-control.v1.json +27 -3
  10. package/benchmarks/production-v1/suite.json +5 -1
  11. package/benchmarks/production-v2/grade.mjs +892 -0
  12. package/benchmarks/production-v2/project/README.md +7 -0
  13. package/benchmarks/production-v2/project/config/service.json +4 -0
  14. package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
  15. package/benchmarks/production-v2/project/docs/ops.md +3 -0
  16. package/benchmarks/production-v2/project/logs/incident.log +2 -0
  17. package/benchmarks/production-v2/project/package.json +11 -0
  18. package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
  19. package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
  20. package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
  21. package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
  22. package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
  23. package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
  24. package/benchmarks/production-v2/project/src/data/csv.js +3 -0
  25. package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
  26. package/benchmarks/production-v2/project/src/data/migration.js +9 -0
  27. package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
  28. package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
  29. package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
  30. package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
  31. package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
  32. package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
  33. package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
  34. package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
  35. package/benchmarks/production-v2/project/src/platform/args.js +10 -0
  36. package/benchmarks/production-v2/project/src/platform/config.js +7 -0
  37. package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
  38. package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
  39. package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
  40. package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
  41. package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
  42. package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
  43. package/benchmarks/production-v2/project/vendor/README.md +1 -0
  44. package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
  45. package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
  46. package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
  47. package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
  48. package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
  49. package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
  50. package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
  51. package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
  52. package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
  53. package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
  54. package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
  55. package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
  56. package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
  57. package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
  58. package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
  59. package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
  60. package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
  61. package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
  62. package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
  63. package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
  64. package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
  65. package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
  66. package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
  67. package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
  68. package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
  69. package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
  70. package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
  71. package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
  72. package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
  73. package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
  74. package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
  75. package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
  76. package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
  77. package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
  78. package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
  79. package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
  80. package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
  81. package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
  82. package/benchmarks/production-v2/spend-control.v1.json +115 -0
  83. package/benchmarks/production-v2/suite.json +1009 -0
  84. package/benchmarks/production-v2/variant.mjs +169 -0
  85. package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
  86. package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
  87. package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
  88. package/benchmarks/production-v3/grade.mjs +1017 -0
  89. package/benchmarks/production-v3/project/README.md +7 -0
  90. package/benchmarks/production-v3/project/config/service.json +4 -0
  91. package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
  92. package/benchmarks/production-v3/project/docs/ops.md +3 -0
  93. package/benchmarks/production-v3/project/logs/incident.log +2 -0
  94. package/benchmarks/production-v3/project/package.json +11 -0
  95. package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
  96. package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
  97. package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
  98. package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
  99. package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
  100. package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
  101. package/benchmarks/production-v3/project/src/data/csv.js +3 -0
  102. package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
  103. package/benchmarks/production-v3/project/src/data/migration.js +9 -0
  104. package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
  105. package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
  106. package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
  107. package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
  108. package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
  109. package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
  110. package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
  111. package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
  112. package/benchmarks/production-v3/project/src/platform/args.js +10 -0
  113. package/benchmarks/production-v3/project/src/platform/config.js +7 -0
  114. package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
  115. package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
  116. package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
  117. package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
  118. package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
  119. package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
  120. package/benchmarks/production-v3/project/vendor/README.md +1 -0
  121. package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
  122. package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
  123. package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
  124. package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
  125. package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
  126. package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
  127. package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
  128. package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
  129. package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
  130. package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
  131. package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
  132. package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
  133. package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
  134. package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
  135. package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
  136. package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
  137. package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
  138. package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
  139. package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
  140. package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
  141. package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
  142. package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
  143. package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
  144. package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
  145. package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
  146. package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
  147. package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
  148. package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
  149. package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
  150. package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
  151. package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
  152. package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
  153. package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
  154. package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
  155. package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
  156. package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
  157. package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
  158. package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
  159. package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
  160. package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
  161. package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
  162. package/benchmarks/production-v3/spend-control.v1.json +121 -0
  163. package/benchmarks/production-v3/suite.json +1010 -0
  164. package/benchmarks/production-v3/variant.mjs +178 -0
  165. package/catalog/capabilities.json +16 -16
  166. package/docs/authority-manifest.md +1 -1
  167. package/docs/auto-delegation-policy.md +38 -73
  168. package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
  169. package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
  170. package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
  171. package/docs/benchmark-public-contract-addendum-v2.md +85 -0
  172. package/docs/benchmark-public-contract-addendum-v3.md +116 -0
  173. package/docs/command-reference-vietnamese.md +30 -35
  174. package/docs/context-engine.md +5 -3
  175. package/docs/context-window-policy.md +37 -1
  176. package/docs/distribution-standard.md +9 -9
  177. package/docs/en/architecture.md +26 -13
  178. package/docs/languages.json +4 -4
  179. package/docs/operator-manual-vietnamese.md +26 -46
  180. package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
  181. package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
  182. package/docs/piagent-core-improvement-plan.md +528 -0
  183. package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
  184. package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
  185. package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
  186. package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
