dsh-memu-memory 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 (184) hide show
  1. package/AGENTS.md +1 -1
  2. package/README.md +39 -6
  3. package/README.zh.md +39 -7
  4. package/docs/active-memory-ab.md +385 -0
  5. package/docs/binding-lifecycle.md +99 -0
  6. package/docs/contributing.md +3 -3
  7. package/docs/emotion-model-spec.md +281 -0
  8. package/docs/env-and-commands.md +8 -0
  9. package/docs/host-api.md +2 -2
  10. package/docs/memory-semantics.md +12 -10
  11. package/docs/recall-benchmark.md +238 -0
  12. package/docs/troubleshooting.md +396 -395
  13. package/lib/active-memory.js +165 -0
  14. package/lib/activity-log.js +23 -6
  15. package/lib/binding-diagnostics.js +8 -3
  16. package/lib/client.js +335 -35
  17. package/lib/consolidation-ledger.js +347 -0
  18. package/lib/consolidation-receipt.js +284 -0
  19. package/lib/consolidation-snapshot.js +274 -0
  20. package/lib/consolidation-worker.js +144 -0
  21. package/lib/curator.js +86 -31
  22. package/lib/cycle.js +40 -6
  23. package/lib/graph.js +139 -5
  24. package/lib/index.js +33 -1
  25. package/lib/inject.js +104 -76
  26. package/lib/local-embedding-manager.js +24 -6
  27. package/lib/memory.js +534 -88
  28. package/lib/mood.js +268 -39
  29. package/lib/preset-scope.js +1 -1
  30. package/lib/private.js +430 -43
  31. package/lib/prompts/private-consolidation.coding.md +1 -0
  32. package/lib/prompts/private-consolidation.md +1 -0
  33. package/lib/prompts/private-consolidation.mixed.md +1 -0
  34. package/lib/prompts/private-consolidation.rp.md +1 -0
  35. package/lib/prompts.js +4 -4
  36. package/lib/remote-summary.js +51 -25
  37. package/lib/rpc.js +54 -5
  38. package/lib/state.js +29 -12
  39. package/lib/text-similarity.js +106 -0
  40. package/lib/tools.js +113 -10
  41. package/lib/vector-migration.js +28 -3
  42. package/lib/vector-profile.js +51 -0
  43. package/package.json +4 -2
  44. package/scripts/acceptance-local.mjs +2 -1
  45. package/scripts/active-memory-ab-live.mjs +655 -0
  46. package/scripts/active-memory-ab.mjs +1302 -0
  47. package/scripts/chain-matrix.mjs +11 -1
  48. package/scripts/check-all.mjs +10 -1
  49. package/scripts/replay.mjs +573 -0
  50. package/tests/activation.test.mjs +89 -0
  51. package/tests/active-memory-ab.test.mjs +745 -0
  52. package/tests/active-memory.test.mjs +129 -0
  53. package/tests/activity-log.test.mjs +126 -0
  54. package/tests/activity-wiring.test.mjs +103 -0
  55. package/tests/appendix-b.test.mjs +216 -0
  56. package/tests/appendix-b2.test.mjs +166 -0
  57. package/tests/audit-regression.test.mjs +272 -0
  58. package/tests/audit-safety.test.mjs +227 -0
  59. package/tests/backend-compensation.test.mjs +272 -0
  60. package/tests/backup.test.mjs +793 -0
  61. package/tests/binding-diagnostics.test.mjs +720 -0
  62. package/tests/browser.test.mjs +366 -0
  63. package/tests/client-render.test.mjs +892 -0
  64. package/tests/coalesce.test.mjs +113 -0
  65. package/tests/config.test.mjs +159 -0
  66. package/tests/consolidation-credential.test.mjs +331 -0
  67. package/tests/consolidation-feed.test.mjs +96 -0
  68. package/tests/consolidation-fork.test.mjs +489 -0
  69. package/tests/consolidation-ledger.test.mjs +298 -0
  70. package/tests/consolidation-snapshot.test.mjs +360 -0
  71. package/tests/consolidation-value.test.mjs +89 -0
  72. package/tests/context-scope.test.mjs +40 -0
  73. package/tests/cue.test.mjs +44 -0
  74. package/tests/curator-commit.test.mjs +277 -0
