audrey 1.1.3 → 1.2.1

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 (285) hide show
  1. package/CHANGELOG.md +97 -0
  2. package/README.md +39 -12
  3. package/dist/mcp-server/codex-hooks-probe.d.ts +22 -0
  4. package/dist/mcp-server/codex-hooks-probe.d.ts.map +1 -0
  5. package/dist/mcp-server/codex-hooks-probe.js +160 -0
  6. package/dist/mcp-server/codex-hooks-probe.js.map +1 -0
  7. package/dist/mcp-server/config.d.ts +26 -2
  8. package/dist/mcp-server/config.d.ts.map +1 -1
  9. package/dist/mcp-server/config.js +54 -22
  10. package/dist/mcp-server/config.js.map +1 -1
  11. package/dist/mcp-server/hooks.d.ts +28 -0
  12. package/dist/mcp-server/hooks.d.ts.map +1 -1
  13. package/dist/mcp-server/hooks.js +91 -2
  14. package/dist/mcp-server/hooks.js.map +1 -1
  15. package/dist/mcp-server/index.d.ts +242 -2
  16. package/dist/mcp-server/index.d.ts.map +1 -1
  17. package/dist/mcp-server/index.js +806 -240
  18. package/dist/mcp-server/index.js.map +1 -1
  19. package/dist/mcp-server/tool-schemas.d.ts +32 -2
  20. package/dist/mcp-server/tool-schemas.d.ts.map +1 -1
  21. package/dist/mcp-server/tool-schemas.js +55 -0
  22. package/dist/mcp-server/tool-schemas.js.map +1 -1
  23. package/dist/src/audrey.d.ts +14 -2
  24. package/dist/src/audrey.d.ts.map +1 -1
  25. package/dist/src/audrey.js +152 -28
  26. package/dist/src/audrey.js.map +1 -1
  27. package/dist/src/autopilot.d.ts +22 -0
  28. package/dist/src/autopilot.d.ts.map +1 -1
  29. package/dist/src/autopilot.js +271 -50
  30. package/dist/src/autopilot.js.map +1 -1
  31. package/dist/src/capsule.d.ts +23 -0
  32. package/dist/src/capsule.d.ts.map +1 -1
  33. package/dist/src/capsule.js +56 -11
  34. package/dist/src/capsule.js.map +1 -1
  35. package/dist/src/confidence.d.ts +9 -1
  36. package/dist/src/confidence.d.ts.map +1 -1
  37. package/dist/src/confidence.js +16 -2
  38. package/dist/src/confidence.js.map +1 -1
  39. package/dist/src/consolidate.d.ts +4 -0
  40. package/dist/src/consolidate.d.ts.map +1 -1
  41. package/dist/src/consolidate.js +206 -34
  42. package/dist/src/consolidate.js.map +1 -1
  43. package/dist/src/controller.d.ts +17 -1
  44. package/dist/src/controller.d.ts.map +1 -1
  45. package/dist/src/controller.js +244 -128
  46. package/dist/src/controller.js.map +1 -1
  47. package/dist/src/db.d.ts.map +1 -1
  48. package/dist/src/db.js +233 -56
  49. package/dist/src/db.js.map +1 -1
  50. package/dist/src/encode.d.ts +2 -1
  51. package/dist/src/encode.d.ts.map +1 -1
  52. package/dist/src/encode.js +41 -4
  53. package/dist/src/encode.js.map +1 -1
  54. package/dist/src/events.d.ts +48 -1
  55. package/dist/src/events.d.ts.map +1 -1
  56. package/dist/src/events.js +160 -62
  57. package/dist/src/events.js.map +1 -1
  58. package/dist/src/export.d.ts.map +1 -1
  59. package/dist/src/export.js +4 -4
  60. package/dist/src/export.js.map +1 -1
  61. package/dist/src/grounding.d.ts +111 -0
  62. package/dist/src/grounding.d.ts.map +1 -0
  63. package/dist/src/grounding.js +259 -0
  64. package/dist/src/grounding.js.map +1 -0
  65. package/dist/src/hybrid-recall.d.ts +16 -7
  66. package/dist/src/hybrid-recall.d.ts.map +1 -1
  67. package/dist/src/hybrid-recall.js +22 -11
  68. package/dist/src/hybrid-recall.js.map +1 -1
  69. package/dist/src/import.d.ts +10 -0
  70. package/dist/src/import.d.ts.map +1 -1
  71. package/dist/src/import.js +122 -22
  72. package/dist/src/import.js.map +1 -1
  73. package/dist/src/llm.d.ts.map +1 -1
  74. package/dist/src/llm.js +16 -0
  75. package/dist/src/llm.js.map +1 -1
  76. package/dist/src/preflight.d.ts.map +1 -1
  77. package/dist/src/preflight.js +89 -47
  78. package/dist/src/preflight.js.map +1 -1
  79. package/dist/src/project.d.ts +4 -1
  80. package/dist/src/project.d.ts.map +1 -1
  81. package/dist/src/project.js +65 -5
  82. package/dist/src/project.js.map +1 -1
  83. package/dist/src/recall.d.ts.map +1 -1
  84. package/dist/src/recall.js +138 -13
  85. package/dist/src/recall.js.map +1 -1
  86. package/dist/src/redact.d.ts +1 -1
  87. package/dist/src/redact.d.ts.map +1 -1
  88. package/dist/src/redact.js +185 -5
  89. package/dist/src/redact.js.map +1 -1
  90. package/dist/src/routes.d.ts.map +1 -1
  91. package/dist/src/routes.js +91 -18
  92. package/dist/src/routes.js.map +1 -1
  93. package/dist/src/trust.d.ts +35 -0
  94. package/dist/src/trust.d.ts.map +1 -0
  95. package/dist/src/trust.js +77 -0
  96. package/dist/src/trust.js.map +1 -0
  97. package/dist/src/types.d.ts +24 -0