  187. package/docs/piagent-open-session-wait.md +19 -0
  188. package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
  189. package/docs/publishing-for-teams.md +7 -0
  190. package/docs/quality-benchmark.md +299 -23
  191. package/docs/quickstart-vietnamese.md +2 -2
  192. package/docs/release-install-policy.md +32 -9
  193. package/docs/research/harness-next-evidence-review.md +113 -0
  194. package/docs/runtime-harness-standard.md +3 -3
  195. package/docs/security-threat-model.md +1 -1
  196. package/docs/subagent-orchestration-capabilities.md +19 -51
  197. package/docs/subagents-and-multiagent.md +21 -47
  198. package/docs/task-implementation-contract.md +3 -3
  199. package/docs/task-lifecycle-vietnamese.md +1 -1
  200. package/docs/team-onboarding.md +7 -7
  201. package/docs/third-party-neutrality.md +7 -0
  202. package/docs/usage-observability.md +1 -1
  203. package/docs/vi/architecture.md +26 -13
  204. package/docs-site/architecture.html +1 -1
  205. package/docs-site/benchmark.html +89 -95
  206. package/docs-site/commands.html +1 -1
  207. package/docs-site/content/benchmark.html +88 -94
  208. package/docs-site/content/en/benchmark.html +134 -90
  209. package/docs-site/content/en/index.html +19 -19
  210. package/docs-site/content/en/runtime.html +5 -5
  211. package/docs-site/content/en/whats-new.html +23 -1
  212. package/docs-site/content/en/workflows.html +2 -2
  213. package/docs-site/content/index.html +19 -19
  214. package/docs-site/content/runtime.html +6 -6
  215. package/docs-site/content/whats-new.html +23 -1
  216. package/docs-site/content/workflows.html +2 -2
  217. package/docs-site/context-engine.html +1 -1
  218. package/docs-site/en/architecture.html +1 -1
  219. package/docs-site/en/benchmark.html +135 -91
  220. package/docs-site/en/commands.html +1 -1
  221. package/docs-site/en/context-engine.html +1 -1
  222. package/docs-site/en/index.html +20 -20
  223. package/docs-site/en/mcp-approval.html +1 -1
  224. package/docs-site/en/mcp-auth.html +1 -1
  225. package/docs-site/en/mcp-servers.html +1 -1
  226. package/docs-site/en/mcp.html +1 -1
  227. package/docs-site/en/permissions.html +1 -1
  228. package/docs-site/en/profiles.html +1 -1
  229. package/docs-site/en/quickstart.html +1 -1
  230. package/docs-site/en/runtime.html +6 -6
  231. package/docs-site/en/scope.html +1 -1
  232. package/docs-site/en/security.html +1 -1
  233. package/docs-site/en/team.html +1 -1
  234. package/docs-site/en/whats-new.html +24 -2
  235. package/docs-site/en/workflows.html +3 -3
  236. package/docs-site/index.html +20 -20
  237. package/docs-site/mcp-approval.html +1 -1
  238. package/docs-site/mcp-auth.html +1 -1
  239. package/docs-site/mcp-servers.html +1 -1
  240. package/docs-site/mcp.html +1 -1
  241. package/docs-site/permissions.html +1 -1
  242. package/docs-site/profiles.html +1 -1
  243. package/docs-site/quickstart.html +1 -1
  244. package/docs-site/runtime.html +7 -7
  245. package/docs-site/scope.html +1 -1
  246. package/docs-site/security.html +1 -1
  247. package/docs-site/team.html +1 -1
  248. package/docs-site/whats-new.html +24 -2
  249. package/docs-site/workflows.html +3 -3
  250. package/evals/architecture-conformance-v1/lane.json +18 -0
  251. package/evals/architecture-conformance-v1/runner.mjs +77 -0
  252. package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
  253. package/evals/fixtures/approved-host-contracts.valid.json +43 -0
  254. package/evals/fixtures/benchmark-assurance-evidence.invalid.json +1 -1
  255. package/evals/fixtures/benchmark-assurance-evidence.valid.json +1 -1
  256. package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
  257. package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
  258. package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
  259. package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
  260. package/evals/fixtures/codex-relative-efficiency-v1.invalid.json +4 -0
  261. package/evals/fixtures/codex-relative-efficiency-v1.valid.json +94 -0
  262. package/evals/fixtures/composite-criterion.invalid.json +10 -0
  263. package/evals/fixtures/composite-criterion.valid.json +33 -0
  264. package/evals/fixtures/contract-family-library.invalid.json +42 -0
  265. package/evals/fixtures/contract-family-library.valid.json +41 -0
  266. package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
  267. package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
  268. package/evals/fixtures/helper-request.invalid.json +1 -1
  269. package/evals/fixtures/helper-request.valid.json +1 -1
  270. package/evals/fixtures/host-contract-plan.invalid.json +42 -0
  271. package/evals/fixtures/host-contract-plan.valid.json +41 -0
  272. package/evals/fixtures/piagent-webui/subagent-tree-v1.valid.json +7 -0
  273. package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
  274. package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
  275. package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
  276. package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
  277. package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
  278. package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
  279. package/evals/fixtures/task-authority-snapshot.valid.json +4 -4
  280. package/evals/fs4-readiness-matrix.v1.json +2 -2
  281. package/evals/fs5-artifact-archive/README.md +32 -0
  282. package/evals/fs5-artifact-archive/manifest.json +148 -0
  283. package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
  284. package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
  285. package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
  286. package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
  287. package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
  288. package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
  289. package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
  290. package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
  291. package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
  292. package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
  293. package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
  294. package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
  295. package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
  296. package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
  297. package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
  298. package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
  299. package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
  300. package/evals/fs5-causal-arm.v1.json +10 -6
  301. package/evals/fs5-pilot-protocol.v1.json +3 -3
  302. package/evals/fs5-pilot-protocol.v2.json +8 -8
  303. package/evals/fs5-pilot-protocol.v3.json +8 -8
  304. package/evals/fs5-pilot-protocol.v4.json +7 -7
  305. package/evals/fs5-pilot-protocol.v5.json +10 -8
  306. package/evals/harness-next/development-corpus.mjs +106 -0
  307. package/evals/harness-next/resource-budget-probe.mjs +38 -0
  308. package/evals/harness-next/run-calibration.mjs +83 -0
  309. package/evals/harness-next/run-resource-probe.mjs +55 -0
  310. package/evals/harness-next/selected-development-corpus.mjs +27 -0
  311. package/evals/harness-next.v1.json +51 -0
  312. package/evals/long-horizon-v1/README.md +19 -0
  313. package/evals/long-horizon-v1/runner.mjs +116 -48
  314. package/evals/long-horizon-v1/worker.mjs +77 -30
  315. package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
  316. package/evals/private-holdout-v1/README.md +5 -2
  317. package/evals/private-holdout-v1/public-exposure.v1.json +1 -1
  318. package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
  319. package/evals/runtime-conformance-v1/README.md +43 -0
  320. package/evals/runtime-conformance-v1/evaluate-paid-canary.mjs +198 -0
  321. package/evals/runtime-conformance-v1/lane.json +57 -0
  322. package/evals/runtime-conformance-v1/paid-canary.json +31 -0
  323. package/evals/runtime-conformance-v1/runner.mjs +456 -0
  324. package/package.json +9 -4
  325. package/packages/piagent-core/README.md +15 -6
  326. package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
  327. package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
  328. package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
  329. package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
  330. package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
  331. package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
  332. package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
  333. package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
  334. package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
  335. package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
  336. package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
  337. package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +699 -0
  338. package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
  339. package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
  340. package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
  341. package/packages/piagent-core/benchmark/benchmark-core.js +111 -110
  342. package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
  343. package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
  344. package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
  345. package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
  346. package/packages/piagent-core/benchmark/benchmark-forensics.js +67 -52
  347. package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
  348. package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
  349. package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
  350. package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
  351. package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
  352. package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
  353. package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
  354. package/packages/piagent-core/benchmark/benchmark-preflight.js +70 -7
  355. package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
  356. package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
  357. package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
  358. package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +462 -0
  359. package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
  360. package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
  361. package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
  362. package/packages/piagent-core/benchmark/benchmark-record-validation.js +343 -8
  363. package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
  364. package/packages/piagent-core/benchmark/benchmark-report.js +94 -5
  365. package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
  366. package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
  367. package/packages/piagent-core/benchmark/benchmark-runtime-causal.js +244 -0
  368. package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
  369. package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
  370. package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
  371. package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
  372. package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +246 -33
  373. package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
  374. package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
  375. package/packages/piagent-core/benchmark/benchmark-suite.js +232 -4
  376. package/packages/piagent-core/benchmark/benchmark-summary-support.js +33 -1