  75. package/tests/curator-probe.test.mjs +433 -0
  76. package/tests/curator.test.mjs +150 -0
  77. package/tests/cycle.test.mjs +411 -0
  78. package/tests/disclosure.test.mjs +231 -0
  79. package/tests/doc-drift.test.mjs +222 -0
  80. package/tests/embedding-catalog.test.mjs +266 -0
  81. package/tests/embedding-download.test.mjs +391 -0
  82. package/tests/embedding-runtime.test.mjs +389 -0
  83. package/tests/emotion-chain.test.mjs +84 -0
  84. package/tests/emotion-contract.test.mjs +445 -0
  85. package/tests/engine.test.mjs +35 -0
  86. package/tests/fixtures/active-memory-ab/README.md +56 -0
  87. package/tests/fixtures/active-memory-ab/cases/01-clean-dense-turn.json +351 -0
  88. package/tests/fixtures/active-memory-ab/cases/02-clean-thin-turn.json +122 -0
  89. package/tests/fixtures/active-memory-ab/cases/03-clean-starved-small.json +125 -0
  90. package/tests/fixtures/active-memory-ab/cases/04-clean-large-thin.json +137 -0
  91. package/tests/fixtures/active-memory-ab/cases/05-clean-half-step.json +166 -0
  92. package/tests/fixtures/active-memory-ab/cases/06-clean-duplicate-reread.json +197 -0
  93. package/tests/fixtures/active-memory-ab/cases/07-clean-recent-recall.json +125 -0
  94. package/tests/fixtures/active-memory-ab/cases/08-global-relevant.json +111 -0
  95. package/tests/fixtures/active-memory-ab/cases/09-global-irrelevant-conflict.json +158 -0
  96. package/tests/fixtures/active-memory-ab/cases/10-no-human-turn.json +97 -0
  97. package/tests/fixtures/active-memory-ab/cases/11-clean-scale-100.json +137 -0
  98. package/tests/fixtures/active-memory-ab/cases/12-clean-scale-1000.json +137 -0
  99. package/tests/fixtures/active-memory-ab/cases/13-clean-scale-5000.json +137 -0
  100. package/tests/fixtures/active-memory-ab/cases/14-clean-scale-20000.json +137 -0
  101. package/tests/fixtures/active-memory-ab/lead-ai-reviewed-reference.json +191 -0
  102. package/tests/fixtures/active-memory-ab/manifest.json +199 -0
  103. package/tests/fixtures/recall-bench/01-paraphrase.json +184 -0
  104. package/tests/fixtures/recall-bench/02-exact.json +153 -0
  105. package/tests/fixtures/recall-bench/03-negation.json +173 -0
  106. package/tests/fixtures/recall-bench/04-no-answer.json +162 -0
  107. package/tests/fixtures/recall-bench/05-conflict.json +194 -0
  108. package/tests/fixtures/recall-bench/06-boundary.json +273 -0
  109. package/tests/fixtures/recall-bench/07-secret.json +191 -0
  110. package/tests/fixtures/recall-bench/08-coding.json +164 -0
  111. package/tests/fixtures/recall-bench-gold-candidates/01-paraphrase.json +193 -0
  112. package/tests/fixtures/recall-bench-gold-candidates/02-exact.json +197 -0
  113. package/tests/fixtures/recall-bench-gold-candidates/03-negation.json +244 -0
  114. package/tests/fixtures/recall-bench-gold-candidates/04-no-answer.json +189 -0
  115. package/tests/fixtures/recall-bench-gold-candidates/05-conflict.json +245 -0
  116. package/tests/fixtures/recall-bench-gold-candidates/06-boundary.json +279 -0
  117. package/tests/fixtures/recall-bench-gold-candidates/07-secret.json +224 -0
  118. package/tests/fixtures/recall-bench-gold-candidates/08-coding.json +218 -0
  119. package/tests/fixtures/repetition-eval/01-zh-rp.json +93 -0
  120. package/tests/fixtures/repetition-eval/02-en.json +97 -0
  121. package/tests/fixtures/repetition-eval/03-code.json +91 -0
  122. package/tests/fixtures/repetition-eval/04-names.json +116 -0
  123. package/tests/fixtures/repetition-eval/05-user-quote.json +96 -0
