@furongjun1999/dsh-memory 0.9.22 → 0.10.1

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 (179) hide show
  1. package/CHANGELOG.en.md +450 -0
  2. package/CHANGELOG.md +112 -0
  3. package/README.en.md +244 -0
  4. package/README.md +272 -260
  5. package/codebuddy/CODEBUDDY.md +1 -1
  6. package/docs/README.md +2 -0
  7. package/docs/discipline/templates/full.md.tmpl +1 -1
  8. package/docs/discipline/templates/rules.mdc.tmpl +1 -1
  9. package/docs/en/README.md +190 -0
  10. package/docs/en/TRANSLATION.md +78 -0
  11. package/docs/en/TRANSLATION_MANIFEST.json +170 -0
  12. package/docs/en/TRANSLATION_REPORT.md +11 -0
  13. package/docs/en/dev/README.md +10 -0
  14. package/docs/en/dev/coding-testing.md +72 -0
  15. package/docs/en/dev/switch-matrix.md +10 -0
  16. package/docs/en/glossary.md +97 -0
  17. package/docs/en/harnesses/claude.md +100 -0
  18. package/docs/en/harnesses/codebuddy.md +93 -0
  19. package/docs/en/harnesses/codex.md +88 -0
  20. package/docs/en/harnesses/dsh.md +84 -0
  21. package/docs/en/harnesses/zcode.md +42 -0
  22. package/docs/en/mdcg/ccg.md +32 -0
  23. package/docs/en/mdcg/generated-docs.md +20 -0
  24. package/docs/en/mdcg/guardrail-charter.md +170 -0
  25. package/docs/en/mdcg/mcp-tool-names.md +43 -0
  26. package/docs/en/mdcg/multi-harness.md +43 -0
  27. package/docs/en/mdcg/recall-metadata.md +30 -0
  28. package/docs/en/mdcg/world-model-endpoint.md +379 -0
  29. package/docs/en/mdcg/write-aliases.md +130 -0
  30. package/docs/en/ops/README.md +11 -0
  31. package/docs/en/ops/environment-known-issues.md +244 -0
  32. package/docs/en/ops/install.md +72 -0
  33. package/docs/en/ops/operations-troubleshooting.md +86 -0
  34. package/docs/en/ops/permissions-review.md +79 -0
  35. package/docs/en/ops/release-governance.md +49 -0
  36. package/docs/en/ops/tool-name-migration.md +30 -0
  37. package/docs/en/ops/ui-install-troubleshooting.md +53 -0
  38. package/docs/en/standards/ai-coding-standard-checklist.md +71 -0
  39. package/docs/en/standards/ai-coding-standard-lingshu.md +47 -0
  40. package/docs/en/standards/ai-coding-standard.md +185 -0
  41. package/docs/en/ui/minimal.md +62 -0
  42. package/docs/mdcg/cogmap_ops.md +16 -15
  43. package/lib/client.js +29 -3
  44. package/lib/hooks.d.ts +34 -0
  45. package/lib/hooks.js +167 -33
  46. package/lib/index.js +19 -3
  47. package/lib/lib/console_web.mjs +30 -3
  48. package/lib/lib/hook_audit.d.ts +3 -2
  49. package/lib/lib/hook_audit.js +14 -4
  50. package/lib/lib/mdcg_client.d.ts +5 -2
  51. package/lib/lib/mdcg_client.js +8 -4
  52. package/lib/lib/prompt_safety.d.ts +24 -0
  53. package/lib/lib/prompt_safety.js +77 -1
  54. package/lib/lib/token_store.js +5 -2
  55. package/lib/tools.js +5 -3
  56. package/md_cg/bucketindex.py +231 -0
  57. package/md_cg/bucketscope.py +339 -0
  58. package/md_cg/config_registry_bulk.py +19 -10
  59. package/md_cg/config_validate.py +14 -2
  60. package/md_cg/detrerank.py +137 -0
  61. package/md_cg/fsutil.py +215 -25
  62. package/md_cg/govern/__init__.py +23 -0
  63. package/md_cg/iface/__init__.py +46 -0
  64. package/md_cg/iface/_verify.py +609 -0
  65. package/md_cg/iface/contracts.py +337 -0
  66. package/md_cg/judgment_manifest.py +14 -0
  67. package/md_cg/kernel/__init__.py +27 -0
  68. package/md_cg/mcp_server.py +457 -23