  377. package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
  378. package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
  379. package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
  380. package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
  381. package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
  382. package/packages/piagent-core/benchmark/benchmark-usage.js +410 -37
  383. package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
  384. package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
  385. package/packages/piagent-core/capabilities/runtime-integrity.js +6 -0
  386. package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
  387. package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
  388. package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
  389. package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
  390. package/packages/piagent-core/extensions/acceptance-behavior-proof.js +1000 -0
  391. package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
  392. package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
  393. package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
  394. package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
  395. package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
  396. package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
  397. package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
  398. package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
  399. package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
  400. package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
  401. package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
  402. package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
  403. package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
  404. package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
  405. package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
  406. package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
  407. package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
  408. package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
  409. package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
  410. package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
  411. package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
  412. package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
  413. package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
  414. package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
  415. package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
  416. package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
  417. package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
  418. package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
  419. package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
  420. package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
  421. package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
  422. package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
  423. package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
  424. package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
  425. package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
  426. package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
  427. package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
  428. package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
  429. package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
  430. package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
  431. package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
  432. package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
  433. package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
  434. package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
  435. package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
  436. package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
  437. package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
  438. package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
  439. package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
  440. package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
  441. package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
  442. package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
  443. package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
  444. package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
  445. package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
  446. package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
  447. package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
  448. package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
  449. package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
  450. package/packages/piagent-core/extensions/acceptance-receipt.js +208 -180
  451. package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
  452. package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
  453. package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
  454. package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
  455. package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
  456. package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
  457. package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
  458. package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
  459. package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
  460. package/packages/piagent-core/extensions/context-engine.js +12 -17
  461. package/packages/piagent-core/extensions/context-index-policy.js +4 -0
  462. package/packages/piagent-core/extensions/context-query-paths.js +57 -0
  463. package/packages/piagent-core/extensions/criterion-context-pack.js +109 -4
  464. package/packages/piagent-core/extensions/criterion-graph.js +4 -1
  465. package/packages/piagent-core/extensions/failure-types.ts +2 -2
  466. package/packages/piagent-core/extensions/fallback-base-policy.ts +101 -0
  467. package/packages/piagent-core/extensions/guard-types.ts +9 -1
  468. package/packages/piagent-core/extensions/piagent-guard.ts +374 -364
  469. package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
  470. package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
  471. package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
  472. package/packages/piagent-core/extensions/task-contract-view.js +4 -4
  473. package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
  474. package/packages/piagent-core/extensions/task-journal.js +48 -3
  475. package/packages/piagent-core/extensions/task-lifecycle.js +32 -0
  476. package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
  477. package/packages/piagent-core/extensions/task-state.js +140 -43
  478. package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
  479. package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
  480. package/packages/piagent-core/extensions/verification-intelligence.js +109 -20
  481. package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
  482. package/packages/piagent-core/extensions/workspace-evidence-roots.js +52 -16
  483. package/packages/piagent-core/extensions/workspace-revision.js +48 -0
  484. package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
  485. package/packages/piagent-core/package.json +3 -3
  486. package/packages/piagent-core/policies/base-policy.json +43 -2
  487. package/packages/piagent-core/policy/authority-manifest.v1.json +3 -3
  488. package/packages/piagent-core/prompts/be-to-fe.md +2 -8
  489. package/packages/piagent-core/prompts/plan.md +1 -1
  490. package/packages/piagent-core/prompts/platform-improve.md +2 -6
  491. package/packages/piagent-core/prompts/review.md +2 -7
  492. package/packages/piagent-core/prompts/scout.md +1 -1
  493. package/packages/piagent-core/prompts/task.md +4 -3
  494. package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
  495. package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
  496. package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
  497. package/packages/piagent-core/runtime/hooks/completion-hook.ts +83 -89
  498. package/packages/piagent-core/runtime/hooks/input-hook.ts +98 -7
  499. package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
  500. package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
  501. package/packages/piagent-core/runtime/hooks/session-start-hook.ts +150 -19
  502. package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +111 -92
  503. package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
  504. package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
  505. package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
  506. package/packages/piagent-core/runtime/hooks/tool-result-value-helpers.ts +66 -0
  507. package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
  508. package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
  509. package/packages/piagent-core/runtime/inspection/bounded-emission-guard.ts +286 -0
  510. package/packages/piagent-core/runtime/inspection/runtime-event-store.ts +37 -7
  511. package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
  512. package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
  513. package/packages/piagent-core/runtime/inspection/task-subagent-tree.ts +27 -4
  514. package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +69 -11
  515. package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +104 -22
  516. package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
  517. package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
  518. package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
  519. package/packages/piagent-core/runtime/orchestration/helper-lifecycle.ts +16 -12
  520. package/packages/piagent-core/runtime/orchestration/minimal-delegation-policy.ts +71 -0
  521. package/packages/piagent-core/runtime/orchestration/owned-work-budget.ts +28 -20
  522. package/packages/piagent-core/runtime/orchestration/role-policy.ts +27 -9
  523. package/packages/piagent-core/runtime/orchestration/subagent-tool-policy.ts +329 -0
  524. package/packages/piagent-core/runtime/policy/block-remedy.ts +0 -4
  525. package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
  526. package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
  527. package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
  528. package/packages/piagent-core/runtime/product/efficiency-metrics.ts +15 -7
  529. package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
  530. package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
  531. package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
  532. package/packages/piagent-core/runtime/quality/edit-freshness-guard.ts +102 -0
  533. package/packages/piagent-core/runtime/quality/exact-output-contract.ts +20 -10
  534. package/packages/piagent-core/runtime/quality/performance-assurance.ts +3 -3
  535. package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
  536. package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
  537. package/packages/piagent-core/runtime/recovery/continuation-budget.ts +12 -3
  538. package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
  539. package/packages/piagent-core/runtime/recovery/handoff-projection.ts +196 -58
  540. package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
  541. package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
  542. package/packages/piagent-core/runtime/recovery/pristine-task-policy-refresh.ts +91 -0