  124. package/tests/fixtures/repetition-eval/06-agreement.json +86 -0
  125. package/tests/graph-cascade.test.mjs +190 -0
  126. package/tests/graph-view.test.mjs +187 -0
  127. package/tests/graph.test.mjs +94 -0
  128. package/tests/hook-order.test.mjs +72 -0
  129. package/tests/humanlike.test.mjs +232 -0
  130. package/tests/hybrid-recall.test.mjs +122 -0
  131. package/tests/inbound.test.mjs +393 -0
  132. package/tests/inject-guide.test.mjs +364 -0
  133. package/tests/inject.test.mjs +222 -0
  134. package/tests/injection-budget-settings.test.mjs +35 -0
  135. package/tests/injection-budget.test.mjs +64 -0
  136. package/tests/injection-chain.test.mjs +191 -0
  137. package/tests/loader-register.mjs +3 -0
  138. package/tests/loader.mjs +12 -0
  139. package/tests/local-embedding-entry.test.mjs +95 -0
  140. package/tests/local-embedding-manager.test.mjs +476 -0
  141. package/tests/local-embedding-protocol.test.mjs +95 -0
  142. package/tests/local-embedding-worker.test.mjs +346 -0
  143. package/tests/managed-memory.test.mjs +591 -0
  144. package/tests/managed-runtime.test.mjs +40 -0
  145. package/tests/media.test.mjs +132 -0
  146. package/tests/migration-payload.test.mjs +42 -0
  147. package/tests/mood.test.mjs +109 -0
  148. package/tests/multiprocess.test.mjs +29 -0
  149. package/tests/new-session-gate.test.mjs +361 -0
  150. package/tests/package-shipping.test.mjs +84 -0
  151. package/tests/preset-scope.test.mjs +212 -0
  152. package/tests/prompts.test.mjs +250 -0
  153. package/tests/react-snapshot.test.mjs +23 -0
  154. package/tests/recall-bench.test.mjs +306 -0
  155. package/tests/recall-dedup.test.mjs +417 -0
  156. package/tests/remote.test.mjs +140 -0
  157. package/tests/rename.test.mjs +30 -0
  158. package/tests/repetition-detector-snapshot.test.mjs +231 -0
  159. package/tests/repetition-eval-live.test.mjs +177 -0
  160. package/tests/repetition-eval.test.mjs +373 -0
  161. package/tests/salient-date.test.mjs +142 -0
  162. package/tests/settings-beat.test.mjs +221 -0
  163. package/tests/short-term.test.mjs +166 -0
  164. package/tests/sidebar-beat.test.mjs +212 -0
  165. package/tests/sidebar-refresh.test.mjs +123 -0
  166. package/tests/sidebar.test.mjs +70 -0
  167. package/tests/snippet.test.mjs +47 -0
  168. package/tests/source-kind.test.mjs +52 -0
  169. package/tests/state.test.mjs +150 -0
  170. package/tests/stats.test.mjs +112 -0
  171. package/tests/storage.test.mjs +125 -0
  172. package/tests/stubs/dsh-llm.js +5 -0
  173. package/tests/stubs/dsh-session.js +7 -0
  174. package/tests/stubs/dsh-tools.js +101 -0
  175. package/tests/tool-schema.test.mjs +73 -0
  176. package/tests/tools.test.mjs +184 -0
  177. package/tests/ui-defaults.test.mjs +57 -0
  178. package/tests/update-fields.test.mjs +138 -0
  179. package/tests/vector-migration.test.mjs +790 -0
  180. package/tests/vector-profile.test.mjs +425 -0
  181. package/tests/visibility-migration.test.mjs +152 -0
  182. package/tests/webui-auth.test.mjs +454 -0
  183. package/tests/webui.test.mjs +165 -0
  184. package/tests/windows-env.test.mjs +89 -0
@@ -0,0 +1,137 @@
1
+ {
2
+ "schemaVersion": 1,
3
+ "kind": "active-memory-ab-case",
4
+ "id": "clean-scale-5000",
5
+ "draft": true,
6
+ "notGold": true,
7
+ "gold": false,
8
+ "review": "pending",
9
+ "annotationSource": "synthetic-proxy",
10
+ "interference": "off",
11
+ "title": "规模梯度 5000 事件:越过 largeTranscriptEvents,薄回合不再吃饥饿豁免",
12
