@hassangameryt/freeflow 0.5.0 → 0.7.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (301) hide show
  1. package/.claude-plugin/marketplace.json +1 -1
  2. package/.claude-plugin/plugin.json +2 -2
  3. package/.codex-plugin/plugin.json +4 -5
  4. package/README.md +141 -163
  5. package/capabilities/cognitive-routing/SKILL.md +387 -0
  6. package/capabilities/context-virtualization/SKILL.md +103 -0
  7. package/capabilities/conversation-history/SKILL.md +53 -0
  8. package/command-surface.json +6 -37
  9. package/hooks/freeflow-runtime-context.mjs +50 -541
  10. package/hooks/hooks.json +1 -26
  11. package/hooks/tests/check-runtime-context-hook.sh +114 -308
  12. package/package.json +27 -17
  13. package/pi-extension/dist/cognitive-routing/commands.js +96 -0
  14. package/pi-extension/dist/cognitive-routing/config.js +232 -0
  15. package/pi-extension/dist/cognitive-routing/controller.js +1292 -0
  16. package/pi-extension/dist/cognitive-routing/history.js +451 -0
  17. package/pi-extension/dist/cognitive-routing/host.js +188 -0
  18. package/pi-extension/dist/cognitive-routing/pi-routing-session.js +414 -0
  19. package/pi-extension/dist/cognitive-routing/pi-session-control.js +320 -0
  20. package/pi-extension/dist/cognitive-routing/runtime.js +43 -0
  21. package/pi-extension/dist/cognitive-routing/tool.js +212 -0
  22. package/pi-extension/dist/cognitive-routing/types.js +6 -0
  23. package/pi-extension/dist/context-virtualization/commands.js +89 -0
  24. package/pi-extension/dist/context-virtualization/projector.js +42 -0
  25. package/pi-extension/dist/context-virtualization/runtime.js +356 -0
  26. package/pi-extension/dist/context-virtualization/state.js +57 -0
  27. package/pi-extension/dist/context-virtualization/types.js +10 -0
  28. package/pi-extension/dist/conversation-history/passages.js +66 -0
  29. package/pi-extension/dist/conversation-history/ranking.js +105 -0
  30. package/pi-extension/dist/conversation-history/runtime.js +226 -0
  31. package/pi-extension/dist/conversation-history/search.js +56 -0
  32. package/pi-extension/dist/conversation-history/types.js +1 -0
  33. package/pi-extension/dist/freeflow-context/resolver.js +120 -0
  34. package/pi-extension/dist/freeflow-context/runtime.js +80 -0
  35. package/pi-extension/dist/freeflow-context/source-projector.js +95 -0
  36. package/pi-extension/dist/freeflow-context/tool.js +308 -0
  37. package/pi-extension/dist/freeflow-context/types.js +12 -0
  38. package/pi-extension/dist/freeflow-context/visibility.js +28 -0
  39. package/pi-extension/dist/index.js +564 -129
  40. package/pi-extension/dist/runtime/runtime-context.js +357 -470
  41. package/pi-extension/dist/runtime/runtime-identity.js +5 -0
  42. package/pi-extension/dist/settings/settings-tui.js +78 -0
  43. package/pi-extension/dist/settings/settings-ui.js +572 -826
  44. package/pi-extension/dist/ui/text-component.js +57 -0
  45. package/pi-extension/src/cognitive-routing/commands.ts +129 -0
  46. package/pi-extension/src/cognitive-routing/config.ts +303 -0
  47. package/pi-extension/src/cognitive-routing/controller.ts +1628 -0
  48. package/pi-extension/src/cognitive-routing/history.ts +650 -0
  49. package/pi-extension/src/cognitive-routing/host.ts +266 -0
  50. package/pi-extension/src/cognitive-routing/pi-routing-session.ts +538 -0
  51. package/pi-extension/src/cognitive-routing/pi-session-control.ts +444 -0
  52. package/pi-extension/src/cognitive-routing/runtime.ts +57 -0
  53. package/pi-extension/src/cognitive-routing/tool.ts +273 -0
  54. package/pi-extension/src/cognitive-routing/types.ts +89 -0
  55. package/pi-extension/src/context-virtualization/commands.ts +111 -0
  56. package/pi-extension/src/context-virtualization/projector.ts +62 -0
  57. package/pi-extension/src/context-virtualization/runtime.ts +406 -0
  58. package/pi-extension/src/context-virtualization/state.ts +79 -0
  59. package/pi-extension/src/context-virtualization/types.ts +92 -0
  60. package/pi-extension/src/conversation-history/passages.ts +81 -0
  61. package/pi-extension/src/conversation-history/ranking.ts +155 -0
  62. package/pi-extension/src/conversation-history/runtime.ts +250 -0
  63. package/pi-extension/src/conversation-history/search.ts +79 -0
  64. package/pi-extension/src/conversation-history/types.ts +31 -0
  65. package/pi-extension/src/freeflow-context/resolver.ts +139 -0
  66. package/pi-extension/src/freeflow-context/runtime.ts +100 -0
  67. package/pi-extension/src/freeflow-context/source-projector.ts +106 -0
  68. package/pi-extension/src/freeflow-context/tool.ts +360 -0
  69. package/pi-extension/src/freeflow-context/types.ts +70 -0
  70. package/pi-extension/src/freeflow-context/visibility.ts +53 -0
  71. package/pi-extension/src/index.ts +597 -135
  72. package/pi-extension/src/runtime/runtime-context.ts +373 -491
  73. package/pi-extension/src/runtime/runtime-identity.ts +8 -0
  74. package/pi-extension/src/settings/settings-tui.ts +93 -0
  75. package/pi-extension/src/settings/settings-ui.ts +679 -910
  76. package/pi-extension/src/ui/text-component.ts +66 -0
  77. package/pi-extension/tests/cognitive-routing/commands.test.js +134 -0
  78. package/pi-extension/tests/cognitive-routing/controller.test.js +1128 -0
  79. package/pi-extension/tests/cognitive-routing/extension-lifecycle.test.js +1142 -0
  80. package/pi-extension/tests/cognitive-routing/history.test.js +274 -0
  81. package/pi-extension/tests/cognitive-routing/host-fixture.js +4 -0
  82. package/pi-extension/tests/cognitive-routing/host-identity.test.js +63 -0
  83. package/pi-extension/tests/cognitive-routing/pi-normal-lifecycle.test.js +434 -0
  84. package/pi-extension/tests/cognitive-routing/pi-session-control.test.js +174 -0
  85. package/pi-extension/tests/cognitive-routing/pi-session-history.test.js +91 -0