  543. package/packages/piagent-core/runtime/recovery/recovery-policy.ts +31 -40
  544. package/packages/piagent-core/runtime/recovery/resume-state.ts +120 -19
  545. package/packages/piagent-core/runtime/recovery/semantic-repair-handshake.ts +7 -13
  546. package/packages/piagent-core/runtime/recovery/task-evidence-coverage-refresh.ts +217 -0
  547. package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
  548. package/packages/piagent-core/runtime/registration/context-commands.ts +28 -10
  549. package/packages/piagent-core/runtime/registration/knowledge-tools.ts +9 -5
  550. package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
  551. package/packages/piagent-core/runtime/registration/operator-request-intake.ts +49 -0
  552. package/packages/piagent-core/runtime/registration/policy-tools.ts +8 -6
  553. package/packages/piagent-core/runtime/registration/session-commands.ts +10 -8
  554. package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
  555. package/packages/piagent-core/runtime/registration/task-completion-tools.ts +39 -20
  556. package/packages/piagent-core/runtime/registration/task-evidence-tools.ts +21 -12
  557. package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
  558. package/packages/piagent-core/runtime/registration/task-start-guidance.ts +24 -15
  559. package/packages/piagent-core/runtime/registration/task-start-manifest.ts +54 -0
  560. package/packages/piagent-core/runtime/registration/task-start-tool.ts +79 -79
  561. package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
  562. package/packages/piagent-core/runtime/runtime-limits.ts +16 -2
  563. package/packages/piagent-core/runtime/session/adaptive-context-analysis.ts +461 -0
  564. package/packages/piagent-core/runtime/session/adaptive-context-governor.ts +464 -0
  565. package/packages/piagent-core/runtime/session/adaptive-context-ledger.ts +265 -0
  566. package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
  567. package/packages/piagent-core/runtime/session/message-signals.ts +6 -18
  568. package/packages/piagent-core/runtime/session/operator-request-carry.ts +44 -0
  569. package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
  570. package/packages/piagent-core/runtime/session/runtime-state.ts +37 -39
  571. package/packages/piagent-core/runtime/session/source-checkout-read-grants.ts +35 -0
  572. package/packages/piagent-core/runtime/session/system-prompt.ts +76 -14
  573. package/packages/piagent-core/runtime/session/task-contract-presentation.ts +39 -0
  574. package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
  575. package/packages/piagent-core/runtime/session/usage.ts +4 -3
  576. package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
  577. package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
  578. package/packages/piagent-core/runtime/tools/phase-tools.ts +10 -9
  579. package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
  580. package/packages/piagent-core/runtime/tools/tool-groups.ts +22 -5
  581. package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
  582. package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
  583. package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
  584. package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
  585. package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
  586. package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
  587. package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
  588. package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
  589. package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
  590. package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
  591. package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +101 -0
  592. package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
  593. package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
  594. package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
  595. package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
  596. package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
  597. package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
  598. package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
  599. package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +201 -0
  600. package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
  601. package/packages/piagent-core/runtime/workflows/task-intake.ts +276 -169
  602. package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
  603. package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
  604. package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
  605. package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
  606. package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
  607. package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
  608. package/packages/piagent-core/subagents/piagent-oracle.md +5 -3
  609. package/packages/piagent-core/subagents/piagent-planner.md +6 -4
  610. package/packages/piagent-core/subagents/piagent-reviewer.md +5 -3
  611. package/packages/piagent-core/subagents/piagent-scout.md +4 -2
  612. package/packages/piagent-core/subagents/piagent-worker.md +6 -4
  613. package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-CxeZqTte.js → MarkdownMessage-DGk34zJT.js} +1 -1
  614. package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
  615. package/packages/piagent-webui/dist/client/assets/index-DiKRmCEP.js +176 -0
  616. package/packages/piagent-webui/dist/client/index.html +2 -2
  617. package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
  618. package/packages/piagent-webui/extension/piagent-webui.ts +13 -3
  619. package/packages/piagent-webui/extension/review-controller.ts +29 -6
  620. package/packages/piagent-webui/extension/same-session-bridge.ts +7 -8
  621. package/packages/piagent-webui/extension/session-stream-adapter.ts +88 -16
  622. package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
  623. package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
  624. package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
  625. package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
  626. package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
  627. package/packages/piagent-webui/gateway/gateway-session-stream.ts +170 -32
  628. package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
  629. package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
  630. package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
  631. package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
  632. package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
  633. package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
  634. package/packages/piagent-webui/gateway/session-inspection-registry.ts +11 -2
  635. package/packages/piagent-webui/gateway/session-operation-lifecycle.ts +237 -0
  636. package/packages/piagent-webui/gateway/session-operation-retry-control.ts +38 -0
  637. package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
  638. package/packages/piagent-webui/gateway/session-operation-watchdog.ts +9 -2
  639. package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
  640. package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
  641. package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
  642. package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
  643. package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
  644. package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +118 -117
  645. package/packages/piagent-webui/ownership/session-lease-store.ts +5 -2
  646. package/packages/piagent-webui/server/core-inspection-provider.ts +75 -15
  647. package/packages/piagent-webui/server/loopback-server.ts +9 -7
  648. package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
  649. package/packages/piagent-webui/server/transcript-projection.ts +150 -11
  650. package/packages/piagent-webui/shared/message-correlation.ts +25 -0
  651. package/schemas/approved-host-contracts.schema.json +973 -0
  652. package/schemas/benchmark-attempt-outcome.schema.json +303 -0
  653. package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
  654. package/schemas/codex-relative-efficiency-v1.schema.json +155 -0
  655. package/schemas/composite-criterion.schema.json +788 -0
  656. package/schemas/contract-family-library.schema.json +28 -0
  657. package/schemas/contract-selection-recipe.schema.json +31 -0
  658. package/schemas/failure-evidence.schema.json +2 -2
  659. package/schemas/helper-request.schema.json +16 -4
  660. package/schemas/host-contract-plan.schema.json +25 -0
  661. package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
  662. package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
  663. package/schemas/piagent-webui/runtime-event-v2.schema.json +46 -5
  664. package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
  665. package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
  666. package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
  667. package/schemas/piagent-webui/subagent-tree-v1.schema.json +11 -1
  668. package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
  669. package/schemas/project-profile.schema.json +1 -1
  670. package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
  671. package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
  672. package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
  673. package/schemas/public-contracts/revocation-v1.schema.json +87 -0
  674. package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
  675. package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
  676. package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
  677. package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
  678. package/schemas/task-contract.schema.json +45 -0
  679. package/scripts/approve-independent-verification.mjs +57 -0
  680. package/scripts/benchmark-arm-observer.mjs +320 -0
  681. package/scripts/benchmark-budget-ipc.mjs +34 -0
  682. package/scripts/benchmark-budget-launcher.mjs +91 -0
  683. package/scripts/benchmark-budget-processes.mjs +133 -0
  684. package/scripts/benchmark-budget-publication.mjs +228 -0
  685. package/scripts/benchmark-budget-runtime.mjs +136 -0
  686. package/scripts/benchmark-candidate-ablation.mjs +265 -0
  687. package/scripts/benchmark-codex-journey.mjs +406 -0
  688. package/scripts/benchmark-independent-verification.mjs +628 -0
  689. package/scripts/benchmark-journey-outcome.mjs +92 -0
  690. package/scripts/benchmark-production-v3-human-review.mjs +192 -0
  691. package/scripts/benchmark-public-input-catalog.mjs +158 -0
  692. package/scripts/benchmark-registered-plan-author.mjs +355 -0
  693. package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
  694. package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
  695. package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
  696. package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
  697. package/scripts/benchmark-runner-completion.mjs +133 -0
  698. package/scripts/benchmark-runner-configuration.mjs +159 -0
  699. package/scripts/benchmark-runner-core.mjs +224 -387
  700. package/scripts/benchmark-runner-finalization.mjs +144 -27
  701. package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
  702. package/scripts/benchmark-runner-schedule.mjs +256 -0
  703. package/scripts/benchmark-runner-support.mjs +136 -4
  704. package/scripts/benchmark-runner.mjs +95 -18
  705. package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
  706. package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
  707. package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