+ "annotationNotes": "合成代理标注。与 clean-large-thin 同证据、同标签、同 expect 形状,唯一变量是 padding.count,用于标定 largeTranscriptEvents 阈值。padding 是确定性工具轮询,文本完全相同,去重后只算 1 个工作单元。",
13
+ "snapshot": {
14
+ "domain": "coding",
15
+ "idleAtSnapshotEndMinutes": 20,
16
+ "lastInjectedAt": null,
17
+ "events": [
18
+ {
19
+ "type": "turn/start",
20
+ "at": "2026-10-30T02:00:00.000Z"
21
+ },
22
+ {
23
+ "type": "user/message",
24
+ "at": "2026-10-30T02:00:00.000Z",
25
+ "data": {
26
+ "source": {
27
+ "kind": "user"
28
+ },
29
+ "content": [
30
+ {
31
+ "type": "text",
32
+ "text": "先按计划把迁移脚本跑通,日志我看产出就行。"
33
+ }
34
+ ]
35
+ }
36
+ },
37
+ {
38
+ "type": "assistant/message",
39
+ "at": "2026-10-30T02:00:00.000Z",
40
+ "data": {
41
+ "message": {
42
+ "content": [
43
+ {
44
+ "type": "text",
45
+ "text": "开始跑。"
46
+ }
47
+ ]
48
+ }
49
+ }
50
+ },
51
+ {
52
+ "type": "step/end",
53
+ "at": "2026-10-30T02:00:00.000Z",
54
+ "step": 1
55
+ },
56
+ {
57
+ "type": "turn/end",
58
+ "at": "2026-10-30T02:00:30.000Z"
59
+ },
60
+ {
61
+ "type": "padding",
62
+ "at": "2026-10-30T02:01:00.000Z",
63
+ "startAt": "2026-10-30T02:01:00.000Z",
64
+ "count": 1247,
65
+ "intervalMs": 1000,
66
+ "tool": "read",
67
+ "text": "polling: no change"
68
+ },
69
+ {
70
+ "type": "turn/start",
71
+ "at": "2026-10-30T02:10:00.000Z"
72
+ },
73
+ {
74
+ "type": "user/message",
75
+ "at": "2026-10-30T02:10:00.000Z",
76
+ "data": {
77
+ "source": {
78
+ "kind": "user"
79
+ },
80
+ "content": [
81
+ {
82
+ "type": "text",
83
+ "text": "继续"
84
+ }
85
+ ]
86
+ }
87
+ },
88
+ {
89
+ "type": "assistant/message",
90
+ "at": "2026-10-30T02:10:00.000Z",
91
+ "data": {
92
+ "message": {
93
+ "content": [
94
+ {
95
+ "type": "text",
96
+ "text": "还在跑。"
97
+ }
98
+ ]
99
+ }
100
+ }
101
+ },
102
+ {
103
+ "type": "step/end",
104
+ "at": "2026-10-30T02:10:00.000Z",
105
+ "step": 1247
106
+ },
107
+ {
108
+ "type": "turn/end",
109
+ "at": "2026-10-30T02:10:00.000Z"
110
+ }
111
+ ]
112
+ },
113
+ "labels": {
114
+ "expected": [
115
+ {
116
+ "id": "mem-ab-scale-plan-scope",
117
+ "class": "decision",
118
+ "sourceSeq": 1,
119
+ "note": "迁移脚本先跑通,日志以产出为准"
120
+ },
121
+ {
122
+ "id": "mem-ab-scale-thin-tail",
123
+ "class": "promise",
124
+ "sourceSeq": 4994,
125
+ "note": "末尾薄回合(“继续”)本身没有新事实,用于量漏记"
126
+ }
127
+ ],
128
+ "mainProactive": [],
129
+ "userReminders": []
130
+ },
131
+ "expect": {
132
+ "maxBackgroundChecks": 0,
133
+ "blockedReasons": [
134
+ "large-transcript-thin"
135
+ ]
136
+ }
137
+ }
@@ -0,0 +1,137 @@
1
+ {
2
+ "schemaVersion": 1,
3
+ "kind": "active-memory-ab-case",
4
+ "id": "clean-scale-20000",
5
+ "draft": true,
6
+ "notGold": true,
7
+ "gold": false,
8
+ "review": "pending",
9
+ "annotationSource": "synthetic-proxy",
10
+ "interference": "off",
11
+ "title": "规模梯度 20000 事件:上限区,薄回合被拦且完整前缀种子成本最高",
12
+ "annotationNotes": "合成代理标注。与 clean-large-thin 同证据、同标签、同 expect 形状,唯一变量是 padding.count,用于标定 largeTranscriptEvents 阈值。padding 是确定性工具轮询,文本完全相同,去重后只算 1 个工作单元。",
13
+ "snapshot": {
14
+ "domain": "coding",
15
+ "idleAtSnapshotEndMinutes": 20,
16
+ "lastInjectedAt": null,
17
+ "events": [
18
+ {
19
+ "type": "turn/start",
20
+ "at": "2026-10-30T02:00:00.000Z"
21
+ },
22
+ {
23
+ "type": "user/message",
24
+ "at": "2026-10-30T02:00:00.000Z",
25
+ "data": {
26
+ "source": {
27
+ "kind": "user"
28
+ },
29
+ "content": [
30
+ {
31
+ "type": "text",
32
+ "text": "先按计划把迁移脚本跑通,日志我看产出就行。"
33
+ }
34
+ ]
35
+ }
36
+ },
37
+ {
38
+ "type": "assistant/message",
39
+ "at": "2026-10-30T02:00:00.000Z",
40
+ "data": {
41
+ "message": {
42
+ "content": [
43
+ {
44