  98. package/dist/src/types.d.ts.map +1 -1
  99. package/package.json +3 -37
  100. package/benchmarks/adapter-kit.mjs +0 -20
  101. package/benchmarks/adapter-self-test.mjs +0 -170
  102. package/benchmarks/adapters/example-allow.mjs +0 -31
  103. package/benchmarks/adapters/mem0-platform.mjs +0 -274
  104. package/benchmarks/adapters/registry.json +0 -51
  105. package/benchmarks/adapters/zep-cloud.mjs +0 -304
  106. package/benchmarks/baselines.js +0 -174
  107. package/benchmarks/build-leaderboard.mjs +0 -183
  108. package/benchmarks/cases.js +0 -549
  109. package/benchmarks/create-conformance-card.mjs +0 -148
  110. package/benchmarks/create-submission-bundle.mjs +0 -190
  111. package/benchmarks/dry-run-external-adapters.mjs +0 -177
  112. package/benchmarks/guardbench.js +0 -1265
  113. package/benchmarks/output/adapter-self-test/guardbench-adapter-self-test.json +0 -50
  114. package/benchmarks/output/external/guardbench-external-dry-run.json +0 -69
  115. package/benchmarks/output/external/guardbench-external-evidence.json +0 -56
  116. package/benchmarks/output/guardbench-conformance-card.json +0 -63
  117. package/benchmarks/output/guardbench-manifest.json +0 -414
  118. package/benchmarks/output/guardbench-raw.json +0 -1279
  119. package/benchmarks/output/guardbench-summary.json +0 -2123
  120. package/benchmarks/output/leaderboard/guardbench-leaderboard.json +0 -93
  121. package/benchmarks/output/leaderboard/guardbench-leaderboard.md +0 -7
  122. package/benchmarks/output/submission-bundle/guardbench-conformance-card.json +0 -63
  123. package/benchmarks/output/submission-bundle/guardbench-manifest.json +0 -414
  124. package/benchmarks/output/submission-bundle/guardbench-raw.json +0 -1279
  125. package/benchmarks/output/submission-bundle/guardbench-summary.json +0 -2123
  126. package/benchmarks/output/submission-bundle/schemas/guardbench-adapter-registry.schema.json +0 -69
  127. package/benchmarks/output/submission-bundle/schemas/guardbench-adapter-self-test.schema.json +0 -156
  128. package/benchmarks/output/submission-bundle/schemas/guardbench-conformance-card.schema.json +0 -184
  129. package/benchmarks/output/submission-bundle/schemas/guardbench-external-dry-run.schema.json +0 -74
  130. package/benchmarks/output/submission-bundle/schemas/guardbench-external-evidence.schema.json +0 -108
  131. package/benchmarks/output/submission-bundle/schemas/guardbench-external-run.schema.json +0 -160
  132. package/benchmarks/output/submission-bundle/schemas/guardbench-leaderboard.schema.json +0 -179
  133. package/benchmarks/output/submission-bundle/schemas/guardbench-manifest.schema.json +0 -213
  134. package/benchmarks/output/submission-bundle/schemas/guardbench-publication-verification.schema.json +0 -47
  135. package/benchmarks/output/submission-bundle/schemas/guardbench-raw.schema.json +0 -184
  136. package/benchmarks/output/submission-bundle/schemas/guardbench-submission-manifest.schema.json +0 -151
  137. package/benchmarks/output/submission-bundle/schemas/guardbench-summary.schema.json +0 -249
  138. package/benchmarks/output/submission-bundle/submission-manifest.json +0 -131
  139. package/benchmarks/output/submission-bundle/validation-report.json +0 -31
  140. package/benchmarks/output/summary.json +0 -2354
  141. package/benchmarks/perf-snapshot.js +0 -307
  142. package/benchmarks/perf.bench.js +0 -169
  143. package/benchmarks/public-paths.mjs +0 -97
  144. package/benchmarks/reference-results.js +0 -75
  145. package/benchmarks/report.js +0 -280
  146. package/benchmarks/run-external-guardbench.mjs +0 -303
  147. package/benchmarks/run.js +0 -735
  148. package/benchmarks/schemas/guardbench-adapter-registry.schema.json +0 -69
  149. package/benchmarks/schemas/guardbench-adapter-self-test.schema.json +0 -156
  150. package/benchmarks/schemas/guardbench-conformance-card.schema.json +0 -184
  151. package/benchmarks/schemas/guardbench-external-dry-run.schema.json +0 -74
  152. package/benchmarks/schemas/guardbench-external-evidence.schema.json +0 -108
  153. package/benchmarks/schemas/guardbench-external-run.schema.json +0 -160
  154. package/benchmarks/schemas/guardbench-leaderboard.schema.json +0 -179
  155. package/benchmarks/schemas/guardbench-manifest.schema.json +0 -213
  156. package/benchmarks/schemas/guardbench-publication-verification.schema.json +0 -47
  157. package/benchmarks/schemas/guardbench-raw.schema.json +0 -184
  158. package/benchmarks/schemas/guardbench-submission-manifest.schema.json +0 -151