  708. package/scripts/benchmark-scoped-pi-router.mjs +196 -0
  709. package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
  710. package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
  711. package/scripts/benchmark-scoped-session-custody.mjs +599 -0
  712. package/scripts/benchmark-scoped-session-factory.mjs +342 -0
  713. package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
  714. package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
  715. package/scripts/benchmark-session-evaluator.mjs +169 -0
  716. package/scripts/benchmark-session-provider-boundary.mjs +62 -0
  717. package/scripts/benchmark-session-usage.mjs +28 -0
  718. package/scripts/benchmark-session.mjs +343 -50
  719. package/scripts/benchmark-webui-journey.mjs +678 -0
  720. package/scripts/check-public-wording.mjs +46 -0
  721. package/scripts/configure-subagents.sh +42 -34
  722. package/scripts/fs4-readiness-evaluation.mjs +7 -1
  723. package/scripts/helper-evaluation.mjs +1 -1
  724. package/scripts/phase-tools-evaluation.mjs +1 -0
  725. package/scripts/pi-auto.sh +7 -1
  726. package/scripts/pi-usage-history.mjs +201 -11
  727. package/scripts/piagent-cli.mjs +8 -0
  728. package/scripts/private-holdout-readiness.mjs +7 -1
  729. package/scripts/public-evaluation-exposure.mjs +140 -0
  730. package/scripts/select-independent-verification.mjs +62 -0
  731. package/scripts/team-doctor.sh +1 -1
  732. package/scripts/typescript-loader.mjs +135 -1
  733. package/scripts/update-global.mjs +6 -7
  734. package/scripts/verify-local.sh +41 -55
  735. package/scripts/verify-release-identity.mjs +5 -3
  736. package/packages/piagent-webui/dist/client/assets/index-DynIMe8I.js +0 -176
@@ -0,0 +1,942 @@
1
+ import { StringDecoder } from "node:string_decoder";
2
+
3
+ import { rounded } from "./benchmark-statistics.js";
4
+ import {
5
+ BENCHMARK_EXECUTION_COUNTER_FIELDS as EXECUTION_COUNTER_FIELDS,
6
+ BENCHMARK_TIMING_PHASES as PHASES
7
+ } from "./benchmark-timing-contract.js";
8
+
9
+ export {
10
+ canonicalBenchmarkTimingDiagnostics,
11
+ summarizeBenchmarkTimingDiagnostics,
12
+ validBenchmarkTimingDiagnostics
13
+ } from "./benchmark-timing-contract.js";
14
+
15
+ const CODEX_MODEL_OUTPUT_ITEMS = new Set(["agent_message", "reasoning"]);
16
+ const CODEX_CONTROL_ITEMS = new Set(["todo_list"]);
17
+ const CODEX_COMPLETED_ONLY_CONTROL_ITEMS = new Set(["error"]);
18
+ const CODEX_COMPLETED_ONLY_TOOL_ITEMS = new Set(["file_change"]);
19
+ const CODEX_TIMED_TOOL_ITEMS = new Set(["command_execution", "mcp_tool_call", "collab_tool_call", "web_search"]);
20
+ const CODEX_COMMAND_TERMINAL_STATUSES = new Set(["completed", "failed", "declined"]);
21
+ const CODEX_TOOL_TERMINAL_STATUSES = new Set(["completed", "failed"]);
22
+ const CODEX_COLLAB_TOOLS = new Set(["spawn_agent", "send_input", "wait", "close_agent"]);
23
+ const CODEX_COLLAB_AGENT_STATUSES = new Set([
24
+ "pending_init", "running", "interrupted", "completed", "errored", "shutdown", "not_found"
25
+ ]);
26
+ const CODEX_USAGE_KEYS = ["input_tokens", "cached_input_tokens", "output_tokens", "reasoning_output_tokens"];
27
+ const PI_CONTROL_EVENTS = new Set([
28
+ "queue_update", "compaction_start", "entry_appended", "session_info_changed",
29
+ "thinking_level_changed", "compaction_end", "auto_retry_start", "auto_retry_end",
30
+ "summarization_retry_scheduled", "summarization_retry_attempt_start", "summarization_retry_finished",
31
+ "bash_execution_update"
32
+ ]);
33
+ const PI_MESSAGE_ROLES = new Set(["user", "assistant", "toolResult", "custom"]);
34
+ const SURFACES = new Set(["raw-pi", "piagent", "codex-cli"]);
35
+ const MAX_JSONL_BYTES = 64 * 1024 * 1024;
36
+ const MAX_TIMING_REPLAY_BYTES = 4 * 1024 * 1024;
37
+
38
+ function plainObject(value) {
39
+ return Boolean(value) && typeof value === "object" && !Array.isArray(value);
40
+ }
41
+
42
+ function stableValue(value) {
43
+ if (Array.isArray(value)) return value.map(stableValue);
44
+ if (!plainObject(value)) return value;
45
+ return Object.fromEntries(Object.keys(value).sort().map((key) => [key, stableValue(value[key])]));
46
+ }
47
+
48
+ function executionFingerprint(name, args) {
49
+ try { return JSON.stringify([name, stableValue(args)]); }
50
+ catch { return String(name); }
51
+ }
52
+
53
+ function piSubagentAttempt(toolName) {
54
+ const name = String(toolName ?? "").toLowerCase();
55
+ return ["spawn_agent", "create_agent", "delegate_task", "subagent"].includes(name);
56
+ }
57
+
58
+ function codexExecutionIdentity(item) {
59
+ if (item.type === "command_execution") return item.command;
60
+ if (item.type === "mcp_tool_call") return [item.server, item.tool, item.arguments];
61
+ if (item.type === "collab_tool_call") return [item.tool, item.prompt];
62
+ if (item.type === "web_search") return item.action ?? item.query;
63
+ return item.type;
64
+ }
65
+
66
+ function structuredStatus(value) {
67
+ const status = [value?.status, value?.details?.status, value?.result?.status]
68
+ .find((candidate) => typeof candidate === "string");
69
+ return status?.toLowerCase() ?? "";
70
+ }
71
+
72
+ function emptyExecution(source) {
73
+ return {
74
+ schemaVersion: 1,
75
+ source,
76
+ completeness: {
77
+ providerAttempts: "unavailable",
78
+ tools: "unavailable",
79
+ retries: "unavailable",
80
+ compactions: "unavailable",
81
+ subagents: "unavailable"
82
+ },
83
+ ...Object.fromEntries(EXECUTION_COUNTER_FIELDS.map((field) => [field, 0]))
84
+ };
85
+ }
86
+
87
+ function finiteNonnegative(value) {
88
+ return Number.isFinite(value) && value >= 0;
89
+ }
90
+
91
+ function phase(status, seconds, reason, boundary) {
92
+ return { status, seconds: status === "available" ? rounded(seconds, 6) : null, reason, boundary };
93
+ }
94
+
95
+ function intervalDuration(intervals) {
96
+ if (intervals.length === 0) return 0;
97
+ const ordered = intervals.slice().sort((left, right) => left[0] - right[0] || left[1] - right[1]);
98
+ let total = 0;
99
+ let [start, end] = ordered[0];
100
+ for (const [nextStart, nextEnd] of ordered.slice(1)) {
101
+ if (nextStart <= end) end = Math.max(end, nextEnd);
102
+ else { total += end - start; start = nextStart; end = nextEnd; }
103
+ }
104
+ return total + end - start;
105
+ }
106
+
107
+ function addInterval(state, target, start, end) {
108
+ if (!finiteNonnegative(start) || !finiteNonnegative(end) || end < start) {
109
+ state.invalidClockBoundaries += 1;
110
+ return false;
111
+ }
112
+ if (end === start) {
113
+ state.coalescedBoundaries += 1;
114
+ return false;
115
+ }
116
+ target.push([start, end]);
117
+ return true;
118
+ }
119
+
120
+ function safeId(value) {
121
+ return typeof value === "string" && value.length > 0 && value.length <= 1_000 ? value : null;
122
+ }
123
+
124
+ function sameIdentityMaps(left, right) {
125
+ if (left.size !== right.size) return false;
126
+ for (const [id, name] of left) if (right.get(id) !== name) return false;
127
+ return true;
128
+ }
129
+
130
+ function piToolResultIdentity(message) {
131
+ if (!plainObject(message) || message.role !== "toolResult") return null;
132
+ const id = safeId(message.toolCallId);
133
+ return id && typeof message.toolName === "string" && message.toolName ? { id, name: message.toolName } : null;
134
+ }
135
+
136
+ function validPiCustomMessage(message) {
137
+ return typeof message.customType === "string" && message.customType.length > 0
138
+ && (typeof message.content === "string" || Array.isArray(message.content))
139
+ && typeof message.display === "boolean" && finiteNonnegative(message.timestamp);
140
+ }
141
+
142
+ function invalidateEvent(state) {
143
+ state.malformedLines += 1;
144
+ }
145
+
146
+ function validPiControlEvent(event) {
147
+ if (event.type === "summarization_retry_finished") return true;
148
+ if (event.type === "queue_update") return Array.isArray(event.steering) && Array.isArray(event.followUp);
149
+ if (event.type === "compaction_start") return ["manual", "threshold", "overflow"].includes(event.reason);
150
+ if (event.type === "entry_appended") return plainObject(event.entry);
151
+ if (event.type === "session_info_changed") return event.name === undefined || typeof event.name === "string";
152
+ if (event.type === "thinking_level_changed") return typeof event.level === "string" && event.level.length > 0;
153
+ if (event.type === "compaction_end") return ["manual", "threshold", "overflow"].includes(event.reason)
154
+ && typeof event.aborted === "boolean" && typeof event.willRetry === "boolean"
155
+ && (event.result === undefined || event.result === null || plainObject(event.result));
156
+ if (event.type === "auto_retry_start" || event.type === "summarization_retry_scheduled") {
157
+ return Number.isSafeInteger(event.attempt) && event.attempt > 0
158
+ && Number.isSafeInteger(event.maxAttempts) && event.maxAttempts > 0
159
+ && finiteNonnegative(event.delayMs) && typeof event.errorMessage === "string";
160
+ }
161
+ if (event.type === "auto_retry_end") return typeof event.success === "boolean"
162
+ && Number.isSafeInteger(event.attempt) && event.attempt > 0
163
+ && (event.finalError === undefined || typeof event.finalError === "string");
164
+ if (event.type === "summarization_retry_attempt_start") return ["branchSummary", "compaction"].includes(event.source)
165
+ && (event.source !== "compaction" || ["manual", "threshold", "overflow"].includes(event.reason));
166
+ if (event.type === "bash_execution_update") return (event.id === undefined || safeId(event.id)) && typeof event.delta === "string";
167
+ return false;
168
+ }
169
+
170
+ function consumePiEvent(state, event, at) {
171
+ if (state.agentSettled) {
172
+ invalidateEvent(state);
173
+ return;
174
+ }
175
+ if (event.type === "session") {
176
+ if (state.jsonEvents !== 1 || state.sessionSeen || !Number.isSafeInteger(event.version) || event.version <= 0 || !safeId(event.id)
177
+ || typeof event.timestamp !== "string" || !event.timestamp || typeof event.cwd !== "string" || !event.cwd) invalidateEvent(state);
178
+ else {
179
+ state.sessionSeen = true;
180
+ state.sessionHeaders += 1;
181
+ }
182
+ return;
183
+ }
184
+ if (event.type === "agent_start") {
185
+ if (!state.sessionSeen || state.agentCycleOpen || state.turnOpen || state.modelTurnStart !== null
186
+ || state.openMessageRole !== null || state.openTools.size > 0) invalidateEvent(state);
187
+ else {
188
+ state.agentStarted = true;
189
+ state.agentCycleOpen = true;
190
+ state.agentEnded = false;
191
+ state.lastAgentEndWillRetry = null;
192
+ state.agentStarts += 1;
193
+ }
194
+ return;
195
+ }
196
+ if (event.type === "turn_start") {
197
+ if (!state.agentCycleOpen || state.turnOpen || state.modelTurnStart !== null) {
198
+ invalidateEvent(state);
199
+ return;
200
+ }
201
+ state.turnStarts += 1;
202
+ state.execution.providerStartedAttempts += 1;
203
+ if (state.startupBoundary === null) state.startupBoundary = at;
204
+ state.turnOpen = true;
205
+ state.assistantSeenInTurn = false;
206
+ state.assistantMessageEnded = false;
207
+ state.turnToolStarts = 0;
208
+ state.turnToolCompletions = 0;
209
+ state.turnToolResultMessages = 0;
210
+ state.completedTurnTools.clear();
211
+ state.turnToolResultIdentities.clear();
212
+ state.openToolResultIdentity = null;
213
+ state.modelTurnStart = at;
214
+ return;
215
+ }
216
+ if (event.type === "message_start" || event.type === "message_end") {
217
+ if (!plainObject(event.message) || typeof event.message.role !== "string" || !state.turnOpen) {
218
+ invalidateEvent(state);
219
+ return;
220
+ }
221
+ const role = event.message.role;
222
+ if (!PI_MESSAGE_ROLES.has(role) || role === "custom" && !validPiCustomMessage(event.message)) { invalidateEvent(state); return; }
223
+ if (event.type === "message_start") {
224