  86. package/pi-extension/tests/cognitive-routing/profile-resolution.test.js +279 -0
  87. package/pi-extension/tests/cognitive-routing/tool.test.js +230 -0
  88. package/pi-extension/tests/context-virtualization/extension.test.js +749 -0
  89. package/pi-extension/tests/context-virtualization/projector.test.js +110 -0
  90. package/pi-extension/tests/context-virtualization/runtime.test.js +216 -0
  91. package/pi-extension/tests/conversation-history/runtime.test.js +98 -0
  92. package/pi-extension/tests/conversation-history/search.test.js +97 -0
  93. package/pi-extension/tests/integration/mode-free-core.test.js +218 -0
  94. package/pi-extension/tests/integration/pi-extension.test.js +532 -3701
  95. package/pi-extension/tests/integration/prompt-architecture.test.js +350 -0
  96. package/pi-extension/tests/integration/unaffected-regressions.test.js +411 -0
  97. package/runtime/prompts/cognitive-routing.md +10 -0
  98. package/runtime/prompts/context-virtualization.md +3 -0
  99. package/runtime/prompts/conversation-history.md +3 -0
  100. package/runtime/prompts/core.md +84 -0
  101. package/runtime/prompts/interaction-contract.md +11 -0
  102. package/skills/action-selection/SKILL.md +140 -0
  103. package/skills/action-selection/references/trajectory-stalls.md +71 -0
  104. package/skills/bypass/SKILL.md +109 -38
  105. package/skills/commit-work/SKILL.md +94 -55
  106. package/skills/commit-work/references/git-state-edges.md +94 -0
  107. package/skills/commit-work/references/simple-push.md +92 -0
  108. package/skills/commit-work/references/staging-decisions.md +79 -46
  109. package/skills/decision-gate/SKILL.md +80 -22
  110. package/skills/design-for-depth/SKILL.md +206 -60
  111. package/skills/design-for-depth/references/design-pressure-signals.md +108 -240
  112. package/skills/design-for-depth/references/interface-design-loop.md +77 -97
  113. package/skills/design-for-depth/references/module-and-dependency-design.md +147 -0
  114. package/skills/design-for-depth/references/software-design-philosophy.md +128 -174
  115. package/skills/design-for-depth/references/state-and-failure-boundaries.md +136 -0
  116. package/skills/diagnose-failure/SKILL.md +200 -48
  117. package/skills/diagnose-failure/references/diagnostic-loop-catalog.md +18 -6
  118. package/skills/diagnose-failure/references/{flaky-and-performance.md → flaky-and-performance-diagnosis.md} +13 -5
  119. package/skills/discuss/SKILL.md +119 -65
  120. package/skills/discuss/references/discussion-continuity.md +123 -0
  121. package/skills/evaluate-skill/SKILL.md +166 -52
  122. package/skills/evaluate-skill/references/{definition-schema.md → evaluation-definition-schema.md} +43 -3
  123. package/skills/evaluate-skill/references/evaluation-design.md +56 -30
  124. package/skills/evaluate-skill/references/execution-and-evidence.md +8 -3
  125. package/skills/evaluate-skill/references/review-and-revision.md +62 -21
  126. package/skills/evaluate-skill/scripts/lib/definitions.mjs +114 -3
  127. package/skills/evaluate-skill/scripts/lib/grade-interaction.mjs +156 -1
  128. package/skills/evaluate-skill/scripts/lib/grade.mjs +9 -1
  129. package/skills/evaluate-skill/scripts/lib/pi.mjs +147 -32
  130. package/skills/evaluate-skill/scripts/lib/rpc.mjs +4 -4
  131. package/skills/evaluate-skill/scripts/lib/runner.mjs +36 -14
  132. package/skills/evaluate-skill/scripts/lib/sandbox.mjs +92 -0
  133. package/skills/evaluate-skill/scripts/pi-guard.mjs +1 -1
  134. package/skills/evaluate-skill/scripts/pi-observer.mjs +66 -0
  135. package/skills/evaluate-skill/scripts/skill-eval.mjs +6 -5
  136. package/skills/execute-work/SKILL.md +99 -57
  137. package/skills/execute-work/references/code-practices.md +4 -4
  138. package/skills/execute-work/references/execute-work-edges.md +62 -0
  139. package/skills/finish-branch/SKILL.md +2 -2
  140. package/skills/finish-branch/references/integration-options.md +1 -1
  141. package/skills/handoff/SKILL.md +3 -3
  142. package/skills/launch-work/SKILL.md +1 -1
  143. package/skills/launch-work/references/launch-readiness.md +2 -2
  144. package/skills/migration-work/SKILL.md +2 -2
  145. package/skills/migration-work/references/migration-lifecycle.md +1 -1
  146. package/skills/release-work/SKILL.md +1 -1
  147. package/skills/release-work/references/release-evidence.md +1 -1
  148. package/skills/review-artifact/SKILL.md +129 -66
  149. package/skills/review-artifact/references/adjudicate-artifact-review.md +101 -0
  150. package/skills/review-artifact/references/independent-artifact-reviewer-contract.md +159 -0
  151. package/skills/review-work/SKILL.md +113 -68
  152. package/skills/review-work/references/adjudicate-work-review.md +97 -0
  153. package/skills/review-work/references/independent-work-reviewer-contract.md +139 -0
  154. package/skills/review-work/references/security-risk-lens.md +18 -16
  155. package/skills/setup-freeflow/SKILL.md +16 -17
  156. package/skills/setup-freeflow/references/activation-contract.md +24 -30
  157. package/skills/setup-freeflow/references/host-setup.md +6 -7
  158. package/skills/simplify-code/SKILL.md +97 -39
  159. package/skills/simplify-code/references/simplification-patterns.md +1 -1
  160. package/skills/tdd/SKILL.md +100 -53
  161. package/skills/tdd/references/test-design.md +3 -3
  162. package/skills/track-work/SKILL.md +195 -58
  163. package/skills/track-work/references/working-record-format.md +515 -0
  164. package/skills/track-work/scripts/lib/document.mjs +713 -0
  165. package/skills/track-work/scripts/lib/format.mjs +166 -0
  166. package/skills/track-work/scripts/lib/store.mjs +233 -0
  167. package/skills/track-work/scripts/lib/transitions.mjs +790 -0
  168. package/skills/track-work/scripts/lib/views.mjs +40 -0
  169. package/skills/track-work/scripts/working-record.mjs +316 -0