  159. package/benchmarks/schemas/guardbench-summary.schema.json +0 -249
  160. package/benchmarks/snapshots/perf-0.22.2.json +0 -123
  161. package/benchmarks/snapshots/perf-0.23.0.json +0 -123
  162. package/benchmarks/validate-adapter-module.mjs +0 -107
  163. package/benchmarks/validate-adapter-registry.mjs +0 -145
  164. package/benchmarks/validate-adapter-self-test.mjs +0 -96
  165. package/benchmarks/validate-guardbench-artifacts.mjs +0 -400
  166. package/benchmarks/verify-external-evidence.mjs +0 -351
  167. package/benchmarks/verify-publication-artifacts.mjs +0 -309
  168. package/benchmarks/verify-submission-bundle.mjs +0 -172
  169. package/docs/AUDREY_PAPER_OUTLINE.md +0 -178
  170. package/docs/MEMORY_BENCHMARKING.md +0 -59
  171. package/docs/PRODUCTION_BACKLOG.md +0 -321
  172. package/docs/assets/audrey-feature-grid.jpg +0 -0
  173. package/docs/assets/audrey-logo.svg +0 -45
  174. package/docs/assets/audrey-wordmark.png +0 -0
  175. package/docs/paper/00-master.md +0 -48
  176. package/docs/paper/01-introduction.md +0 -27
  177. package/docs/paper/02-related-work.md +0 -47
  178. package/docs/paper/03-problem-definition.md +0 -108
  179. package/docs/paper/04-design.md +0 -164
  180. package/docs/paper/05-guardbench-spec.md +0 -412
  181. package/docs/paper/06-implementation.md +0 -115
  182. package/docs/paper/07-evaluation.md +0 -168
  183. package/docs/paper/08-discussion-limitations.md +0 -61
  184. package/docs/paper/09-conclusion.md +0 -11
  185. package/docs/paper/SUBMISSION_README.md +0 -162
  186. package/docs/paper/appendix-a-demo-transcript.md +0 -114
  187. package/docs/paper/arxiv-compile-report.schema.json +0 -116
  188. package/docs/paper/arxiv-source.schema.json +0 -61
  189. package/docs/paper/audrey-paper-v1.md +0 -1108
  190. package/docs/paper/browser-launch-plan.json +0 -209
  191. package/docs/paper/browser-launch-plan.schema.json +0 -100
  192. package/docs/paper/browser-launch-results.json +0 -86
  193. package/docs/paper/browser-launch-results.schema.json +0 -66
  194. package/docs/paper/claim-register.json +0 -138
  195. package/docs/paper/claim-register.schema.json +0 -81
  196. package/docs/paper/evidence-ledger.md +0 -103
  197. package/docs/paper/output/arxiv/README-arxiv.txt +0 -8
  198. package/docs/paper/output/arxiv/arxiv-manifest.json +0 -41
  199. package/docs/paper/output/arxiv/main.tex +0 -951
  200. package/docs/paper/output/arxiv/references.bib +0 -222
  201. package/docs/paper/output/arxiv-compile-report.json +0 -24
  202. package/docs/paper/output/submission-bundle/LICENSE +0 -21
  203. package/docs/paper/output/submission-bundle/README.md +0 -440
  204. package/docs/paper/output/submission-bundle/benchmarks/output/adapter-self-test/guardbench-adapter-self-test.json +0 -50
  205. package/docs/paper/output/submission-bundle/benchmarks/output/external/guardbench-external-dry-run.json +0 -69
  206. package/docs/paper/output/submission-bundle/benchmarks/output/external/guardbench-external-evidence.json +0 -56
  207. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-conformance-card.json +0 -63
  208. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-manifest.json +0 -414
  209. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-raw.json +0 -1279
  210. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-summary.json +0 -2123
  211. package/docs/paper/output/submission-bundle/benchmarks/output/leaderboard/guardbench-leaderboard.json +0 -93
  212. package/docs/paper/output/submission-bundle/benchmarks/output/leaderboard/guardbench-leaderboard.md +0 -7
  213. package/docs/paper/output/submission-bundle/benchmarks/output/submission-bundle/submission-manifest.json +0 -131
  214. package/docs/paper/output/submission-bundle/benchmarks/output/submission-bundle/validation-report.json +0 -31
  215. package/docs/paper/output/submission-bundle/benchmarks/output/summary.json +0 -2354
  216. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-adapter-registry.schema.json +0 -69
  217. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-adapter-self-test.schema.json +0 -156
  218. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-conformance-card.schema.json +0 -184
  219. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-external-dry-run.schema.json +0 -74
  220. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-external-evidence.schema.json +0 -108
  221. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-external-run.schema.json +0 -160