+ "type": "text",
45
+ "text": "开始跑。"
46
+ }
47
+ ]
48
+ }
49
+ }
50
+ },
51
+ {
52
+ "type": "step/end",
53
+ "at": "2026-10-30T02:00:00.000Z",
54
+ "step": 1
55
+ },
56
+ {
57
+ "type": "turn/end",
58
+ "at": "2026-10-30T02:00:30.000Z"
59
+ },
60
+ {
61
+ "type": "padding",
62
+ "at": "2026-10-30T02:01:00.000Z",
63
+ "startAt": "2026-10-30T02:01:00.000Z",
64
+ "count": 4997,
65
+ "intervalMs": 1000,
66
+ "tool": "read",
67
+ "text": "polling: no change"
68
+ },
69
+ {
70
+ "type": "turn/start",
71
+ "at": "2026-10-30T02:10:00.000Z"
72
+ },
73
+ {
74
+ "type": "user/message",
75
+ "at": "2026-10-30T02:10:00.000Z",
76
+ "data": {
77
+ "source": {
78
+ "kind": "user"
79
+ },
80
+ "content": [
81
+ {
82
+ "type": "text",
83
+ "text": "继续"
84
+ }
85
+ ]
86
+ }
87
+ },
88
+ {
89
+ "type": "assistant/message",
90
+ "at": "2026-10-30T02:10:00.000Z",
91
+ "data": {
92
+ "message": {
93
+ "content": [
94
+ {
95
+ "type": "text",
96
+ "text": "还在跑。"
97
+ }
98
+ ]
99
+ }
100
+ }
101
+ },
102
+ {
103
+ "type": "step/end",
104
+ "at": "2026-10-30T02:10:00.000Z",
105
+ "step": 4997
106
+ },
107
+ {
108
+ "type": "turn/end",
109
+ "at": "2026-10-30T02:10:00.000Z"
110
+ }
111
+ ]
112
+ },
113
+ "labels": {
114
+ "expected": [
115
+ {
116
+ "id": "mem-ab-scale-plan-scope",
117
+ "class": "decision",
118
+ "sourceSeq": 1,
119
+ "note": "迁移脚本先跑通,日志以产出为准"
120
+ },
121
+ {
122
+ "id": "mem-ab-scale-thin-tail",
123
+ "class": "promise",
124
+ "sourceSeq": 19994,
125
+ "note": "末尾薄回合(“继续”)本身没有新事实,用于量漏记"
126
+ }
127
+ ],
128
+ "mainProactive": [],
129
+ "userReminders": []
130
+ },
131
+ "expect": {
132
+ "maxBackgroundChecks": 0,
133
+ "blockedReasons": [
134
+ "large-transcript-thin"
135
+ ]
136
+ }
137
+ }
@@ -0,0 +1,191 @@
1
+ {
2
+ "schemaVersion": 1,
3
+ "kind": "active-memory-ab-lead-reviewed-reference",
4
+ "status": "lead-ai-reviewed",
5
+ "gold": false,
6
+ "annotationSource": "lead-ai-reviewed",
7
+ "reviewer": "lead-ai",
8
+ "reviewedAt": "2026-10-30",
9
+ "sourceDataset": {
10
+ "status": "synthetic-proxy-not-gold",
11
+ "gold": false,
12
+ "source": "local fixture cases/ (unchanged; no synthetic label was promoted)"
13
+ },
14
+ "policy": {
15
+ "seedMode": "complete-contiguous-prefix",
16
+ "windowedSeed": false,
17
+ "heldCountsAsCovered": false,
18
+ "halfStep": "deferred without model scoring",
19
+ "noHumanTurn": "skipped",
20
+ "thinNoFact": "no_value only when normal completed check returns no candidates"
21
+ },
22
+ "cases": [
23
+ {
24
+ "id": "clean-dense-turn",
25
+ "expectedStatus": "completed",
26
+ "labels": [
27
+ {
28
+ "class": "decision",
29
+ "sourceSeq": 1,
30
+ "canonical": "snapshot-worker-does-not-block-main-session"
31
+ },
32
+ {
33
+ "class": "root-cause",
34
+ "sourceSeq": 1,
35
+ "canonical": "repeated-full-transcript-forking-drives-input-cost"
36
+ },
37
+ {
38
+ "class": "preference",
39
+ "sourceSeq": 5,
40
+ "canonical": "python-installation-is-confined-to-engine-venv"
41
+ },
42
+ {
43
+ "class": "promise",
44
+ "sourceSeq": 9,
45
+ "canonical": "synthetic-labels-never-become-gold-without-review"
46
+ }
47
+ ],
48
+ "notes": [
49
+ "The worker/root-cause decision is at seq 1; the venv constraint is at seq 5; the gold-gating promise is at seq 9."
50
+ ]
51
+ },
52
+ {
53
+ "id": "clean-thin-turn",
54
+ "expectedStatus": "completed",
55
+ "labels": [],
56
+ "outcome": "no_value",
57
+ "notes": [
58
+ "The two human turns contain no durable decision or preference; the fixture label marks an uncovered interval, not a memory."