  170. package/skills/verify-work/SKILL.md +57 -19
  171. package/skills/verify-work/references/browser-runtime-evidence.md +2 -2
  172. package/skills/verify-work/references/integration-evidence.md +1 -7
  173. package/skills/verify-work/references/performance-evidence.md +2 -2
  174. package/skills/workflow/SKILL.md +137 -55
  175. package/skills/workflow/references/domain-skill-composition.md +26 -40
  176. package/skills/write-plan/SKILL.md +1 -1
  177. package/skills/write-skill/SKILL.md +218 -61
  178. package/skills/write-skill/references/activation-boundaries.md +63 -21
  179. package/skills/write-skill/references/agent-first-instructions.md +61 -27
  180. package/skills/write-skill/references/progressive-disclosure.md +97 -33
  181. package/skills/write-skill/references/skill-dependency-graphs.md +190 -0
  182. package/capabilities/output-router/SKILL.md +0 -206
  183. package/capabilities/output-router/references/safety-policy.md +0 -40
  184. package/capabilities/output-router/references/setup.md +0 -110
  185. package/pi-extension/dist/output-router/host-producer-identification.js +0 -232
  186. package/pi-extension/dist/output-router/native-safety-net.js +0 -216
  187. package/pi-extension/dist/output-router/observed-tool-routing.js +0 -122
  188. package/pi-extension/dist/output-router/renderers.js +0 -1023
  189. package/pi-extension/dist/output-router/router-tools.js +0 -554
  190. package/pi-extension/dist/output-router/schemas.js +0 -383
  191. package/pi-extension/dist/output-router/status.js +0 -744
  192. package/pi-extension/dist/output-router/utils.js +0 -502
  193. package/pi-extension/src/output-router/host-producer-identification.ts +0 -256
  194. package/pi-extension/src/output-router/native-safety-net.ts +0 -248
  195. package/pi-extension/src/output-router/observed-tool-routing.ts +0 -134
  196. package/pi-extension/src/output-router/renderers.ts +0 -1108
  197. package/pi-extension/src/output-router/router-tools.ts +0 -579
  198. package/pi-extension/src/output-router/schemas.ts +0 -387
  199. package/pi-extension/src/output-router/status.ts +0 -791
  200. package/pi-extension/src/output-router/utils.ts +0 -554
  201. package/pi-extension/tests/output-router/observed-producers.test.js +0 -132
  202. package/pi-extension/tests/output-router/search.test.js +0 -722
  203. package/router/dist/benchmarks/benchmark-harness.d.ts +0 -35
  204. package/router/dist/benchmarks/benchmark-harness.js +0 -92
  205. package/router/dist/benchmarks/benchmarks.d.ts +0 -109
  206. package/router/dist/benchmarks/benchmarks.js +0 -979
  207. package/router/dist/benchmarks/codex-qa-benchmarks.d.ts +0 -95
  208. package/router/dist/benchmarks/codex-qa-benchmarks.js +0 -500
  209. package/router/dist/benchmarks/command-benchmarks.d.ts +0 -143
  210. package/router/dist/benchmarks/command-benchmarks.js +0 -698
  211. package/router/dist/benchmarks/context-mode-normalized-benchmarks.d.ts +0 -82
  212. package/router/dist/benchmarks/context-mode-normalized-benchmarks.js +0 -773
  213. package/router/dist/benchmarks/context-mode-real-deep-benchmark.d.ts +0 -132
  214. package/router/dist/benchmarks/context-mode-real-deep-benchmark.js +0 -1380
  215. package/router/dist/benchmarks/context-saving-baseline.d.ts +0 -60
  216. package/router/dist/benchmarks/context-saving-baseline.js +0 -470
  217. package/router/dist/benchmarks/index-benchmarks.d.ts +0 -103
  218. package/router/dist/benchmarks/index-benchmarks.js +0 -862
  219. package/router/dist/benchmarks/storage-policy-benchmarks.d.ts +0 -101
  220. package/router/dist/benchmarks/storage-policy-benchmarks.js +0 -618
  221. package/router/dist/benchmarks/vault-index-benchmarks.d.ts +0 -56
  222. package/router/dist/benchmarks/vault-index-benchmarks.js +0 -322
  223. package/router/dist/config/config.d.ts +0 -103
  224. package/router/dist/config/config.js +0 -632
  225. package/router/dist/config/router-contract.d.ts +0 -10
  226. package/router/dist/config/router-contract.js +0 -54
  227. package/router/dist/config/schema.d.ts +0 -27
  228. package/router/dist/config/schema.js +0 -648
  229. package/router/dist/config/types.d.ts +0 -573
  230. package/router/dist/config/types.js +0 -41
  231. package/router/dist/evidence/bounded-evidence.d.ts +0 -40
  232. package/router/dist/evidence/bounded-evidence.js +0 -323
  233. package/router/dist/evidence/evidence-range-selector.d.ts +0 -24
  234. package/router/dist/evidence/evidence-range-selector.js +0 -232
  235. package/router/dist/evidence/evidence-search.d.ts +0 -27
  236. package/router/dist/evidence/evidence-search.js +0 -490
  237. package/router/dist/evidence/evidence.d.ts +0 -38
  238. package/router/dist/evidence/evidence.js +0 -200
  239. package/router/dist/evidence/failure-contracts.d.ts +0 -37
  240. package/router/dist/evidence/failure-contracts.js +0 -166
  241. package/router/dist/evidence/line-ranges.d.ts +0 -24
  242. package/router/dist/evidence/line-ranges.js +0 -22
  243. package/router/dist/experiments/local-index.d.ts +0 -64
  244. package/router/dist/experiments/local-index.js +0 -581
  245. package/router/dist/index.d.ts +0 -30
  246. package/router/dist/index.js +0 -32
  247. package/router/dist/local/local-traversal.d.ts +0 -18
  248. package/router/dist/local/local-traversal.js +0 -334
  249. package/router/dist/processing/engine.d.ts +0 -132
  250. package/router/dist/processing/engine.js +0 -533
  251. package/router/dist/processing/reducers.d.ts +0 -280
  252. package/router/dist/processing/reducers.js +0 -1673
  253. package/router/dist/processing/renderers.d.ts +0 -31
  254. package/router/dist/processing/renderers.js +0 -195
  255. package/router/dist/processing/scripts.d.ts +0 -91
  256. package/router/dist/processing/scripts.js +0 -386
  257. package/router/dist/repo/repo-traversal.d.ts +0 -18