  222. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-leaderboard.schema.json +0 -179
  223. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-manifest.schema.json +0 -213
  224. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-publication-verification.schema.json +0 -47
  225. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-raw.schema.json +0 -184
  226. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-submission-manifest.schema.json +0 -151
  227. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-summary.schema.json +0 -249
  228. package/docs/paper/output/submission-bundle/docs/AUDREY_PAPER_OUTLINE.md +0 -178
  229. package/docs/paper/output/submission-bundle/docs/paper/00-master.md +0 -48
  230. package/docs/paper/output/submission-bundle/docs/paper/01-introduction.md +0 -27
  231. package/docs/paper/output/submission-bundle/docs/paper/02-related-work.md +0 -47
  232. package/docs/paper/output/submission-bundle/docs/paper/03-problem-definition.md +0 -108
  233. package/docs/paper/output/submission-bundle/docs/paper/04-design.md +0 -164
  234. package/docs/paper/output/submission-bundle/docs/paper/05-guardbench-spec.md +0 -412
  235. package/docs/paper/output/submission-bundle/docs/paper/06-implementation.md +0 -115
  236. package/docs/paper/output/submission-bundle/docs/paper/07-evaluation.md +0 -168
  237. package/docs/paper/output/submission-bundle/docs/paper/08-discussion-limitations.md +0 -61
  238. package/docs/paper/output/submission-bundle/docs/paper/09-conclusion.md +0 -11
  239. package/docs/paper/output/submission-bundle/docs/paper/SUBMISSION_README.md +0 -162
  240. package/docs/paper/output/submission-bundle/docs/paper/appendix-a-demo-transcript.md +0 -114
  241. package/docs/paper/output/submission-bundle/docs/paper/arxiv-compile-report.schema.json +0 -116
  242. package/docs/paper/output/submission-bundle/docs/paper/arxiv-source.schema.json +0 -61
  243. package/docs/paper/output/submission-bundle/docs/paper/audrey-paper-v1.md +0 -1108
  244. package/docs/paper/output/submission-bundle/docs/paper/browser-launch-plan.json +0 -209
  245. package/docs/paper/output/submission-bundle/docs/paper/browser-launch-plan.schema.json +0 -100
  246. package/docs/paper/output/submission-bundle/docs/paper/browser-launch-results.json +0 -86
  247. package/docs/paper/output/submission-bundle/docs/paper/browser-launch-results.schema.json +0 -66
  248. package/docs/paper/output/submission-bundle/docs/paper/claim-register.json +0 -138
  249. package/docs/paper/output/submission-bundle/docs/paper/claim-register.schema.json +0 -81
  250. package/docs/paper/output/submission-bundle/docs/paper/evidence-ledger.md +0 -103
  251. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv/README-arxiv.txt +0 -8
  252. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv/arxiv-manifest.json +0 -41
  253. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv/main.tex +0 -951
  254. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv/references.bib +0 -222
  255. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv-compile-report.json +0 -24
  256. package/docs/paper/output/submission-bundle/docs/paper/paper-submission-bundle.schema.json +0 -70
  257. package/docs/paper/output/submission-bundle/docs/paper/publication-pack.json +0 -81
  258. package/docs/paper/output/submission-bundle/docs/paper/publication-pack.schema.json +0 -60
  259. package/docs/paper/output/submission-bundle/docs/paper/references.bib +0 -222
  260. package/docs/paper/output/submission-bundle/package.json +0 -235
  261. package/docs/paper/output/submission-bundle/paper-submission-manifest.json +0 -379
  262. package/docs/paper/paper-submission-bundle.schema.json +0 -70
  263. package/docs/paper/publication-pack.json +0 -81
  264. package/docs/paper/publication-pack.schema.json +0 -60
  265. package/docs/paper/references.bib +0 -222
  266. package/scripts/audit-release-completion.mjs +0 -440
  267. package/scripts/create-arxiv-source.mjs +0 -368
  268. package/scripts/create-paper-submission-bundle.mjs +0 -214
  269. package/scripts/finalize-release.mjs +0 -601
  270. package/scripts/prepare-release-cut.mjs +0 -277
  271. package/scripts/publish-release-bundle.mjs +0 -252
  272. package/scripts/publish-release-github-api.mjs +0 -494
  273. package/scripts/run-vitest.mjs +0 -34
  274. package/scripts/smoke-cli.js +0 -92
  275. package/scripts/sync-paper-artifacts.mjs +0 -112
  276. package/scripts/verify-arxiv-compile.mjs +0 -476