+ if (state.openMessageRole !== null || (["user", "custom"].includes(role) && state.assistantSeenInTurn)
225
+ || (role === "toolResult" && (!state.assistantMessageEnded || state.openTools.size > 0))) {
226
+ invalidateEvent(state);
227
+ return;
228
+ }
229
+ if (role === "toolResult") {
230
+ const identity = piToolResultIdentity(event.message);
231
+ if (!identity || state.completedTurnTools.get(identity.id) !== identity.name
232
+ || state.turnToolResultIdentities.has(identity.id)) {
233
+ invalidateEvent(state);
234
+ return;
235
+ }
236
+ state.openToolResultIdentity = identity;
237
+ }
238
+ if (role === "custom") state.openCustomType = event.message.customType;
239
+ state.openMessageRole = role;
240
+ if (role !== "assistant") return;
241
+ if (state.assistantSeenInTurn || state.modelTurnStart === null) {
242
+ invalidateEvent(state);
243
+ return;
244
+ }
245
+ state.assistantSeenInTurn = true;
246
+ return;
247
+ }
248
+ if (state.openMessageRole !== role || role === "custom" && state.openCustomType !== event.message.customType) {
249
+ invalidateEvent(state);
250
+ return;
251
+ }
252
+ if (role === "toolResult") {
253
+ const identity = piToolResultIdentity(event.message);
254
+ if (!identity || state.openToolResultIdentity === null
255
+ || identity.id !== state.openToolResultIdentity.id || identity.name !== state.openToolResultIdentity.name) {
256
+ invalidateEvent(state);
257
+ return;
258
+ }
259
+ state.turnToolResultIdentities.set(identity.id, identity.name);
260
+ state.openToolResultIdentity = null;
261
+ }
262
+ state.openMessageRole = null;
263
+ if (role === "custom") state.openCustomType = null;
264
+ if (role === "user" || role === "custom") {
265
+ if (state.assistantSeenInTurn || state.modelTurnStart === null) invalidateEvent(state);
266
+ else state.modelTurnStart = at;
267
+ return;
268
+ }
269
+ if (role === "toolResult") {
270
+ if (!state.assistantMessageEnded || state.openTools.size > 0) invalidateEvent(state);
271
+ else state.turnToolResultMessages += 1;
272
+ }
273
+ if (role === "assistant") {
274
+ if (!state.assistantSeenInTurn || state.assistantMessageEnded || state.modelTurnStart === null) invalidateEvent(state);
275
+ else {
276
+ state.assistantBoundaries += 1;
277
+ if (addInterval(state, state.modelTurnIntervals, state.modelTurnStart, at)) state.matchedModelTurnIntervals += 1;
278
+ else state.modelTurnIntervalsInvalid = true;
279
+ state.modelTurnStart = null;
280
+ state.assistantMessageEnded = true;
281
+ }
282
+ }
283
+ return;
284
+ }
285
+ if (event.type === "message_update") {
286
+ if (!state.turnOpen || state.openMessageRole !== "assistant" || !plainObject(event.assistantMessageEvent)) invalidateEvent(state);
287
+ return;
288
+ }
289
+ if (event.type === "tool_execution_start") {
290
+ const id = safeId(event.toolCallId);
291
+ if (!state.turnOpen || !state.assistantMessageEnded || !id || typeof event.toolName !== "string" || !event.toolName
292
+ || state.openTools.has(id)) {
293
+ invalidateEvent(state);
294
+ return;
295
+ }
296
+ if (state.completedTurnTools.has(id)) {
297
+ invalidateEvent(state);
298
+ return;
299
+ }
300
+ state.toolStarts += 1;
301
+ state.execution.toolCalls += 1;
302
+ state.turnToolStarts += 1;
303
+ const subagentAttempt = piSubagentAttempt(event.toolName);
304
+ const fingerprint = executionFingerprint(event.toolName, event.args);
305
+ if (state.toolFingerprints.has(fingerprint)) state.execution.repeatedToolCalls += 1;
306
+ else state.toolFingerprints.add(fingerprint);
307
+ if (subagentAttempt) state.execution.subagentAttempts += 1;
308
+ state.openTools.set(id, { at, toolName: event.toolName, subagentAttempt });
309
+ return;
310
+ }
311
+ if (event.type === "tool_execution_end") {
312
+ const id = safeId(event.toolCallId);
313
+ if (!state.turnOpen || !state.assistantMessageEnded || !id || typeof event.toolName !== "string" || !event.toolName
314
+ || typeof event.isError !== "boolean") {
315
+ invalidateEvent(state);
316
+ return;
317
+ }
318
+ state.toolCompletions += 1;
319
+ state.execution.toolResults += 1;
320
+ state.turnToolCompletions += 1;
321
+ const started = state.openTools.get(id);
322
+ const status = structuredStatus(event.result);
323
+ const failed = event.isError === true || status === "failed" || status === "error";
324
+ if (failed) state.execution.toolFailures += 1;
325
+ if (status === "blocked") state.execution.blockedToolCalls += 1;
326
+ if (status === "declined") state.execution.declinedToolCalls += 1;
327
+ if (started?.subagentAttempt && (failed || status === "blocked" || status === "declined")) state.execution.subagentFailures += 1;
328
+ if (started === undefined || started.toolName !== event.toolName || state.completedTurnTools.has(id)) {
329
+ if (started !== undefined && started.toolName !== event.toolName) invalidateEvent(state);
330
+ state.unmatchedToolCompletions += 1;
331
+ }
332
+ else {
333
+ if (!addInterval(state, state.toolIntervals, started.at, at)) state.toolIntervalsInvalid = true;
334
+ state.matchedToolIntervals += 1;
335
+ state.openTools.delete(id);
336
+ state.completedTurnTools.set(id, event.toolName);
337
+ }
338
+ return;
339
+ }
340
+ if (event.type === "tool_execution_update") {
341
+ const id = safeId(event.toolCallId);
342
+ const started = id ? state.openTools.get(id) : undefined;
343
+ if (!state.turnOpen || !state.assistantMessageEnded || !id || started === undefined
344
+ || typeof event.toolName !== "string" || !event.toolName || started.toolName !== event.toolName) invalidateEvent(state);
345
+ return;
346
+ }
347
+ if (event.type === "turn_end") {
348
+ const receiptIdentities = new Map();
349
+ let receiptValid = Array.isArray(event.toolResults);
350
+ if (receiptValid) {
351
+ for (const result of event.toolResults) {
352
+ const identity = piToolResultIdentity(result);
353
+ if (!identity || receiptIdentities.has(identity.id)) { receiptValid = false; break; }
354
+ receiptIdentities.set(identity.id, identity.name);
355
+ }
356
+ }
357
+ if (!state.turnOpen || !state.assistantMessageEnded || !plainObject(event.message) || !Array.isArray(event.toolResults)
358
+ || event.message.role !== "assistant" || state.openMessageRole !== null
359
+ || state.openToolResultIdentity !== null || !receiptValid
360
+ || state.openTools.size > 0 || state.modelTurnStart !== null
361
+ || state.turnToolStarts !== state.turnToolCompletions
362
+ || state.turnToolCompletions !== state.turnToolResultMessages
363
+ || state.turnToolCompletions !== event.toolResults.length
364
+ || !sameIdentityMaps(state.completedTurnTools, state.turnToolResultIdentities)
365
+ || !sameIdentityMaps(state.completedTurnTools, receiptIdentities)) {
366
+ invalidateEvent(state);
367
+ return;
368
+ }
369
+ state.turnCompletions += 1;
370
+ state.turnOpen = false;
371
+ state.assistantSeenInTurn = false;
372
+ state.assistantMessageEnded = false;
373
+ state.turnToolStarts = 0;
374
+ state.turnToolCompletions = 0;
375
+ state.turnToolResultMessages = 0;
376
+ state.completedTurnTools.clear();
377
+ state.turnToolResultIdentities.clear();
378
+ state.openToolResultIdentity = null;
379
+ return;
380
+ }
381
+ if (event.type === "agent_end") {
382
+ if (!state.agentCycleOpen || typeof event.willRetry !== "boolean" || !Array.isArray(event.messages)
383
+ || state.turnOpen || state.modelTurnStart !== null || state.openMessageRole !== null || state.openTools.size > 0) invalidateEvent(state);
384
+ else {
385
+ state.agentCycleOpen = false;
386
+ state.agentEnded = true;
387
+ state.lastAgentEndWillRetry = event.willRetry;
388
+ state.agentEnds += 1;
389
+ }
390
+ return;
391
+ }
392
+ if (event.type === "agent_settled") {
393
+ if (!state.sessionSeen || !state.agentStarted || !state.agentEnded || state.agentCycleOpen
394
+ || state.lastAgentEndWillRetry !== false
395
+ || state.turnOpen || state.openMessageRole !== null || state.openTools.size > 0) invalidateEvent(state);
396
+ else {
397
+ state.agentSettled = true;
398
+ state.agentSettledEvents += 1;
399
+ }
400
+ return;
401
+ }
402
+ if (PI_CONTROL_EVENTS.has(event.type)) {
403
+ if (!validPiControlEvent(event)) invalidateEvent(state);
404
+ else if (event.type === "auto_retry_start") state.execution.explicitRetries += 1;
405
+ else if (event.type === "auto_retry_end" && event.success === false) state.execution.retryFailures += 1;
406
+ else if (event.type === "compaction_start") state.execution.compactions += 1;
407
+ else if (event.type === "compaction_end" && event.aborted === true) state.execution.abortedCompactions += 1;
408
+ else if (event.type === "summarization_retry_scheduled") state.execution.summarizationRetries += 1;
409
+ return;
410
+ }
411
+ invalidateEvent(state);
412
+ }
413
+
414
+ function codexItemKind(type) {
415
+ if (CODEX_MODEL_OUTPUT_ITEMS.has(type)) return "model-output";
416
+ if (CODEX_CONTROL_ITEMS.has(type)) return "control";
417
+ if (CODEX_COMPLETED_ONLY_CONTROL_ITEMS.has(type)) return "completed-only-control";
418
+ if (CODEX_COMPLETED_ONLY_TOOL_ITEMS.has(type)) return "completed-only-tool";
419
+ if (CODEX_TIMED_TOOL_ITEMS.has(type)) return "timed-tool";
420
+ return null;
421
+ }
422
+
423
+ function validCodexTodoList(item) {
424
+ return Array.isArray(item.items) && item.items.every((todo) => plainObject(todo)
425
+ && typeof todo.text === "string" && typeof todo.completed === "boolean");
426
+ }
427
+
428
+ function validCodexUsage(usage) {
429
+ return plainObject(usage) && CODEX_USAGE_KEYS.every((key) => Number.isSafeInteger(usage[key]) && usage[key] >= 0)
430
+ && (usage.cache_write_input_tokens === undefined
431
+ || Number.isSafeInteger(usage.cache_write_input_tokens) && usage.cache_write_input_tokens >= 0);
432
+ }
433
+
434
+ function validCodexMcpResult(result) {
435
+ return plainObject(result) && Array.isArray(result.content);
436
+ }
437
+
438
+ function validCodexCollabAgents(states) {
439
+ return plainObject(states) && Object.entries(states).every(([id, agent]) => safeId(id) && plainObject(agent)
440
+ && CODEX_COLLAB_AGENT_STATUSES.has(agent.status)
441
+ && (agent.message === null || typeof agent.message === "string"));
442
+ }
443
+
444
+ function validCodexWebSearchAction(action) {
445
+ if (!plainObject(action) || !["search", "open_page", "find_in_page", "other"].includes(action.type)) return false;
446
+ if (action.type === "search") return (action.query === undefined || typeof action.query === "string")
447
+ && (action.queries === undefined || Array.isArray(action.queries) && action.queries.every((query) => typeof query === "string"));
448
+ if (action.type === "open_page") return action.url === undefined || typeof action.url === "string";
449
+ if (action.type === "find_in_page") return (action.url === undefined || typeof action.url === "string")
450
+ && (action.pattern === undefined || typeof action.pattern === "string");
451
+ return true;
452
+ }
453
+
454
+ function validCodexItemForEvent(eventType, item) {
455
+ if (CODEX_MODEL_OUTPUT_ITEMS.has(item.type)) {
456
+ return eventType === "item.completed" && typeof item.text === "string"
457
+ && (item.type !== "reasoning" || item.text.trim().length > 0);
458
+ }
459
+ if (item.type === "todo_list") {
460
+ return ["item.started", "item.updated", "item.completed"].includes(eventType) && validCodexTodoList(item);
461
+ }
462
+ if (item.type === "error") {
463
+ return eventType === "item.completed" && typeof item.message === "string" && item.message.length > 0;
464
+ }
465
+ if (item.type === "file_change") {
466
+ return eventType === "item.completed" && ["completed", "failed"].includes(item.status)
467
+ && Array.isArray(item.changes) && item.changes.every((change) => plainObject(change)
468
+ && typeof change.path === "string" && ["add", "delete", "update"].includes(change.kind));
469
+ }
470