  258. package/router/dist/repo/repo-traversal.js +0 -245
  259. package/router/dist/routing/observed-reducers.d.ts +0 -41
  260. package/router/dist/routing/observed-reducers.js +0 -661
  261. package/router/dist/routing/observed-routing.d.ts +0 -26
  262. package/router/dist/routing/observed-routing.js +0 -378
  263. package/router/dist/routing/parsers.d.ts +0 -16
  264. package/router/dist/routing/parsers.js +0 -521
  265. package/router/dist/routing/run-filters.d.ts +0 -60
  266. package/router/dist/routing/run-filters.js +0 -238
  267. package/router/dist/routing/run-reducers.d.ts +0 -49
  268. package/router/dist/routing/run-reducers.js +0 -122
  269. package/router/dist/sandbox/adapter-roots.d.ts +0 -46
  270. package/router/dist/sandbox/adapter-roots.js +0 -66
  271. package/router/dist/sandbox/eryx-python-adapter.d.ts +0 -13
  272. package/router/dist/sandbox/eryx-python-adapter.js +0 -868
  273. package/router/dist/sandbox/jq-wasm-adapter.d.ts +0 -9
  274. package/router/dist/sandbox/jq-wasm-adapter.js +0 -429
  275. package/router/dist/sandbox/quickjs-wasi-adapter.d.ts +0 -13
  276. package/router/dist/sandbox/quickjs-wasi-adapter.js +0 -384
  277. package/router/dist/sandbox/script-sandbox.d.ts +0 -134
  278. package/router/dist/sandbox/script-sandbox.js +0 -337
  279. package/router/dist/setup/script-transform-adapters.d.ts +0 -32
  280. package/router/dist/setup/script-transform-adapters.js +0 -278
  281. package/router/dist/tools/batch.d.ts +0 -33
  282. package/router/dist/tools/batch.js +0 -586
  283. package/router/dist/tools/run.d.ts +0 -62
  284. package/router/dist/tools/run.js +0 -1304
  285. package/router/dist/tools/search.d.ts +0 -66
  286. package/router/dist/tools/search.js +0 -1364
  287. package/router/dist/transform/engine.d.ts +0 -168
  288. package/router/dist/transform/engine.js +0 -2072
  289. package/router/dist/vault/vault-index.d.ts +0 -125
  290. package/router/dist/vault/vault-index.js +0 -416
  291. package/router/dist/vault/vault.d.ts +0 -138
  292. package/router/dist/vault/vault.js +0 -630
  293. package/runtime/interaction-contract.md +0 -16
  294. package/skills/discuss/references/checkpoints.md +0 -66
  295. package/skills/execute-work/references/execution-loop.md +0 -77
  296. package/skills/mode-contract/SKILL.md +0 -104
  297. package/skills/review-artifact/references/reviewer-prompt.md +0 -148
  298. package/skills/review-work/references/reviewer-prompt.md +0 -134
  299. package/skills/track-work/references/working-record-schema.md +0 -378
  300. package/skills/workflow/references/workflow-loop.md +0 -92
  301. /package/skills/handoff/references/{templates.md → handoff-templates.md} +0 -0
@@ -1,618 +0,0 @@
1
- import { createHash } from "node:crypto";
2
- import { readdir, rm, stat } from "node:fs/promises";
3
- import { tmpdir } from "node:os";
4
- import { join, resolve } from "node:path";
5
- import { fileURLToPath } from "node:url";
6
- import { mkdtemp } from "node:fs/promises";
7
- import { performance } from "node:perf_hooks";
8
- import {
9
- approximateTokens,
10
- defaultJsonRunReportPath,
11
- escapeMarkdownTableCell as escapeTable,
12
- formatPercent,
13
- latencySummary,
14
- normalizeIterations,
15
- parseBenchmarkCliArgs,
16
- reductionPercent,
17
- writeBenchmarkReportPair,
18
- } from "./benchmark-harness.js";
19
- import {
20
- commandOutputFingerprints,
21
- createVault,
22
- findExactDuplicateCommandOutput,
23
- readOutputText,
24
- storeCommandOutput,
25
- storeMetadataOutput,
26
- } from "../vault/vault.js";
27
- const DEFAULT_ITERATIONS = 2;
28
- const LARGE_EXACT_THRESHOLD_BYTES = 8 * 1024;
29
- const FIXTURES = createStoragePolicyFixtures();
30
- const POLICIES = [
31
- {
32
- id: "store-everything",
33
- label: "Store everything exactly",
34
- description:
35
- "Current baseline: every command output is vaulted exactly, so recovery is maximally useful and storage/privacy surface is largest.",
36
- decide() {
37
- return "exact";
38
- },
39
- },
40
- {
41
- id: "threshold-exact",
42
- label: "Threshold exact storage",
43
- description: `Exact storage only when combined output is at least ${LARGE_EXACT_THRESHOLD_BYTES} bytes; smaller outputs are metadata-only. This intentionally tests whether a threshold alone would drop exact failure/verification recovery.`,
44
- decide(input) {
45
- return input.rawCombinedBytes >= LARGE_EXACT_THRESHOLD_BYTES ? "exact" : "metadata";
46
- },
47
- },
48
- {
49
- id: "metadata-small-exact-large-hybrid",
50
- label: "Metadata-small / exact-large hybrid",
51
- description: `Small non-sensitive outputs become metadata-only; exactness-sensitive or >= ${LARGE_EXACT_THRESHOLD_BYTES} byte outputs remain exactly recoverable.`,
52
- decide(input) {
53
- return input.fixture.exactnessSensitive || input.rawCombinedBytes >= LARGE_EXACT_THRESHOLD_BYTES
54
- ? "exact"
55
- : "metadata";
56
- },
57
- },
58
- {
59
- id: "duplicate-output-dedupe",
60
- label: "Duplicate output dedupe",
61
- description:
62
- "First occurrence is stored exactly; exact duplicates become metadata-only records that point at the prior exact output.",
63
- decide(input) {
64
- return input.duplicate ? "duplicate-metadata" : "exact";
65
- },
66
- },
67
- {
68
- id: "hybrid-dedupe",
69
- label: "Hybrid exactness + duplicate dedupe",
70
- description: `Exactness-sensitive or >= ${LARGE_EXACT_THRESHOLD_BYTES} byte outputs stay exactly recoverable; exact duplicates of those outputs become metadata pointers; small non-sensitive outputs are metadata-only.`,
71
- decide(input) {
72
- const shouldStoreExact =
73
- input.fixture.exactnessSensitive || input.rawCombinedBytes >= LARGE_EXACT_THRESHOLD_BYTES;
74
- if (!shouldStoreExact) {
75
- return "metadata";
76
- }
77
- return input.duplicate ? "duplicate-metadata" : "exact";
78
- },
79
- },
80
- ];
81
- export async function runStoragePolicyBenchmarks(options = {}) {