  277. package/scripts/verify-arxiv-source.mjs +0 -224
  278. package/scripts/verify-browser-launch-plan.mjs +0 -254
  279. package/scripts/verify-browser-launch-results.mjs +0 -303
  280. package/scripts/verify-paper-artifacts.mjs +0 -798
  281. package/scripts/verify-paper-claims.mjs +0 -254
  282. package/scripts/verify-paper-submission-bundle.mjs +0 -218
  283. package/scripts/verify-publication-pack.mjs +0 -210
  284. package/scripts/verify-python-package.py +0 -201
  285. package/scripts/verify-release-readiness.mjs +0 -920
@@ -1,184 +0,0 @@
1
- {
2
- "$schema": "https://json-schema.org/draft/2020-12/schema",
3
- "$id": "https://audrey-memory.org/schemas/guardbench-raw.schema.json",
4
- "title": "GuardBench Raw Output",
5
- "type": "object",
6
- "additionalProperties": false,
7
- "required": [
8
- "suite",
9
- "generatedAt",
10
- "manifestVersion",
11
- "provenance",
12
- "cases",
13
- "artifactRedactionSweep"
14
- ],
15
- "properties": {
16
- "suite": { "const": "GuardBench comparative" },
17
- "generatedAt": { "type": "string", "minLength": 1 },
18
- "manifestVersion": { "type": "string", "minLength": 1 },
19
- "provenance": { "$ref": "#/$defs/provenance" },
20
- "cases": {
21
- "type": "array",
22
- "minItems": 10,
23
- "items": { "$ref": "#/$defs/caseResult" }
24
- },
25
- "artifactRedactionSweep": { "$ref": "#/$defs/artifactRedactionSweep" }
26
- },
27
- "$defs": {
28
- "jsonValue": {
29
- "anyOf": [
30
- { "type": "null" },
31
- { "type": "string" },
32
- { "type": "boolean" },
33
- { "type": "number" },
34
- {
35
- "type": "array",
36
- "items": { "$ref": "#/$defs/jsonValue" }
37
- },
38
- { "$ref": "#/$defs/jsonObject" }
39
- ]
40
- },
41
- "jsonObject": {
42
- "type": "object",
43
- "additionalProperties": { "$ref": "#/$defs/jsonValue" }
44
- },
45
- "provenance": {
46
- "type": "object",
47
- "additionalProperties": false,
48
- "required": [
49
- "generatedAt",
50
- "gitSha",
51
- "gitDirty",
52
- "node",
53
- "v8",
54
- "platform",
55
- "arch",
56
- "osRelease",
57
- "cpuModel",
58
- "cpuCount",
59
- "totalMemoryGb",
60
- "embeddingProvider",
61
- "embeddingDimensions",
62
- "llmProvider"
63
- ],
64
- "properties": {
65
- "generatedAt": { "type": "string", "minLength": 1 },
66
- "gitSha": {
67
- "anyOf": [
68
- { "type": "null" },
69
- { "type": "string", "minLength": 7 }
70
- ]
71
- },
72
- "gitDirty": { "type": "boolean" },
73
- "node": { "type": "string", "minLength": 1 },
74
- "v8": { "type": "string", "minLength": 1 },
75
- "platform": { "type": "string", "minLength": 1 },
76
- "arch": { "type": "string", "minLength": 1 },
77
- "osRelease": { "type": "string", "minLength": 1 },
78
- "cpuModel": {
79
- "anyOf": [
80
- { "type": "null" },
81
- { "type": "string", "minLength": 1 }
82
- ]
83
- },
84
- "cpuCount": { "type": "integer", "minimum": 1 },
85
- "totalMemoryGb": { "type": "number", "minimum": 0 },
86
- "embeddingProvider": { "type": "string", "minLength": 1 },
87
- "embeddingDimensions": { "type": "integer", "minimum": 1 },
88
- "llmProvider": { "type": "string", "minLength": 1 }
89
- }
90
- },
91
- "caseResult": {
92
- "type": "object",
93
- "additionalProperties": false,
94
- "required": ["id", "name", "expectedDecision", "results"],
95
- "properties": {
96
- "id": { "type": "string", "pattern": "^GB-[0-9]{2}$" },
97
- "name": { "type": "string", "minLength": 1 },
98
- "expectedDecision": { "enum": ["allow", "warn", "block"] },
99
- "results": {
100
- "type": "array",
101
- "minItems": 5,
102
- "items": { "$ref": "#/$defs/resultRow" }
103
- }
104
- }
105
- },
106
- "resultRow": {
107
- "type": "object",
108
- "additionalProperties": false,
109
- "required": [
110
- "system",
111
- "id",
112
- "name",
113
- "expectedDecision",
114
- "decision",
115
- "decisionCorrect",
116
- "riskScore",
117
- "passed",
118
- "latencyMs",
119
- "evidenceCount",
120
- "evidenceIds",
121
- "recommendedActions",
122
- "summary",
123
- "recallErrors",
124
- "leakedSecrets",
125
- "requiredEvidenceMatched"
126
- ],
127
- "properties": {
128
- "system": { "type": "string", "minLength": 1 },
129
- "external": { "type": "boolean" },
130
- "id": { "type": "string", "pattern": "^GB-[0-9]{2}$" },
131
- "name": { "type": "string", "minLength": 1 },
132
- "expectedDecision": { "enum": ["allow", "warn", "block"] },
133
- "decision": { "enum": ["allow", "warn", "block"] },
134
- "decisionCorrect": { "type": "boolean" },
135
- "riskScore": { "type": "number", "minimum": 0, "maximum": 1 },
136