+ if (item.type === "command_execution") {
471
+ const statusValid = eventType === "item.started"
472
+ ? item.status === "in_progress"
473
+ : eventType === "item.completed" && CODEX_COMMAND_TERMINAL_STATUSES.has(item.status);
474
+ return statusValid && typeof item.command === "string"
475
+ && (eventType === "item.started"
476
+ ? (item.aggregated_output === undefined || typeof item.aggregated_output === "string")
477
+ && (item.exit_code === undefined || item.exit_code === null)
478
+ : typeof item.aggregated_output === "string" && (item.exit_code === null || Number.isSafeInteger(item.exit_code)));
479
+ }
480
+ if (item.type === "mcp_tool_call") {
481
+ const statusValid = eventType === "item.started"
482
+ ? item.status === "in_progress"
483
+ : eventType === "item.completed" && CODEX_TOOL_TERMINAL_STATUSES.has(item.status);
484
+ return statusValid && typeof item.server === "string" && item.server.length > 0
485
+ && typeof item.tool === "string" && item.tool.length > 0
486
+ && (eventType === "item.started" || Object.hasOwn(item, "arguments"))
487
+ && (item.result === undefined || item.result === null || validCodexMcpResult(item.result))
488
+ && (item.error === undefined || item.error === null
489
+ || plainObject(item.error) && typeof item.error.message === "string" && item.error.message.length > 0);
490
+ }
491
+ if (item.type === "collab_tool_call") {
492
+ const statusValid = eventType === "item.started"
493
+ ? item.status === "in_progress"
494
+ : eventType === "item.completed" && CODEX_TOOL_TERMINAL_STATUSES.has(item.status);
495
+ return statusValid && CODEX_COLLAB_TOOLS.has(item.tool) && Boolean(safeId(item.sender_thread_id))
496
+ && Array.isArray(item.receiver_thread_ids) && item.receiver_thread_ids.every((id) => safeId(id))
497
+ && (item.prompt === undefined || item.prompt === null || typeof item.prompt === "string")
498
+ && (eventType === "item.started"
499
+ ? item.agents_states === undefined || validCodexCollabAgents(item.agents_states)
500
+ : validCodexCollabAgents(item.agents_states));
501
+ }
502
+ return item.type === "web_search" && ["item.started", "item.completed"].includes(eventType) && Boolean(safeId(item.id))
503
+ && (eventType === "item.started"
504
+ ? (item.query === undefined || typeof item.query === "string")
505
+ && (item.action === undefined || validCodexWebSearchAction(item.action))
506
+ : typeof item.query === "string" && validCodexWebSearchAction(item.action));
507
+ }
508
+
509
+ function closeCodexModelTurnInterval(state, at) {
510
+ if (state.modelTurnStart === null) return;
511
+ if (addInterval(state, state.modelTurnIntervals, state.modelTurnStart, at)) state.matchedModelTurnIntervals += 1;
512
+ else state.modelTurnIntervalsInvalid = true;
513
+ state.modelTurnStart = null;
514
+ }
515
+
516
+ function consumeCodexEvent(state, event, at) {
517
+ if (state.terminalTurnSeen) {
518
+ invalidateEvent(state);
519
+ return;
520
+ }
521
+ if (event.type === "thread.started") {
522
+ if (state.jsonEvents !== 1 || !safeId(event.thread_id) || state.threadSeen || state.turnOpen) invalidateEvent(state);
523
+ else {
524
+ state.threadSeen = true;
525
+ state.threadStarts += 1;
526
+ }
527
+ return;
528
+ }
529
+ if (event.type === "turn.started") {
530
+ if (!state.threadSeen || state.terminalTurnSeen || state.turnOpen || state.modelTurnStart !== null) {
531
+ invalidateEvent(state);
532
+ return;
533
+ }
534
+ state.turnStarts += 1;
535
+ state.execution.providerStartedAttempts += 1;
536
+ if (state.startupBoundary === null) state.startupBoundary = at;
537
+ state.turnOpen = true;
538
+ state.modelTurnStart = at;
539
+ return;
540
+ }
541
+ if (event.type === "turn.failed") {
542
+ if (!state.turnOpen || !plainObject(event.error) || typeof event.error.message !== "string" || !event.error.message) {
543
+ invalidateEvent(state);
544
+ return;
545
+ }
546
+ if (state.modelTurnStart !== null) {
547
+ state.unmatchedModelTurnStarts += 1;
548
+ state.modelTurnStart = null;
549
+ }
550
+ state.unmatchedToolStarts += state.openTools.size;
551
+ state.openTools.clear();
552
+ if (state.openTodoItemId !== null) state.openTodoItemId = null;
553
+ state.turnOpen = false;
554
+ state.terminalTurnSeen = true;
555
+ return;
556
+ }
557
+ if (event.type === "error") {
558
+ if (!state.threadSeen || typeof event.message !== "string" || !event.message) {
559
+ invalidateEvent(state);
560
+ return;
561
+ }
562
+ state.providerErrorSeen = true;
563
+ return;
564
+ }
565
+ if (event.type === "turn.completed") {
566
+ if (!state.turnOpen || !validCodexUsage(event.usage) || state.openTools.size > 0 || state.openTodoItemId !== null) {
567
+ invalidateEvent(state);
568
+ return;
569
+ }
570
+ state.turnCompletions += 1;
571
+ closeCodexModelTurnInterval(state, at);
572
+ state.turnOpen = false;
573
+ state.terminalTurnSeen = true;
574
+ return;
575
+ }
576
+ if (event.type === "item.updated") {
577
+ if (!state.turnOpen || !plainObject(event.item) || !safeId(event.item.id)
578
+ || event.item.type !== "todo_list" || !validCodexItemForEvent(event.type, event.item)
579
+ || state.openTodoItemId !== event.item.id || state.seenCodexItemIds.get(event.item.id) !== "todo_list") invalidateEvent(state);
580
+ return;
581
+ }
582
+ if (!["item.started", "item.completed"].includes(event.type)) {
583
+ invalidateEvent(state);
584
+ return;
585
+ }
586
+ if (!plainObject(event.item) || !safeId(event.item.id) || typeof event.item.type !== "string" || !state.turnOpen) {
587
+ invalidateEvent(state);
588
+ return;
589
+ }
590
+ const kind = codexItemKind(event.item.type);
591
+ if (!kind || !validCodexItemForEvent(event.type, event.item)) {
592
+ invalidateEvent(state);
593
+ return;
594
+ }
595
+ if (event.type === "item.started") {
596
+ if (state.seenCodexItemIds.has(event.item.id)) {
597
+ invalidateEvent(state);
598
+ return;
599
+ }
600
+ if (kind === "control") {
601
+ if (state.openTodoItemId !== null) invalidateEvent(state);
602
+ else {
603
+ state.seenCodexItemIds.set(event.item.id, event.item.type);
604
+ state.openTodoItemId = event.item.id;
605
+ }
606
+ return;
607
+ }
608
+ const id = safeId(event.item.id);
609
+ if (!id || state.openTools.has(id)) { invalidateEvent(state); return; }
610
+ state.seenCodexItemIds.set(id, event.item.type);
611
+ closeCodexModelTurnInterval(state, at);
612
+ state.toolStarts += 1;
613
+ state.execution.toolCalls += 1;
614
+ const fingerprint = executionFingerprint(event.item.type, codexExecutionIdentity(event.item));
615
+ if (state.toolFingerprints.has(fingerprint)) state.execution.repeatedToolCalls += 1;
616
+ else state.toolFingerprints.add(fingerprint);
617
+ const subagentAttempt = event.item.type === "collab_tool_call" && event.item.tool === "spawn_agent";
618
+ if (subagentAttempt) state.execution.subagentAttempts += 1;
619
+ state.openTools.set(id, { at, type: event.item.type, subagentAttempt });
620
+ return;
621
+ }
622
+ if (kind === "control") {
623
+ if (state.openTodoItemId !== event.item.id || state.seenCodexItemIds.get(event.item.id) !== event.item.type) invalidateEvent(state);
624
+ else state.openTodoItemId = null;
625
+ return;
626
+ }
627
+ if (kind === "completed-only-control" || kind === "model-output" || kind === "completed-only-tool") {
628
+ if (state.seenCodexItemIds.has(event.item.id)) {
629
+ invalidateEvent(state);
630
+ return;
631
+ }
632
+ state.seenCodexItemIds.set(event.item.id, event.item.type);
633
+ }
634
+ if (kind === "completed-only-control") return;
635
+ if (kind === "completed-only-tool") {
636
+ state.toolCompletions += 1;
637
+ state.execution.toolCalls += 1;
638
+ state.execution.toolResults += 1;
639
+ if (event.item.status === "failed") state.execution.toolFailures += 1;
640
+ state.unmatchedToolCompletions += 1;
641
+ state.modelTurnIntervalsInvalid = true;
642
+ state.toolIntervalsInvalid = true;
643
+ return;
644
+ }
645
+ if (kind === "model-output") return;
646
+ state.toolCompletions += 1;
647
+ state.execution.toolResults += 1;
648
+ const id = safeId(event.item.id);
649
+ if (!id) {
650
+ invalidateEvent(state);
651
+ state.unmatchedToolCompletions += 1;
652
+ return;
653
+ }
654
+ const seenType = state.seenCodexItemIds.get(id);
655
+ const started = state.openTools.get(id);
656
+ const status = typeof event.item.status === "string" ? event.item.status.toLowerCase() : "";
657
+ const failed = status === "failed" || event.item.type === "command_execution"
658
+ && Number.isSafeInteger(event.item.exit_code) && event.item.exit_code !== 0;
659
+ if (failed) state.execution.toolFailures += 1;
660
+ if (status === "blocked") state.execution.blockedToolCalls += 1;
661
+ if (status === "declined") state.execution.declinedToolCalls += 1;
662
+ if (started?.subagentAttempt && (failed || status === "blocked" || status === "declined")) state.execution.subagentFailures += 1;
663
+ let matched = false;
664
+ if (started === undefined) {
665
+ if (seenType !== undefined) invalidateEvent(state);
666
+ else state.seenCodexItemIds.set(id, event.item.type);
667
+ state.unmatchedToolCompletions += 1;
668
+ state.modelTurnIntervalsInvalid = true;
669
+ state.toolIntervalsInvalid = true;
670
+ }
671
+ else if (seenType !== event.item.type || started.type !== event.item.type) {
672
+ invalidateEvent(state);
673
+ state.unmatchedToolCompletions += 1;
674
+ state.modelTurnIntervalsInvalid = true;
675
+ state.toolIntervalsInvalid = true;
676
+ }
677
+ else {
678
+ if (!addInterval(state, state.toolIntervals, started.at, at)) state.toolIntervalsInvalid = true;
679
+ state.matchedToolIntervals += 1;
680
+ state.openTools.delete(id);
681
+ matched = true;
682
+ }
683
+ if (matched && state.openTools.size === 0) {
684
+ if (state.modelTurnStart !== null) state.unmatchedModelTurnStarts += 1;
685
+ state.modelTurnStart = at;
686
+ }
687
+ }
688
+
689
+ function unavailableReason(state, kind) {
690
+ if (state.malformedLines > 0 || state.invalidObservationClock || state.providerErrorSeen) return "invalid-event-stream";
691
+ if (kind === "startup" && state.startupBoundary === null) return "missing-turn-start-boundary";
692
+ if (kind === "modelTurn" && (state.modelTurnIntervalsInvalid || state.unmatchedModelTurnStarts > 0 || state.turnStarts === 0)) return "incomplete-model-turn-boundaries";
693
+ if (kind === "tool" && (state.toolIntervalsInvalid || state.unmatchedToolStarts > 0 || state.unmatchedToolCompletions > 0 || state.openTools.size > 0)) return "incomplete-tool-boundaries";
694
+ return "unavailable";
695
+ }
696
+
697
+ export function unavailableBenchmarkTimingDiagnostics(surface, durationSeconds) {
698
+ if (!SURFACES.has(surface)) throw new Error(`Unsupported benchmark timing surface: ${surface}`);
699
+ const processDurationSeconds = finiteNonnegative(durationSeconds) ? rounded(durationSeconds, 6) : null;
700
+ return {
701
+ schemaVersion: 2,
702
+ authority: "observational-only",
703
+ gateImpact: "none",
704
+ clock: "benchmark-process-monotonic-receipt",
705
+ surface,
706
+ status: "unavailable",
707
+ processDurationSeconds,
708
+ phases: {
709
+ processStartup: phase("unavailable", null, "invalid-event-stream", "process-start-to-turn-start"),
710
+ modelTurnWait: phase("unavailable", null, "invalid-event-stream", "request-ready-to-next-model-output-boundary"),
711
+ toolExecution: phase("unavailable", null, "invalid-event-stream", "matched-tool-start-to-tool-completion"),
712
+ other: phase("unavailable", null, "one-or-more-phase-boundaries-unavailable", "process-duration-minus-available-observed-phases")
713
+ },
714
+ observations: {
715
+ jsonEvents: 0, sessionHeaders: 0, agentStarts: 0, agentEnds: 0, agentSettled: 0, threadStarts: 0,
716
+ turnStarts: 0, turnCompletions: 0, matchedModelTurnIntervals: 0,
717
+ toolStarts: 0, toolCompletions: 0, matchedToolIntervals: 0,