82
- const iterations = normalizeIterations(options.iterations, DEFAULT_ITERATIONS);
83
- const policies = [];
84
- for (const policy of POLICIES) {
85
- policies.push(await runPolicy(policy, iterations));
86
- }
87
- return {
88
- generatedAt: options.generatedAt ?? new Date().toISOString(),
89
- iterations,
90
- summary: summarizeStoragePolicyReport(policies),
91
- fixtures: FIXTURES,
92
- policies,
93
- };
94
- }
95
- export function renderStoragePolicyBenchmarkReport(report) {
96
- const date = report.generatedAt.slice(0, 10);
97
- const lines = [
98
- "# Storage Policy Benchmark Report - Iteration 1",
99
- "",
100
- `Date: ${date}`,
101
- "",
102
- "## Scope",
103
- "",
104
- "Benchmark-only experiment for Freeflow vault storage policies. This report does not change runtime defaults or repo config. It compares exact storage, threshold storage, hybrid metadata/exact storage, and duplicate-output dedupe using deterministic command-output fixtures.",
105
- "",
106
- "## Summary",
107
- "",
108
- `- Fixtures: ${report.summary.fixtures}`,
109
- `- Policies: ${report.summary.policies}`,
110
- `- Safe candidates for further evaluation: ${report.summary.safeCandidateIds.join(", ") || "none"}`,
111
- `- Disqualified by exact-recovery safety: ${report.summary.disqualifiedCandidateIds.join(", ") || "none"}`,
112
- `- Runtime default changed: ${report.summary.defaultUnchanged ? "no" : "yes"}`,
113
- "",
114
- "## Policy Results",
115
- "",
116
- "| Policy | Exact-sensitive recovery | Storage bytes | Index bytes | Exact stored bytes | Storage reduction | Token-surface reduction | Privacy surface | Metadata-only | Duplicate metadata | Latency p50/p95 | Notes |",
117
- "| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | --- |",
118
- ];
119
- for (const policy of report.policies) {
120
- lines.push(
121
- [
122
- escapeTable(policy.label),
123
- `${policy.safety.exactnessSensitiveRecoverable}/${policy.safety.exactnessSensitiveFixtures}`,
124
- String(policy.totals.storageBytes),
125
- String(policy.totals.indexBytes),
126
- String(policy.totals.exactStoredCombinedBytes),
127
- formatPercent(policy.totals.storageReductionPercent),
128
- formatPercent(policy.totals.tokenSurfaceReductionPercent),
129
- formatPercent(policy.totals.privacySurfacePercent),
130
- String(policy.totals.metadataOnlyRecords),
131
- String(policy.totals.duplicateMetadataRecords),
132
- `${policy.totals.latencyMs.p50.toFixed(2)}/${policy.totals.latencyMs.p95.toFixed(2)}`,
133
- escapeTable(policy.notes.join(" ")),
134
- ]
135
- .join(" | ")
136
- .replace(/^/, "| ")
137
- .replace(/$/, " |"),
138
- );
139
- }
140
- lines.push("", "## Candidate Notes", "");
141
- for (const policy of report.policies) {
142
- lines.push(`### ${policy.label}`, "", policy.description, "");
143
- lines.push(`- Exactness-sensitive recovery: ${policy.safety.exactnessSensitiveRecoveryPassed ? "pass" : "fail"}`);
144
- lines.push(`- Metadata-only recovery labeled: ${policy.safety.metadataOnlyRecoveryLabeled ? "yes" : "no"}`);
145
- lines.push(`- Repeated outputs deduped: ${policy.safety.repeatedOutputsDeduped ? "yes" : "no"}`);
146
- lines.push(
147
- "",
148
- "| Fixture | Iteration | Record | Recovery | Exact bytes | Notes |",
149
- "| --- | ---: | --- | --- | ---: | --- |",
150
- );
151
- for (const fixture of policy.fixtures) {
152
- lines.push(
153
- [
154
- escapeTable(fixture.fixtureId),
155
- String(fixture.iteration),
156
- `${fixture.recordKind}/${fixture.recoverability}`,
157
- fixture.usefulRecovery,
158
- String(fixture.exactStoredCombinedBytes),
159
- escapeTable(fixture.notes.join(" ")),
160
- ]
161
- .join(" | ")
162
- .replace(/^/, "| ")
163
- .replace(/$/, " |"),
164
- );
165
- }
166
- lines.push("");
167
- }
168
- lines.push(
169
- "## Decision Boundary",
170
- "",
171
- "This benchmark intentionally stops before changing defaults. Any storage default change still needs a product/safety decision because metadata-only storage changes exact recovery semantics and privacy surface.",
172
- "",
173
- );
174
- return `${lines.join("\n")}\n`;
175
- }
176
- export async function writeStoragePolicyBenchmarkReports(
177
- report,
178
- markdownReportPath = defaultReportPath(),
179
- options = {},
180
- ) {
181
- return writeBenchmarkReportPair({
182
- report,
183
- markdownReportPath,
184
- jsonReportPath: options.jsonReportPath,
185
- renderMarkdown: renderStoragePolicyBenchmarkReport,
186
- });
187
- }
188
- async function runPolicy(policy, iterations) {
189
- const temp = await createTempDir(`freeflow-storage-policy-${policy.id}-`);
190
- const vault = createVault({ root: temp.path });
191
- const sessionId = `storage-policy-${policy.id}`;
192
- const fixtureResults = [];
193
- try {
194
- for (let iteration = 1; iteration <= iterations; iteration += 1) {
195
- for (const fixture of FIXTURES) {
196
- fixtureResults.push(await runPolicyFixture({ policy, vault, sessionId, fixture, iteration }));
197
- }
198
- }
199
- const storageBytes = await directorySize(temp.path);
200
- const indexBytes = await directorySize(join(temp.path, "index"));
201
- return buildPolicyResult({ policy, fixtureResults, storageBytes, indexBytes });
202
- } finally {
203
- await temp.cleanup();
204
- }
205
- }
206
- async function runPolicyFixture(options) {
207
- const combined = combineOutputSections(options.fixture.stdout, options.fixture.stderr);
208
- const rawCombinedBytes = byteLength(combined);
209
- const fingerprints = commandOutputFingerprints({
210
- command: options.fixture.command,
211
- stdout: options.fixture.stdout,
212
- stderr: options.fixture.stderr,
213
- combined,
214
- executionStatus: options.fixture.executionStatus,
215
- exitCode: options.fixture.exitCode,
216
- });
217
- const duplicate = await findExactDuplicateCommandOutput(options.vault, {
218