- "passed": { "type": "boolean" },
137
- "latencyMs": { "type": "number", "minimum": 0 },
138
- "evidenceCount": { "type": "integer", "minimum": 0 },
139
- "evidenceIds": {
140
- "type": "array",
141
- "items": { "type": "string" }
142
- },
143
- "recommendedActions": {
144
- "type": "array",
145
- "items": { "type": "string" }
146
- },
147
- "summary": { "type": "string", "minLength": 1 },
148
- "recallErrors": { "type": "array" },
149
- "adapterExtensions": { "$ref": "#/$defs/jsonObject" },
150
- "leakedSecrets": {
151
- "type": "array",
152
- "items": { "type": "string" }
153
- },
154
- "requiredEvidenceMatched": { "type": "boolean" },
155
- "hasEvidenceForDecision": { "type": "boolean" },
156
- "lineageTextMatched": { "type": "boolean" }
157
- }
158
- },
159
- "artifactRedactionSweep": {
160
- "type": "object",
161
- "additionalProperties": false,
162
- "required": [
163
- "checkedAt",
164
- "filesChecked",
165
- "seededSecretsChecked",
166
- "leakCount",
167
- "leaks",
168
- "passed"
169
- ],
170
- "properties": {
171
- "checkedAt": { "type": "string", "minLength": 1 },
172
- "filesChecked": {
173
- "type": "array",
174
- "minItems": 3,
175
- "items": { "type": "string", "minLength": 1 }
176
- },
177
- "seededSecretsChecked": { "type": "integer", "minimum": 0 },
178
- "leakCount": { "type": "integer", "minimum": 0 },
179
- "leaks": { "type": "array" },
180
- "passed": { "type": "boolean" }
181
- }
182
- }
183
- }
184
- }
@@ -1,151 +0,0 @@
1
- {
2
- "$schema": "https://json-schema.org/draft/2020-12/schema",
3
- "$id": "https://audrey-memory.org/schemas/guardbench-submission-manifest.schema.json",
4
- "title": "GuardBench Submission Manifest",
5
- "type": "object",
6
- "additionalProperties": false,
7
- "required": [
8
- "schemaVersion",
9
- "suite",
10
- "generatedAt",
11
- "sourceDir",
12
- "subject",
13
- "score",
14
- "conformance",
15
- "validation",
16
- "files"
17
- ],
18
- "properties": {
19
- "schemaVersion": { "const": "1.0.0" },
20
- "suite": { "const": "GuardBench submission bundle" },
21
- "generatedAt": { "type": "string", "minLength": 1 },
22
- "sourceDir": { "type": "string", "minLength": 1 },
23
- "subject": { "$ref": "#/$defs/subject" },
24
- "score": { "$ref": "#/$defs/score" },
25
- "conformance": { "$ref": "#/$defs/conformance" },
26
- "validation": { "$ref": "#/$defs/validation" },
27
- "files": {
28
- "type": "array",
29
- "minItems": 1,
30
- "items": { "$ref": "#/$defs/fileRecord" }
31
- }
32
- },
33
- "$defs": {
34
- "subject": {
35
- "type": "object",
36
- "additionalProperties": false,
37
- "required": ["name", "requestedAdapter", "external"],
38
- "properties": {
39
- "name": { "type": "string", "minLength": 1 },
40
- "requestedAdapter": {
41
- "anyOf": [
42
- { "type": "string", "minLength": 1 },
43
- { "type": "null" }
44
- ]
45
- },
46
- "external": { "type": "boolean" }
47
- }
48
- },
49
- "score": {
50
- "type": "object",
51
- "additionalProperties": false,
52
- "required": [
53
- "scenarios",
54
- "fullContractPassed",
55
- "fullContractPassRate",
56
- "decisionAccuracy",
57
- "evidenceRecall",
58
- "redactionLeaks",
59
- "latency"
60
- ],
61
- "properties": {
62
- "scenarios": { "type": "integer", "minimum": 0 },
63
- "fullContractPassed": { "$ref": "#/$defs/nullableNumber" },
64
- "fullContractPassRate": { "$ref": "#/$defs/nullableRate" },
65
- "decisionAccuracy": { "$ref": "#/$defs/nullableRate" },
66
- "evidenceRecall": { "$ref": "#/$defs/nullableRate" },
67
- "redactionLeaks": { "$ref": "#/$defs/nullableNumber" },
68
- "latency": {
69
- "anyOf": [
70
- {
71
- "type": "object",
72
- "additionalProperties": false,
73
- "required": ["p50Ms", "p95Ms", "maxMs"],
74
- "properties": {
75
- "p50Ms": { "type": "number", "minimum": 0 },
76
- "p95Ms": { "type": "number", "minimum": 0 },
77
- "maxMs": { "type": "number", "minimum": 0 }
78
- }
79
- },
80
- { "type": "null" }
81
- ]
82
- }
83
- }
84
- },
85
- "conformance": {
86
- "type": "object",
87
- "additionalProperties": false,
88
- "required": ["ok", "failures", "artifactValidationOk", "artifactValidationFailures"],
89
- "properties": {
90
- "ok": { "type": "boolean" },
91
- "failures": {
92
- "type": "array",
93
- "items": { "type": "string" }
94
- },
95
- "artifactValidationOk": { "type": "boolean" },
96
- "artifactValidationFailures": {
97
- "type": "array",
98
- "items": { "type": "string" }
99
- }
100
- }
101
- },
102
- "validation": {
103
- "type": "object",
104
- "additionalProperties": false,
105