718
+ unmatchedToolStarts: 0, unmatchedToolCompletions: 0, unmatchedModelTurnStarts: 0,
719
+ malformedLines: 1, invalidClockBoundaries: 0, coalescedBoundaries: 0
720
+ },
721
+ execution: emptyExecution("stdout-jsonl-lifecycle"),
722
+ privacy: { rawPayloadStored: false, promptsStored: false, commandsStored: false, pathsStored: false, identifiersStored: false }
723
+ };
724
+ }
725
+
726
+ function finish(state, durationSeconds) {
727
+ const durationAvailable = finiteNonnegative(durationSeconds);
728
+ if (durationAvailable && state.lastObservationAt !== null && state.lastObservationAt > durationSeconds + 1e-6) {
729
+ state.invalidObservationClock = true;
730
+ state.invalidClockBoundaries += 1;
731
+ }
732
+ const lifecycleComplete = state.surface === "codex-cli"
733
+ ? state.threadSeen && state.terminalTurnSeen && state.turnStarts > 0
734
+ && state.turnCompletions === state.turnStarts && !state.turnOpen
735
+ : state.turnStarts > 0 && state.turnCompletions === state.turnStarts && state.assistantBoundaries === state.turnStarts
736
+ && state.sessionSeen && state.agentStarted && state.agentEnded && state.agentSettled
737
+ && !state.agentCycleOpen && state.lastAgentEndWillRetry === false && state.agentStarts === state.agentEnds
738
+ && !state.turnOpen && state.modelTurnStart === null;
739
+ if (state.modelTurnStart !== null) state.unmatchedModelTurnStarts += 1;
740
+ state.unmatchedToolStarts += state.openTools.size;
741
+ const startupAvailable = durationAvailable && state.malformedLines === 0 && !state.invalidObservationClock
742
+ && !state.providerErrorSeen && state.startupBoundary !== null && state.startupBoundary <= durationSeconds;
743
+ const modelTurnAvailable = durationAvailable && state.malformedLines === 0 && !state.invalidObservationClock
744
+ && !state.providerErrorSeen && lifecycleComplete && state.unmatchedModelTurnStarts === 0 && !state.modelTurnIntervalsInvalid;
745
+ const toolAvailable = durationAvailable && state.malformedLines === 0 && !state.invalidObservationClock
746
+ && !state.providerErrorSeen && lifecycleComplete && state.unmatchedToolStarts === 0
747
+ && state.unmatchedToolCompletions === 0 && !state.toolIntervalsInvalid;
748
+ const startupSeconds = startupAvailable ? state.startupBoundary : null;
749
+ const modelTurnSeconds = modelTurnAvailable ? intervalDuration(state.modelTurnIntervals) : null;
750
+ const toolSeconds = toolAvailable ? intervalDuration(state.toolIntervals) : null;
751
+ const attributed = startupAvailable && modelTurnAvailable && toolAvailable
752
+ ? startupSeconds + modelTurnSeconds + toolSeconds
753
+ : null;
754
+ const otherAvailable = Number.isFinite(attributed) && attributed <= durationSeconds + 1e-6;
755
+ const otherSeconds = otherAvailable ? Math.max(0, durationSeconds - attributed) : null;
756
+ const availablePhases = [startupAvailable, modelTurnAvailable, toolAvailable, otherAvailable].filter(Boolean).length;
757
+ const status = availablePhases === PHASES.length ? "complete" : availablePhases > 0 ? "partial" : "unavailable";
758
+ const executionComplete = state.malformedLines === 0 && !state.providerErrorSeen && lifecycleComplete;
759
+ state.execution.completeness.providerAttempts = executionComplete
760
+ ? state.surface === "codex-cli" ? "lower-bound" : "exact"
761
+ : "partial";
762
+ state.execution.completeness.tools = executionComplete && state.unmatchedToolStarts === 0
763
+ && state.unmatchedToolCompletions === 0 ? "exact" : "partial";
764
+ state.execution.completeness.retries = executionComplete && state.surface !== "codex-cli" ? "exact" : "unavailable";
765
+ state.execution.completeness.compactions = executionComplete && state.surface !== "codex-cli" ? "exact" : "unavailable";
766
+ state.execution.completeness.subagents = state.execution.completeness.tools;
767
+ return {
768
+ schemaVersion: 2,
769
+ authority: "observational-only",
770
+ gateImpact: "none",
771
+ clock: "benchmark-process-monotonic-receipt",
772
+ surface: state.surface,
773
+ status,
774
+ processDurationSeconds: durationAvailable ? rounded(durationSeconds, 6) : null,
775
+ phases: {
776
+ processStartup: startupAvailable
777
+ ? phase("available", startupSeconds, "observed-boundaries", "process-start-to-turn-start")
778
+ : phase("unavailable", null, unavailableReason(state, "startup"), "process-start-to-turn-start"),
779
+ modelTurnWait: modelTurnAvailable
780
+ ? phase("available", modelTurnSeconds, "observed-boundaries", "request-ready-to-next-model-output-boundary")
781
+ : phase("unavailable", null, unavailableReason(state, "modelTurn"), "request-ready-to-next-model-output-boundary"),
782
+ toolExecution: toolAvailable
783
+ ? phase("available", toolSeconds, "observed-boundaries", "matched-tool-start-to-tool-completion")
784
+ : phase("unavailable", null, unavailableReason(state, "tool"), "matched-tool-start-to-tool-completion"),
785
+ other: otherAvailable
786
+ ? phase("available", otherSeconds, "exact-unattributed-remainder", "process-duration-minus-available-observed-phases")
787
+ : phase("unavailable", null, "one-or-more-phase-boundaries-unavailable", "process-duration-minus-available-observed-phases")
788
+ },
789
+ observations: {
790
+ jsonEvents: state.jsonEvents,
791
+ sessionHeaders: state.sessionHeaders,
792
+ agentStarts: state.agentStarts,
793
+ agentEnds: state.agentEnds,
794
+ agentSettled: state.agentSettledEvents,
795
+ threadStarts: state.threadStarts,
796
+ turnStarts: state.turnStarts,
797
+ turnCompletions: state.turnCompletions,
798
+ matchedModelTurnIntervals: state.matchedModelTurnIntervals,
799
+ toolStarts: state.toolStarts,
800
+ toolCompletions: state.toolCompletions,
801
+ matchedToolIntervals: state.matchedToolIntervals,
802
+ unmatchedToolStarts: state.unmatchedToolStarts,
803
+ unmatchedToolCompletions: state.unmatchedToolCompletions,
804
+ unmatchedModelTurnStarts: state.unmatchedModelTurnStarts,
805
+ malformedLines: state.malformedLines,
806
+ invalidClockBoundaries: state.invalidClockBoundaries,
807
+ coalescedBoundaries: state.coalescedBoundaries
808
+ },
809
+ execution: state.execution,
810
+ privacy: {
811
+ rawPayloadStored: false,
812
+ promptsStored: false,
813
+ commandsStored: false,
814
+ pathsStored: false,
815
+ identifiersStored: false
816
+ }
817
+ };
818
+ }
819
+
820
+ export function createBenchmarkTimingCollector({ surface }) {
821
+ if (!SURFACES.has(surface)) throw new Error(`Unsupported benchmark timing surface: ${surface}`);
822
+ const state = {
823
+ surface,
824
+ jsonEvents: 0, sessionHeaders: 0, agentStarts: 0, agentEnds: 0, agentSettledEvents: 0,
825
+ threadStarts: 0, turnStarts: 0, turnCompletions: 0, assistantBoundaries: 0,
826
+ toolStarts: 0, toolCompletions: 0, matchedToolIntervals: 0, matchedModelTurnIntervals: 0,
827
+ unmatchedToolStarts: 0, unmatchedToolCompletions: 0, unmatchedModelTurnStarts: 0,
828
+ malformedLines: 0, invalidClockBoundaries: 0, coalescedBoundaries: 0, invalidObservationClock: false,
829
+ startupBoundary: null, modelTurnStart: null,
830
+ modelTurnIntervals: [],
831
+ modelTurnIntervalsInvalid: false,
832
+ toolIntervals: [],
833
+ toolIntervalsInvalid: false,
834
+ openTools: new Map(),
835
+ toolFingerprints: new Set(),
836
+ execution: emptyExecution("stdout-jsonl-lifecycle"),
837
+ completedTurnTools: new Map(),
838
+ turnToolResultIdentities: new Map(),
839
+ openToolResultIdentity: null,
840
+ seenCodexItemIds: new Map(),
841
+ turnOpen: false, assistantSeenInTurn: false, assistantMessageEnded: false, openMessageRole: null, openCustomType: null,
842
+ turnToolStarts: 0, turnToolCompletions: 0, turnToolResultMessages: 0, openTodoItemId: null,
843
+ agentStarted: false, agentCycleOpen: false, agentEnded: false, lastAgentEndWillRetry: null,
844
+ agentSettled: false, sessionSeen: false, threadSeen: false, terminalTurnSeen: false, providerErrorSeen: false,
845
+ lastObservationAt: null
846
+ };
847
+ const decoder = new StringDecoder("utf8");
848
+ let buffer = "";
849
+ let bufferedAt = 0;
850
+ let finished = false;
851
+ const consumeLine = (raw, at) => {
852
+ const line = raw.trim();
853
+ if (!line) return;
854
+ let event;
855
+ try { event = JSON.parse(line); }
856
+ catch { state.malformedLines += 1; return; }
857
+ if (!plainObject(event) || typeof event.type !== "string") { state.malformedLines += 1; return; }
858
+ state.jsonEvents += 1;
859
+ if (surface === "codex-cli") consumeCodexEvent(state, event, at);
860
+ else consumePiEvent(state, event, at);
861
+ };
862
+ return {
863
+ write(chunk, observedAtSeconds) {
864
+ if (finished) throw new Error("Benchmark timing collector is already finished");
865
+ if (!finiteNonnegative(observedAtSeconds) || observedAtSeconds < bufferedAt) state.invalidObservationClock = true;
866
+ bufferedAt = finiteNonnegative(observedAtSeconds) ? observedAtSeconds : bufferedAt;
867
+ if (finiteNonnegative(observedAtSeconds)) state.lastObservationAt = observedAtSeconds;
868
+ buffer += typeof chunk === "string" ? chunk : decoder.write(Buffer.from(chunk));
869
+ let newline;
870
+ while ((newline = buffer.indexOf("\n")) >= 0) {
871
+ consumeLine(buffer.slice(0, newline), bufferedAt);
872
+ buffer = buffer.slice(newline + 1);
873
+ }
874
+ if (Buffer.byteLength(buffer) > MAX_JSONL_BYTES) { state.malformedLines += 1; buffer = ""; }
875
+ },
876
+ finish(durationSeconds) {
877
+ if (finished) throw new Error("Benchmark timing collector is already finished");
878
+ finished = true;
879
+ buffer += decoder.end();
880
+ if (buffer) consumeLine(buffer, bufferedAt);
881
+ buffer = "";
882
+ return finish(state, durationSeconds);
883
+ }
884
+ };
885
+ }
886
+
887
+ export function createDeferredBenchmarkTimingCollector({
888
+ surface,
889
+ maximumBytes = MAX_TIMING_REPLAY_BYTES,
890
+ collectorFactory = createBenchmarkTimingCollector
891
+ }) {
892
+ if (!SURFACES.has(surface)) throw new Error(`Unsupported benchmark timing surface: ${surface}`);
893
+ if (!Number.isSafeInteger(maximumBytes) || maximumBytes <= 0) throw new Error("Benchmark timing replay limit must be a positive safe integer");
894
+ if (typeof collectorFactory !== "function") throw new Error("Benchmark timing collector factory must be a function");
895
+ const entries = [];
896
+ let capturedBytes = 0;
897
+ let overflow = false;
898
+ let finished = false;
899
+ const clear = () => {
900
+ for (const entry of entries) { try { entry.chunk.fill(0); } catch { /* Best-effort zeroing; release references below. */ } }
901
+ entries.length = 0;
902
+ capturedBytes = 0;
903
+ };
904
+ return {
905
+ write(chunk, observedAtSeconds) {
906
+ if (finished) throw new Error("Deferred benchmark timing collector is already finished");
907
+ if (overflow) return;
908
+ try {
909
+ const byteLength = Buffer.byteLength(chunk);
910
+ if (byteLength > maximumBytes - capturedBytes) {
911
+ overflow = true;
912
+ clear();
913
+ return;
914
+ }
915
+ const copy = Buffer.from(chunk);
916
+ entries.push({ chunk: copy, observedAtSeconds });
917
+ capturedBytes += copy.length;
918
+ } catch {
919
+ overflow = true;
920
+ clear();
921
+ }
922
+ },
923
+ discard() {
924
+ finished = true;
925
+ clear();
926
+ },
927
+ finish(durationSeconds) {
928
+ if (finished) throw new Error("Deferred benchmark timing collector is already finished");
929
+ finished = true;
930
+ try {
931
+ const collector = collectorFactory({ surface });
932
+ if (overflow) collector.write("invalid-bounded-timing-replay\n", 0);
933
+ else for (const entry of entries) collector.write(entry.chunk, entry.observedAtSeconds);
934
+ return collector.finish(durationSeconds);
935
+ } catch {
936
+ return unavailableBenchmarkTimingDiagnostics(surface, durationSeconds);
937
+ } finally {
938
+ clear();
939
+ }
940
+ }
941
+ };
942
+ }