- sessionId: options.sessionId,
219
- fingerprints,
220
- }).catch(() => undefined);
221
- const decisionInput = { fixture: options.fixture, combined, rawCombinedBytes };
222
- if (duplicate !== undefined) {
223
- decisionInput.duplicate = duplicate;
224
- }
225
- const decision = options.policy.decide(decisionInput);
226
- const storeOptions = { ...options, combined, rawCombinedBytes, fingerprints, decision };
227
- if (duplicate !== undefined) {
228
- Object.assign(storeOptions, { duplicate });
229
- }
230
- const observation = await storeWithPolicy(storeOptions);
231
- const recovery = await checkRecovery({ ...options, observation, combined });
232
- return {
233
- fixtureId: options.fixture.id,
234
- iteration: options.iteration,
235
- outputId: observation.record.outputId,
236
- recordKind: observation.record.kind,
237
- persistence: observation.record.persistence.status,
238
- recoverability: observation.record.persistence.recoverability,
239
- rawCombinedBytes,
240
- exactStoredCombinedBytes: observation.exactStoredCombinedBytes,
241
- latencyMs: observation.latencyMs,
242
- exactRecovery: recovery.exactRecovery,
243
- usefulRecovery: recovery.usefulRecovery,
244
- ...(observation.duplicateOf !== undefined ? { duplicateOf: observation.duplicateOf } : {}),
245
- exactnessSensitive: options.fixture.exactnessSensitive,
246
- notes: recovery.notes,
247
- };
248
- }
249
- async function storeWithPolicy(options) {
250
- const startedAt = performance.now();
251
- if (options.decision === "exact") {
252
- const record = await storeCommandOutput(options.vault, {
253
- sessionId: options.sessionId,
254
- command: options.fixture.command,
255
- stdout: options.fixture.stdout,
256
- stderr: options.fixture.stderr,
257
- combined: options.combined,
258
- executionStatus: options.fixture.executionStatus,
259
- exitCode: options.fixture.exitCode,
260
- decisionIds: [
261
- decisionId("storage-policy", options.policy.id, options.fixture.id, String(options.iteration), "exact"),
262
- ],
263
- producer: { kind: "command", name: "storage-policy-experiment" },
264
- });
265
- return {
266
- record,
267
- decision: options.decision,
268
- latencyMs: performance.now() - startedAt,
269
- rawCombinedBytes: options.rawCombinedBytes,
270
- exactStoredCombinedBytes: options.rawCombinedBytes,
271
- };
272
- }
273
- const duplicateOf = options.decision === "duplicate-metadata" ? options.duplicate?.outputId : undefined;
274
- const record = await storeMetadataOutput(options.vault, {
275
- sessionId: options.sessionId,
276
- sourceKind: "other",
277
- rawLineCount: countLines(options.combined),
278
- rawByteCount: options.rawCombinedBytes,
279
- rawSha256: sha256Text(options.combined),
280
- decisionIds: [
281
- decisionId("storage-policy", options.policy.id, options.fixture.id, String(options.iteration), options.decision),
282
- ],
283
- producer: { kind: "command", name: "storage-policy-experiment" },
284
- metadata: {
285
- policy: options.policy.id,
286
- command: options.fixture.command,
287
- executionStatus: options.fixture.executionStatus,
288
- exitCode: options.fixture.exitCode,
289
- exactnessSensitive: options.fixture.exactnessSensitive,
290
- ...(duplicateOf !== undefined ? { duplicateOf } : {}),
291
- },
292
- });
293
- return {
294
- record,
295
- decision: options.decision,
296
- latencyMs: performance.now() - startedAt,
297
- rawCombinedBytes: options.rawCombinedBytes,
298
- exactStoredCombinedBytes: 0,
299
- ...(duplicateOf !== undefined ? { duplicateOf } : {}),
300
- };
301
- }
302
- async function checkRecovery(options) {
303
- const notes = [];
304
- if (options.observation.record.kind === "command") {
305
- try {
306
- const recovered = await readOutputText(
307
- options.vault,
308
- options.sessionId,
309
- options.observation.record.outputId,
310
- "combined",
311
- );
312
- const passed = recovered === options.combined;
313
- notes.push(passed ? "exact combined recovery passed" : "exact combined recovery mismatch");
314
- return { exactRecovery: passed ? "passed" : "failed", usefulRecovery: passed ? "exact" : "none", notes };
315
- } catch (error) {
316
- notes.push(`exact recovery failed: ${errorMessage(error)}`);
317
- return { exactRecovery: "failed", usefulRecovery: "none", notes };
318
- }
319
- }
320
- if (options.observation.duplicateOf !== undefined) {
321
- try {
322
- const recovered = await readOutputText(
323
- options.vault,
324
- options.sessionId,
325
- options.observation.duplicateOf,
326
- "combined",
327
- );
328
- const passed = recovered === options.combined;
329
- notes.push(
330
- passed
331
- ? `metadata duplicate points to exact outputId=${options.observation.duplicateOf}`
332
- : `metadata duplicate pointer mismatch outputId=${options.observation.duplicateOf}`,
333
- );
334
- return { exactRecovery: "not-exact", usefulRecovery: passed ? "duplicate-ref" : "none", notes };
335
- } catch (error) {
336
- notes.push(`duplicate reference recovery failed: ${errorMessage(error)}`);
337
- return { exactRecovery: "not-exact", usefulRecovery: "none", notes };
338
- }
339
- }
340
- notes.push("metadata-only record intentionally has no raw exact recovery");
341
- return { exactRecovery: "not-exact", usefulRecovery: "metadata-only", notes };
342
- }
343
- function buildPolicyResult(options) {
344
- const rawCombinedBytes = sum(options.fixtureResults.map((fixture) => fixture.rawCombinedBytes));
345
- const exactStoredCombinedBytes = sum(options.fixtureResults.map((fixture) => fixture.exactStoredCombinedBytes));
346
- const latency = latencySummary(options.fixtureResults.map((fixture) => fixture.latencyMs));
347
- const exactnessSensitiveFixtures = options.fixtureResults.filter((fixture) => fixture.exactnessSensitive).length;
348
- const exactnessSensitiveRecoverable = options.fixtureResults.filter(
349
- (fixture) =>
350