59
+ ]
60
+ },
61
+ {
62
+ "id": "clean-starved-small",
63
+ "expectedStatus": "completed",
64
+ "labels": [
65
+ {
66
+ "class": "decision",
67
+ "sourceSeq": 1,
68
+ "canonical": "no_value-requires-normal-completed-empty-result"
69
+ }
70
+ ],
71
+ "notes": [
72
+ "The no_value rule is an explicit human decision at seq 1; idle/starvation does not itself create a memory."
73
+ ]
74
+ },
75
+ {
76
+ "id": "clean-large-thin",
77
+ "expectedStatus": "completed",
78
+ "labels": [
79
+ {
80
+ "class": "decision",
81
+ "sourceSeq": 1,
82
+ "canonical": "run-migration-plan-and-use-produced-logs"
83
+ }
84
+ ],
85
+ "notes": [
86
+ "The later “继续” at seq 7 adds no fact; the synthetic tail label is not a gold memory."
87
+ ]
88
+ },
89
+ {
90
+ "id": "clean-half-step",
91
+ "expectedStatus": "deferred",
92
+ "guardReason": "half-step",
93
+ "labels": [],
94
+ "notes": [
95
+ "The prefix contains an unpaired tool call and snapshot.pendingCheck=true; hard guard prevents scoring, writes, and cursor advance."
96
+ ]
97
+ },
98
+ {
99
+ "id": "clean-duplicate-reread",
100
+ "expectedStatus": "completed",
101
+ "labels": [
102
+ {
103
+ "class": "decision",
104
+ "sourceSeq": 1,
105
+ "canonical": "do-not-use-windowed-seed"
106
+ },
107
+ {
108
+ "class": "decision",
109
+ "sourceSeq": 8,
110
+ "canonical": "receipts-bind-snapshot-digest-and-scope-generation"
111
+ }
112
+ ],
113
+ "notes": [
114
+ "The credential requirement is stated by the human at seq 8; seq 7 is turn/start and cannot be the source."
115
+ ]
116
+ },
117
+ {
118
+ "id": "clean-recent-recall",
119
+ "expectedStatus": "completed",
120
+ "labels": [
121
+ {
122
+ "class": "decision",
123
+ "sourceSeq": 1,
124
+ "canonical": "suppress-checks-within-recall-cooldown"
125
+ }
126
+ ],
127
+ "notes": [
128
+ "The suppression rule is an explicit human instruction at seq 1."
129
+ ]
130
+ },
131
+ {
132
+ "id": "global-relevant",
133
+ "expectedStatus": "completed",
134
+ "labels": [
135
+ {
136
+ "class": "preference",
137
+ "sourceSeq": 1,
138
+ "canonical": "worker-writes-private-only-and-global-is-read-only"
139
+ }
140
+ ],
141
+ "notes": [
142
+ "Relevant global entries are read-only context and must not be copied into private writes."
143
+ ]
144
+ },
145
+ {
146
+ "id": "global-irrelevant-conflict",
147
+ "expectedStatus": "completed",
148
+ "labels": [
149
+ {
150
+ "class": "decision",
151
+ "sourceSeq": 1,
152
+ "canonical": "retain-only-one-backup-after-successful-switch",
153
+ "disposition": "held-conflict"
154
+ },
155
+ {
156
+ "class": "preference",
157
+ "sourceSeq": 6,
158
+ "canonical": "web-gui-port-is-9032",
159
+ "disposition": "covered"
160
+ }
161
+ ],
162
+ "notes": [
163
+ "The backup decision conflicts with the global three-backup entry, so it is held and does not count as covered; the port preference is independent."
164
+ ]
165
+ },
166
+ {
167
+ "id": "no-human-turn",
168
+ "expectedStatus": "skipped",
169
+ "guardReason": "no-human-turn",
170
+ "labels": [],
171
+ "notes": [
172
+ "The user-shaped message is plugin-generated (source.kind=plugin:dsh-memu-memory), so this case is not eligible for human-memory scoring."
173
+ ]
174
+ }
175
+ ],
176
+ "metrics": {
177
+ "cases": 10,
178
+ "completed": 7,
179
+ "deferred": 1,
180
+ "skipped": 1,
181
+ "normalNoValue": 1,
182
+ "labels": 13,
183
+ "heldLabels": 1,
184
+ "coveredLabels": 12
185
+ },
186
+ "caveats": [
187
+ "This lead-AI-reviewed reference is an independent review of local fixture evidence, not a gold set and not a claim that the live model produced these labels.",
188
+ "The 13 labels are non-exhaustive reference annotations; the live run produced no parseable candidates, so model effect and A/B accuracy remain unmeasured.",
189
+ "Raw prompts, responses, transcripts, and credentials are not stored."