- "required": ["ok", "dir", "schemasDir", "files", "optionalFiles", "failures"],
106
- "properties": {
107
- "ok": { "type": "boolean" },
108
- "dir": { "type": "string", "minLength": 1 },
109
- "schemasDir": { "type": "string", "minLength": 1 },
110
- "files": {
111
- "type": "array",
112
- "items": { "type": "string", "minLength": 1 }
113
- },
114
- "optionalFiles": {
115
- "type": "array",
116
- "items": { "type": "string", "minLength": 1 }
117
- },
118
- "failures": {
119
- "type": "array",
120
- "items": { "type": "string" }
121
- }
122
- }
123
- },
124
- "fileRecord": {
125
- "type": "object",
126
- "additionalProperties": false,
127
- "required": ["path", "bytes", "sha256"],
128
- "properties": {
129
- "path": { "type": "string", "minLength": 1 },
130
- "bytes": { "type": "integer", "minimum": 0 },
131
- "sha256": { "$ref": "#/$defs/sha256" }
132
- }
133
- },
134
- "nullableNumber": {
135
- "anyOf": [
136
- { "type": "number", "minimum": 0 },
137
- { "type": "null" }
138
- ]
139
- },
140
- "nullableRate": {
141
- "anyOf": [
142
- { "type": "number", "minimum": 0, "maximum": 1 },
143
- { "type": "null" }
144
- ]
145
- },
146
- "sha256": {
147
- "type": "string",
148
- "pattern": "^[a-f0-9]{64}$"
149
- }
150
- }
151
- }
@@ -1,249 +0,0 @@
1
- {
2
- "$schema": "https://json-schema.org/draft/2020-12/schema",
3
- "$id": "https://audrey-memory.org/schemas/guardbench-summary.schema.json",
4
- "title": "GuardBench Summary",
5
- "type": "object",
6
- "additionalProperties": false,
7
- "required": [
8
- "suite",
9
- "generatedAt",
10
- "manifest",
11
- "provenance",
12
- "subjects",
13
- "scenarios",
14
- "passed",
15
- "passRate",
16
- "preventionRate",
17
- "falseBlockRate",
18
- "decisionAccuracy",
19
- "evidenceRecall",
20
- "redactionLeaks",
21
- "recallDegradationDetectionRate",
22
- "latency",
23
- "systemSummaries",
24
- "comparisons",
25
- "rows",
26
- "cases",
27
- "artifactRedactionSweep"
28
- ],
29
- "properties": {
30
- "suite": { "const": "GuardBench comparative" },
31
- "generatedAt": { "type": "string", "minLength": 1 },
32
- "manifest": { "type": "object" },
33
- "provenance": { "type": "object" },
34
- "subjects": {
35
- "type": "array",
36
- "minItems": 5,
37
- "items": { "type": "string", "minLength": 1 }
38
- },
39
- "scenarios": { "type": "integer", "minimum": 10 },
40
- "passed": { "type": "integer", "minimum": 0 },
41
- "passRate": { "type": "number", "minimum": 0, "maximum": 1 },
42
- "preventionRate": { "type": "number", "minimum": 0, "maximum": 1 },
43
- "falseBlockRate": { "type": "number", "minimum": 0, "maximum": 1 },
44
- "decisionAccuracy": { "type": "number", "minimum": 0, "maximum": 1 },
45
- "usefulWarningPrecision": {
46
- "anyOf": [
47
- { "type": "null" },
48
- { "type": "number", "minimum": 0, "maximum": 1 }
49
- ]
50
- },
51
- "evidenceRecall": { "type": "number", "minimum": 0, "maximum": 1 },
52
- "redactionLeaks": { "type": "integer", "minimum": 0 },
53
- "recallDegradationDetectionRate": { "type": "number", "minimum": 0, "maximum": 1 },
54
- "latency": { "$ref": "#/$defs/latency" },
55
- "systemSummaries": {
56
- "type": "array",
57
- "minItems": 5,
58
- "items": { "$ref": "#/$defs/systemSummary" }
59
- },
60
- "comparisons": {
61
- "type": "object",
62
- "additionalProperties": false,
63
- "required": ["bestBaseline", "audreyMarginOverBestBaseline"],
64
- "properties": {
65
- "bestBaseline": { "$ref": "#/$defs/systemSummary" },
66
- "audreyMarginOverBestBaseline": {
67
- "anyOf": [
68
- { "type": "null" },
69
- { "type": "number", "minimum": 0, "maximum": 1 }
70
- ]
71
- }
72
- }
73
- },
74
- "rows": {
75
- "type": "array",
76
- "minItems": 10,
77
- "items": { "$ref": "#/$defs/resultRow" }
78
- },
79
- "cases": {
80
- "type": "array",
81
- "minItems": 10,
82
- "items": { "$ref": "#/$defs/caseResult" }
83
- },
84
- "artifactRedactionSweep": { "$ref": "#/$defs/artifactRedactionSweep" }
85
- },
86
- "$defs": {
87
- "jsonValue": {
88
- "anyOf": [
89
- { "type": "null" },
90
- { "type": "string" },
91
- { "type": "boolean" },
92
- { "type": "number" },
93
- {
94
- "type": "array",
95
- "items": { "$ref": "#/$defs/jsonValue" }
96
- },
97
- { "$ref": "#/$defs/jsonObject" }
98
- ]
99
- },
100
- "jsonObject": {
101
- "type": "object",
102
- "additionalProperties": { "$ref": "#/$defs/jsonValue" }
103
- },
104
- "latency": {
105
- "type": "object",
106
- "additionalProperties": false,
107
- "required": ["p50Ms", "p95Ms", "maxMs"],
108
- "properties": {
109
- "p50Ms": { "type": "number", "minimum": 0 },
110