- fixture.exactnessSensitive && (fixture.usefulRecovery === "exact" || fixture.usefulRecovery === "duplicate-ref"),
351
- ).length;
352
- const metadataOnlyRecords = options.fixtureResults.filter((fixture) => fixture.recordKind === "metadata").length;
353
- const duplicateMetadataRecords = options.fixtureResults.filter(
354
- (fixture) => fixture.usefulRecovery === "duplicate-ref",
355
- ).length;
356
- const repeatedOutputsDeduped = duplicateMetadataRecords > 0;
357
- const safety = {
358
- exactnessSensitiveFixtures,
359
- exactnessSensitiveRecoverable,
360
- exactnessSensitiveRecoveryPassed: exactnessSensitiveRecoverable === exactnessSensitiveFixtures,
361
- metadataOnlyRecoveryLabeled: options.fixtureResults.every(
362
- (fixture) => fixture.recordKind !== "metadata" || fixture.recoverability === "metadata_only",
363
- ),
364
- repeatedOutputsDeduped,
365
- };
366
- const notes = [];
367
- if (!safety.exactnessSensitiveRecoveryPassed) {
368
- notes.push("disqualified: exactness-sensitive failure/verification output would lose exact recovery");
369
- }
370
- if (options.policy.id === "threshold-exact" && !safety.exactnessSensitiveRecoveryPassed) {
371
- notes.push("threshold-only policy is unsafe without an exactness-sensitive override");
372
- }
373
- if (options.policy.id === "metadata-small-exact-large-hybrid" && safety.exactnessSensitiveRecoveryPassed) {
374
- notes.push("hybrid preserved exact recovery for sensitive fixtures while reducing exact raw storage");
375
- }
376
- if (options.policy.id === "duplicate-output-dedupe" && repeatedOutputsDeduped) {
377
- notes.push("duplicates kept metadata pointers to prior exact output");
378
- }
379
- if (options.policy.id === "hybrid-dedupe" && safety.exactnessSensitiveRecoveryPassed && repeatedOutputsDeduped) {
380
- notes.push(
381
- "hybrid+dedupe preserved exact-sensitive recovery while metadata-only small output and duplicate pointers reduced exact raw storage",
382
- );
383
- }
384
- return {
385
- policyId: options.policy.id,
386
- label: options.policy.label,
387
- description: options.policy.description,
388
- totals: {
389
- rawCombinedBytes,
390
- rawCombinedTokensApprox: approximateTokens(rawCombinedBytes),
391
- exactStoredCombinedBytes,
392
- exactStoredTokensApprox: approximateTokens(exactStoredCombinedBytes),
393
- storageBytes: options.storageBytes,
394
- indexBytes: options.indexBytes,
395
- metadataOnlyRecords,
396
- exactRecords: options.fixtureResults.length - metadataOnlyRecords,
397
- duplicateMetadataRecords,
398
- storageReductionPercent: reductionPercent(rawCombinedBytes, exactStoredCombinedBytes),
399
- tokenSurfaceReductionPercent: reductionPercent(
400
- approximateTokens(rawCombinedBytes),
401
- approximateTokens(exactStoredCombinedBytes),
402
- ),
403
- privacySurfacePercent:
404
- rawCombinedBytes > 0 ? Math.round((exactStoredCombinedBytes / rawCombinedBytes) * 10_000) / 100 : 0,
405
- latencyMs: latency,
406
- },
407
- safety,
408
- fixtures: options.fixtureResults,
409
- notes,
410
- };
411
- }
412
- function summarizeStoragePolicyReport(policies) {
413
- const safeCandidateIds = policies
414
- .filter((policy) => policy.safety.exactnessSensitiveRecoveryPassed && policy.safety.metadataOnlyRecoveryLabeled)
415
- .map((policy) => policy.policyId);
416
- const disqualifiedCandidateIds = policies
417
- .filter((policy) => !policy.safety.exactnessSensitiveRecoveryPassed || !policy.safety.metadataOnlyRecoveryLabeled)
418
- .map((policy) => policy.policyId);
419
- return {
420
- fixtures: FIXTURES.length,
421
- policies: policies.length,
422
- safeCandidateIds,
423
- disqualifiedCandidateIds,
424
- defaultUnchanged: true,
425
- };
426
- }
427
- function createStoragePolicyFixtures() {
428
- const largeLines =
429
- Array.from(
430
- { length: 300 },
431
- (_, index) => `large log line ${String(index + 1).padStart(3, "0")} ${"x".repeat(80)}`,
432
- ).join("\n") + "\nLARGE_LOG_SENTINEL exact tail\n";
433
- return [
434
- {
435
- id: "small-success",
436
- title: "Small successful command output",
437
- command: "echo ok",
438
- stdout: "ok\n",
439
- stderr: "",
440
- executionStatus: "success",
441
- exitCode: 0,
442
- exactnessSensitive: false,
443
- },
444
- {
445
- id: "small-failure",
446
- title: "Small failed command with exact diagnostic",
447
- command: "npm test",
448
- stdout: "Tests: 1 failed, 24 passed, 25 total\n",
449
- stderr: "AssertionError: STORAGE_POLICY_FAILURE_SENTINEL\n",
450
- executionStatus: "failed",
451
- exitCode: 1,
452
- exactnessSensitive: true,
453
- },
454
- {
455
- id: "verification-output",
456
- title: "Verification output with completion claim evidence",
457
- command: "npm run verify",
458
- stdout: "Verification passed: STORAGE_POLICY_VERIFY_SENTINEL\n",
459
- stderr: "",
460
- executionStatus: "success",
461
- exitCode: 0,
462
- exactnessSensitive: true,
463
- },
464
- {
465
- id: "large-log",
466
- title: "Large noisy log output",
467
- command: "tail huge.log",
468
- stdout: largeLines,
469
- stderr: "",
470
- executionStatus: "success",
471
- exitCode: 0,
472
- exactnessSensitive: false,
473
- },
474
- {
475
- id: "large-log-repeat",
476
- title: "Large noisy log repeated output",
477
- command: "tail huge.log",
478
- stdout: largeLines,
479
- stderr: "",
480
- executionStatus: "success",
481
- exitCode: 0,
482
- exactnessSensitive: false,
483
- repeatedGroup: "large-log",
484
- },
485
- {
486
- id: "repeated-failure-a",
487
- title: "Repeated exactness-sensitive failure first occurrence",
488
- command: "npm test -- repeated",
489
- stdout: "Tests: 1 failed, 24 passed, 25 total\n",
490
- stderr: "AssertionError: REPEATED_STORAGE_POLICY_FAILURE_SENTINEL\n",
491
- executionStatus: "failed",
492
- exitCode: 1,
493
- exactnessSensitive: true,
494
- repeatedGroup: "failure",
495
- },
496
- {
497
- id: "repeated-failure-b",
498