190
+ ]
191
+ }
@@ -0,0 +1,199 @@
1
+ {
2
+ "schemaVersion": 1,
3
+ "kind": "active-memory-ab-manifest",
4
+ "name": "active-memory 金标 A/B 候选集(合成代理草稿,未人审)",
5
+ "status": "synthetic-proxy-not-gold",
6
+ "draft": true,
7
+ "notGold": true,
8
+ "gold": false,
9
+ "review": "pending",
10
+ "annotationPolicy": "所有标签都是合成代理(annotationSource=synthetic-proxy),按 seq 指向“若由人来标也应记”的事件;没有人工逐条确认,因此不是 gold set。人审候选与 gold gating 分开:只有 review 改为 human-reviewed 之后才可能进入金标。",
11
+ "axes": [
12
+ "off",
13
+ "relevant",
14
+ "irrelevant-conflict"
15
+ ],
16
+ "conditions": [
17
+ "legacy",
18
+ "guide",
19
+ "guide-weighted"
20
+ ],
21
+ "rows": [
22
+ "legacy",
23
+ "guide",
24
+ "guide-weighted:oracle",
25
+ "guide-weighted:no-value"
26
+ ],
27
+ "paddingGenerator": "ab-polling-padding-v1",
28
+ "realModelCalls": 0,
29
+ "cases": [
30
+ {
31
+ "id": "clean-dense-turn",
32
+ "file": "cases/01-clean-dense-turn.json",
33
+ "throughSeq": 22,
34
+ "eventCount": 23,
35
+ "snapshotSha256": "01b63a470444c849c93239c9b717f3b6049497398c05daaab193d6781093838c",
36
+ "interference": "off",
37
+ "annotationSource": "synthetic-proxy",
38
+ "review": "pending",
39
+ "gold": false,
40
+ "title": "密集回合:分数达标就检查,不等空闲"
41
+ },
42
+ {
43
+ "id": "clean-thin-turn",
44
+ "file": "cases/02-clean-thin-turn.json",
45
+ "throughSeq": 9,
46
+ "eventCount": 10,
47
+ "snapshotSha256": "73cfdd46eb46233ac5129a1a73e5a2a77be34a4bdf3e6d27dc5e182cc39495b7",
48
+ "interference": "off",
49
+ "annotationSource": "synthetic-proxy",
50
+ "review": "pending",
51
+ "gold": false,
52
+ "title": "薄回合:没有空闲窗口,也不该唤醒"
53
+ },
54
+ {
55
+ "id": "clean-starved-small",
56
+ "file": "cases/03-clean-starved-small.json",
57
+ "throughSeq": 9,
58
+ "eventCount": 10,
59
+ "snapshotSha256": "49145876868c032e291dd7964d63ceea6afd8481158bcaf81617561aded8cd82",
60
+ "interference": "off",
61
+ "annotationSource": "synthetic-proxy",
62
+ "review": "pending",
63
+ "gold": false,
64
+ "title": "小会话饥饿:空闲到阈值就无条件检查一次"
65
+ },
66
+ {
67
+ "id": "clean-large-thin",
68
+ "file": "cases/04-clean-large-thin.json",
69
+ "throughSeq": 2089,
70
+ "eventCount": 2090,
71
+ "snapshotSha256": "e828cde9d7ee4242a507c9af294dd0acd0a38df0fd087d42f65351dabb9c5cf1",
72
+ "interference": "off",
73
+ "annotationSource": "synthetic-proxy",
74
+ "review": "pending",
75
+ "gold": false,
76
+ "title": "大 transcript 薄回合:不吃饥饿豁免"
77
+ },
78
+ {
79
+ "id": "clean-half-step",
80
+ "file": "cases/05-clean-half-step.json",
81
+ "throughSeq": 3,
82
+ "eventCount": 4,
83
+ "snapshotSha256": "c6eb87ef676716ddad2e4f5fa32d985bd1c151f70516f72e91e38bd698358b96",
84
+ "interference": "off",
85
+ "annotationSource": "synthetic-proxy",
86
+ "review": "pending",
87
+ "gold": false,
88
+ "title": "半步边界:tool/call 没有结果就 deferred,不推进游标"
89
+ },
90
+ {
91
+ "id": "clean-duplicate-reread",
92
+ "file": "cases/06-clean-duplicate-reread.json",
93
+ "throughSeq": 13,
94
+ "eventCount": 14,
95
+ "snapshotSha256": "d48163e0314151f7ed6e57e267e3f2c4bdf2c7f9879b35e9f5d14d0deab401d0",
96
+ "interference": "off",
97
+ "annotationSource": "synthetic-proxy",
98
+ "review": "pending",
99
+ "gold": false,
100
+ "title": "重复读与已写条目:整段前缀重放会再次提议同一条"
101
+ },
102
+ {
103
+ "id": "clean-recent-recall",
104
+ "file": "cases/07-clean-recent-recall.json",