- "p95Ms": { "type": "number", "minimum": 0 },
111
- "maxMs": { "type": "number", "minimum": 0 }
112
- }
113
- },
114
- "systemSummary": {
115
- "type": "object",
116
- "additionalProperties": false,
117
- "required": [
118
- "system",
119
- "generatedAt",
120
- "scenarios",
121
- "passed",
122
- "passRate",
123
- "decisionCorrect",
124
- "decisionAccuracy",
125
- "preventionRate",
126
- "falseBlockRate",
127
- "usefulWarningPrecision",
128
- "evidenceRecall",
129
- "redactionLeaks",
130
- "recallDegradationDetectionRate",
131
- "latency"
132
- ],
133
- "properties": {
134
- "system": { "type": "string", "minLength": 1 },
135
- "generatedAt": { "type": "string", "minLength": 1 },
136
- "scenarios": { "type": "integer", "minimum": 0 },
137
- "passed": { "type": "integer", "minimum": 0 },
138
- "passRate": { "type": "number", "minimum": 0, "maximum": 1 },
139
- "decisionCorrect": { "type": "integer", "minimum": 0 },
140
- "decisionAccuracy": { "type": "number", "minimum": 0, "maximum": 1 },
141
- "preventionRate": { "type": "number", "minimum": 0, "maximum": 1 },
142
- "falseBlockRate": { "type": "number", "minimum": 0, "maximum": 1 },
143
- "usefulWarningPrecision": {
144
- "anyOf": [
145
- { "type": "null" },
146
- { "type": "number", "minimum": 0, "maximum": 1 }
147
- ]
148
- },
149
- "evidenceRecall": { "type": "number", "minimum": 0, "maximum": 1 },
150
- "redactionLeaks": { "type": "integer", "minimum": 0 },
151
- "recallDegradationDetectionRate": { "type": "number", "minimum": 0, "maximum": 1 },
152
- "latency": { "$ref": "#/$defs/latency" },
153
- "lineageRichness": { "type": "number", "minimum": 0, "maximum": 1 }
154
- }
155
- },
156
- "resultRow": {
157
- "type": "object",
158
- "additionalProperties": false,
159
- "required": [
160
- "system",
161
- "id",
162
- "name",
163
- "expectedDecision",
164
- "decision",
165
- "decisionCorrect",
166
- "riskScore",
167
- "passed",
168
- "latencyMs",
169
- "evidenceCount",
170
- "evidenceIds",
171
- "recommendedActions",
172
- "summary",
173
- "recallErrors",
174
- "leakedSecrets",
175
- "requiredEvidenceMatched"
176
- ],
177
- "properties": {
178
- "system": { "type": "string", "minLength": 1 },
179
- "external": { "type": "boolean" },
180
- "id": { "type": "string", "pattern": "^GB-[0-9]{2}$" },
181
- "name": { "type": "string", "minLength": 1 },
182
- "expectedDecision": { "enum": ["allow", "warn", "block"] },
183
- "decision": { "enum": ["allow", "warn", "block"] },
184
- "decisionCorrect": { "type": "boolean" },
185
- "riskScore": { "type": "number", "minimum": 0, "maximum": 1 },
186
- "passed": { "type": "boolean" },
187
- "latencyMs": { "type": "number", "minimum": 0 },
188
- "evidenceCount": { "type": "integer", "minimum": 0 },
189
- "evidenceIds": {
190
- "type": "array",
191
- "items": { "type": "string" }
192
- },
193
- "recommendedActions": {
194
- "type": "array",
195
- "items": { "type": "string" }
196
- },
197
- "summary": { "type": "string", "minLength": 1 },
198
- "recallErrors": { "type": "array" },
199
- "adapterExtensions": { "$ref": "#/$defs/jsonObject" },
200
- "leakedSecrets": {
201
- "type": "array",
202
- "items": { "type": "string" }
203
- },
204
- "requiredEvidenceMatched": { "type": "boolean" },
205
- "hasEvidenceForDecision": { "type": "boolean" },
206
- "lineageTextMatched": { "type": "boolean" }
207
- }
208
- },
209
- "caseResult": {
210
- "type": "object",
211
- "additionalProperties": false,
212
- "required": ["id", "name", "expectedDecision", "results"],
213
- "properties": {
214
- "id": { "type": "string", "pattern": "^GB-[0-9]{2}$" },
215
- "name": { "type": "string", "minLength": 1 },
216
- "expectedDecision": { "enum": ["allow", "warn", "block"] },
217
- "results": {
218
- "type": "array",
219
- "minItems": 5,
220
- "items": { "$ref": "#/$defs/resultRow" }
221
- }
222
- }
223
- },
224
- "artifactRedactionSweep": {
225
- "type": "object",
226
- "additionalProperties": false,
227
- "required": [
228
- "checkedAt",
229
- "filesChecked",
230
- "seededSecretsChecked",
231
- "leakCount",
232
- "leaks",
233
- "passed"
234
- ],
235
- "properties": {
236
- "checkedAt": { "type": "string", "minLength": 1 },
237
- "filesChecked": {
238
- "type": "array",
239
- "minItems": 3,
240
- "items": { "type": "string", "minLength": 1 }
241
- },
242
- "seededSecretsChecked": { "type": "integer", "minimum": 0 },
243
- "leakCount": { "type": "integer", "minimum": 0 },
244
- "leaks": { "type": "array" },
245
- "passed": { "type": "boolean" }
246
- }
247
- }
248
- }
249
- }