- title: "Repeated exactness-sensitive failure second occurrence",
499
- command: "npm test -- repeated",
500
- stdout: "Tests: 1 failed, 24 passed, 25 total\n",
501
- stderr: "AssertionError: REPEATED_STORAGE_POLICY_FAILURE_SENTINEL\n",
502
- executionStatus: "failed",
503
- exitCode: 1,
504
- exactnessSensitive: true,
505
- repeatedGroup: "failure",
506
- },
507
- {
508
- id: "repeat-a",
509
- title: "Repeated output first occurrence",
510
- command: "fixture repeat",
511
- stdout: "REPEATED_STORAGE_POLICY_SENTINEL\n",
512
- stderr: "",
513
- executionStatus: "success",
514
- exitCode: 0,
515
- exactnessSensitive: false,
516
- repeatedGroup: "repeat",
517
- },
518
- {
519
- id: "repeat-b",
520
- title: "Repeated output second occurrence",
521
- command: "fixture repeat",
522
- stdout: "REPEATED_STORAGE_POLICY_SENTINEL\n",
523
- stderr: "",
524
- executionStatus: "success",
525
- exitCode: 0,
526
- exactnessSensitive: false,
527
- repeatedGroup: "repeat",
528
- },
529
- ];
530
- }
531
- async function directorySize(path) {
532
- try {
533
- const info = await stat(path);
534
- if (info.isFile()) {
535
- return info.size;
536
- }
537
- if (!info.isDirectory()) {
538
- return 0;
539
- }
540
- } catch {
541
- return 0;
542
- }
543
- const entries = await readdir(path, { withFileTypes: true });
544
- let total = 0;
545
- for (const entry of entries) {
546
- total += await directorySize(join(path, entry.name));
547
- }
548
- return total;
549
- }
550
- async function createTempDir(prefix) {
551
- const path = await mkdtemp(resolve(tmpdir(), prefix));
552
- return {
553
- path,
554
- cleanup: async () => {
555
- await rm(path, { recursive: true, force: true });
556
- },
557
- };
558
- }
559
- function combineOutputSections(stdout, stderr) {
560
- if (stdout.length === 0) {
561
- return stderr;
562
- }
563
- if (stderr.length === 0) {
564
- return stdout;
565
- }
566
- return `[stdout]\n${stdout}\n[stderr]\n${stderr}`;
567
- }
568
- function countLines(text) {
569
- if (text.length === 0) {
570
- return 0;
571
- }
572
- return text.split("\n").length;
573
- }
574
- function sum(values) {
575
- return values.reduce((total, value) => total + value, 0);
576
- }
577
- function byteLength(value) {
578
- return Buffer.byteLength(value, "utf8");
579
- }
580
- function sha256Text(value) {
581
- return createHash("sha256").update(value).digest("hex");
582
- }
583
- function decisionId(...parts) {
584
- return `ffdec_${sha256Text(parts.join("\0")).slice(0, 16)}`;
585
- }
586
- function errorMessage(error) {
587
- return error instanceof Error ? error.message : String(error);
588
- }
589
- function defaultReportPath() {
590
- return resolve(process.cwd(), "router/evals/reports/storage-policy-benchmark-1-report.md");
591
- }
592
- async function runCli() {
593
- const { iterations, reportPath, jsonReportPath } = parseBenchmarkCliArgs(process.argv.slice(2), {
594
- reportPath: defaultReportPath(),
595
- });
596
- const options = {};
597
- if (iterations !== undefined) {
598
- options.iterations = iterations;
599
- }
600
- const report = await runStoragePolicyBenchmarks(options);
601
- const reports = await writeStoragePolicyBenchmarkReports(report, reportPath, {
602
- jsonReportPath: jsonReportPath === undefined ? defaultJsonRunReportPath(reportPath) : jsonReportPath,
603
- });
604
- const shortId = createHash("sha256").update(JSON.stringify(report.summary)).digest("hex").slice(0, 8);
605
- console.log(
606
- `Freeflow storage policy benchmark ${shortId}: safe candidates ${report.summary.safeCandidateIds.join(", ") || "none"}`,
607
- );
608
- console.log(`Markdown report: ${reports.markdown}`);
609
- if (reports.json) {
610
- console.log(`JSON run data: ${reports.json}`);
611
- }
612
- if (report.summary.safeCandidateIds.length === 0) {
613
- process.exitCode = 1;
614
- }
615
- }
616
- if (process.argv[1] && fileURLToPath(import.meta.url) === resolve(process.argv[1])) {
617
- await runCli();
618
- }
@@ -1,56 +0,0 @@
1
- export interface RunVaultIndexStorageBenchmarkOptions {
2
- iterations?: number;
3
- generatedAt?: string;
4
- }
5
- export interface VaultIndexStorageBenchmarkReport {
6
- generatedAt: string;
7
- iterations: number;
8
- summary: VaultIndexStorageBenchmarkSummary;
9
- candidates: VaultIndexStorageCandidateResult[];
10
- }
11
- export interface VaultIndexStorageBenchmarkSummary {
12
- selectedEngine: string;
13
- selectedReason: string;
14
- fixtures: number;
15
- localSidecarPassed: boolean;
16
- scannerBaselinePassed: boolean;
17
- sqliteFtsStatus: "not-run" | "available" | "failed";
18
- localAppendMs: LatencyPair;
19
- localQueryMs: LatencyPair;
20
- scanQueryMs: LatencyPair;
21
- localQueryReductionPercent: number;
22
- }
23
- export interface LatencyPair {
24
- p50: number;
25
- p95: number;
26
- }
27
- export interface VaultIndexStorageCandidateResult {
28
- candidate: string;
29
- status: "pass" | "fail" | "not-run";
30
- adopted: boolean;
31
- appendMs?: LatencyPair;
32
- queryMs?: LatencyPair;
33
- rawBytes?: number;
34
- resultBytes?: number;
35
- reductionPercent?: number;
36
- checks: string[];
37
- notes: string[];
38
- }
39
- export declare function runVaultIndexStorageBenchmark(
40
- options?: RunVaultIndexStorageBenchmarkOptions,
41
- ): Promise<VaultIndexStorageBenchmarkReport>;
42
- export declare function renderVaultIndexStorageBenchmarkReport(report: VaultIndexStorageBenchmarkReport): string;
43
- export declare function writeVaultIndexStorageBenchmarkReport(
44
- report: VaultIndexStorageBenchmarkReport,
45
- reportPath: string,
46
- ): Promise<void>;
47
- export declare function writeVaultIndexStorageBenchmarkReports(
48
- report: VaultIndexStorageBenchmarkReport,
49
- markdownReportPath: string,
50
- options?: {
51
- jsonReportPath?: string | false;
52
- },
53
- ): Promise<{
54
- markdown: string;
55
- json?: string;
56
- }>;