105
+ "throughSeq": 9,
106
+ "eventCount": 10,
107
+ "snapshotSha256": "2ea689efcda46021415affa143fe55e167b11ed6784dabadce18c27e1cf26e50",
108
+ "interference": "off",
109
+ "annotationSource": "synthetic-proxy",
110
+ "review": "pending",
111
+ "gold": false,
112
+ "title": "刚注入过记忆:抑制而不是检查"
113
+ },
114
+ {
115
+ "id": "global-relevant",
116
+ "file": "cases/08-global-relevant.json",
117
+ "throughSeq": 4,
118
+ "eventCount": 5,
119
+ "snapshotSha256": "0f4d97adf48488f69f3228a1fc7a746472fed18221f3502186e5e086f9934fb9",
120
+ "interference": "relevant",
121
+ "annotationSource": "synthetic-proxy",
122
+ "review": "pending",
123
+ "gold": false,
124
+ "title": "全局干扰(相关):相关全局条目可以进上下文,但不许被复制进私有写"
125
+ },
126
+ {
127
+ "id": "global-irrelevant-conflict",
128
+ "file": "cases/09-global-irrelevant-conflict.json",
129
+ "throughSeq": 9,
130
+ "eventCount": 10,
131
+ "snapshotSha256": "0a1d5e3a73965d5efb523e967995036d5693be9afaf22bef9c03721bd6a73963",
132
+ "interference": "irrelevant-conflict",
133
+ "annotationSource": "synthetic-proxy",
134
+ "review": "pending",
135
+ "gold": false,
136
+ "title": "全局干扰(无关且冲突):冲突条目只进 held,不进可召回覆盖"
137
+ },
138
+ {
139
+ "id": "no-human-turn",
140
+ "file": "cases/10-no-human-turn.json",
141
+ "throughSeq": 5,
142
+ "eventCount": 6,
143
+ "snapshotSha256": "851fa7965cbe2bb477c2e1c062cf5c13ba198c911cd1f85cb0885cee3d8e0899",
144
+ "interference": "off",
145
+ "annotationSource": "synthetic-proxy",
146
+ "review": "pending",
147
+ "gold": false,
148
+ "title": "没有真人回合:纯工具轮询不构成检查理由"
149
+ },
150
+ {
151
+ "id": "clean-scale-100",
152
+ "file": "cases/11-clean-scale-100.json",
153
+ "throughSeq": 97,
154
+ "eventCount": 98,
155
+ "snapshotSha256": "ec6180fcd444ab6f71345ff571da9ca79c8168f0a036d2e8902b6673e4ee34ce",
156
+ "interference": "off",
157
+ "annotationSource": "synthetic-proxy",
158
+ "review": "pending",
159
+ "gold": false,
160
+ "title": "规模梯度 100 事件:低于 largeTranscriptEvents,饥饿豁免仍然生效"
161
+ },
162
+ {
163
+ "id": "clean-scale-1000",
164
+ "file": "cases/12-clean-scale-1000.json",
165
+ "throughSeq": 997,
166
+ "eventCount": 998,
167
+ "snapshotSha256": "04510b40d08eb03e3b1d04715cff1554324b5f1769bb97c95f78b22e82496f1a",
168
+ "interference": "off",
169
+ "annotationSource": "synthetic-proxy",
170
+ "review": "pending",
171
+ "gold": false,
172
+ "title": "规模梯度 1000 事件:仍在阈值下,饥饿豁免生效"
173
+ },
174
+ {
175
+ "id": "clean-scale-5000",
176
+ "file": "cases/13-clean-scale-5000.json",
177
+ "throughSeq": 4997,
178
+ "eventCount": 4998,
179
+ "snapshotSha256": "dc9095796c6cc122d7b314f59db2efa1963d86226837b7b5234c57ebb671e68f",
180
+ "interference": "off",
181
+ "annotationSource": "synthetic-proxy",
182
+ "review": "pending",
183
+ "gold": false,
184
+ "title": "规模梯度 5000 事件:越过 largeTranscriptEvents,薄回合不再吃饥饿豁免"
185
+ },
186
+ {
187
+ "id": "clean-scale-20000",
188
+ "file": "cases/14-clean-scale-20000.json",
189
+ "throughSeq": 19997,
190
+ "eventCount": 19998,
191
+ "snapshotSha256": "400e916841df1318ea6819bd73b840039b91157b955c9968d08e660c32fc85e4",
192
+ "interference": "off",
193
+ "annotationSource": "synthetic-proxy",
194
+ "review": "pending",
195
+ "gold": false,
196
+ "title": "规模梯度 20000 事件:上限区,薄回合被拦且完整前缀种子成本最高"
197
+ }
198
+ ]
199
+ }