  69. package/md_cg/mdcg.py +1034 -79
  70. package/md_cg/mdcos.py +263 -21
  71. package/md_cg/metacognition.py +66 -13
  72. package/md_cg/mutate_extract_shipped.py +126 -0
  73. package/md_cg/mutate_ledger_e2e.py +100 -0
  74. package/md_cg/nodefile.py +147 -23
  75. package/md_cg/op_table.py +22 -4
  76. package/md_cg/probe_c_ledger_integrity.py +258 -0
  77. package/md_cg/probe_d_retirement_discipline.py +263 -0
  78. package/md_cg/reconcile.py +9 -11
  79. package/md_cg/review_cli.py +6 -1
  80. package/md_cg/rollback.py +16 -1
  81. package/md_cg/run_tests.py +30 -1
  82. package/md_cg/scopeselect.py +214 -0
  83. package/md_cg/scrub.py +160 -23
  84. package/md_cg/security.py +4 -2
  85. package/md_cg/state_extract.py +149 -0
  86. package/md_cg/state_slots.py +58 -10
  87. package/md_cg/structboost.py +179 -0
  88. package/md_cg/test_alternatives_historical.py +162 -0
  89. package/md_cg/test_b2_sensitivity_landing.py +226 -223
  90. package/md_cg/test_b3_merge_keeps_content.py +608 -583
  91. package/md_cg/test_bucket_cards.py +85 -0
  92. package/md_cg/test_bucket_day_order.py +74 -0
  93. package/md_cg/test_bucket_quota.py +79 -0
  94. package/md_cg/test_bucket_recency_guards.py +75 -0
  95. package/md_cg/test_bucket_three_causes.py +83 -0
  96. package/md_cg/test_bucketindex.py +130 -0
  97. package/md_cg/test_bucketscope_early.py +71 -0
  98. package/md_cg/test_cond_term_fold.py +183 -0
  99. package/md_cg/test_cond_term_fold_gate.py +1210 -0
  100. package/md_cg/test_cond_term_neg_scope.py +147 -0
  101. package/md_cg/test_cond_terms_colon.py +348 -0
  102. package/md_cg/test_detrerank.py +157 -0
  103. package/md_cg/test_gate_four_kpi.py +136 -0
  104. package/md_cg/test_generation_guard.py +76 -43
  105. package/md_cg/test_issue192_move_layer_indices_guard.py +48 -0
  106. package/md_cg/test_issue362_neg_condition.py +44 -7
  107. package/md_cg/test_issue39_utf8_stdio.py +572 -572
  108. package/md_cg/test_issue413_preimage_failclosed.py +147 -9
  109. package/md_cg/test_issue444_scrub_evidence_plaintext.py +14 -8
  110. package/md_cg/test_issue677_flush_batch_io.py +139 -0
  111. package/md_cg/test_issue677_info_fast.py +82 -0
  112. package/md_cg/test_issue677_remember_dedupe.py +106 -0
  113. package/md_cg/test_issue712_tokens_sessions_read.py +199 -0
  114. package/md_cg/test_issue722_strict_conditions.py +96 -0
  115. package/md_cg/test_issue730_narrowing_audit.py +121 -0
  116. package/md_cg/test_issue733_verify_override.py +193 -0
  117. package/md_cg/test_issue733_verify_override_and_overwrite_inherit.py +315 -0
  118. package/md_cg/test_issue740_filelock_fd_leak.py +218 -0
  119. package/md_cg/test_issue741_content_hash_same_source.py +158 -0
  120. package/md_cg/test_issue747_slot_contradiction.py +125 -0
  121. package/md_cg/test_issue754_nested_root_guard.py +197 -0
  122. package/md_cg/test_issue757_index_hash_compare.py +85 -0
  123. package/md_cg/test_issue758_disk_hash_single.py +67 -0
  124. package/md_cg/test_issue759_twophase_hash.py +67 -0
  125. package/md_cg/test_issue768_overwrite_time_window_anchor.py +135 -0
  126. package/md_cg/test_issue789_log_plaintext_probe.py +100 -0
  127. package/md_cg/test_issue790_verb_negation.py +77 -0
  128. package/md_cg/test_issue810_root_guard.py +71 -0
  129. package/md_cg/test_issue812_info_index_stale.py +78 -0
  130. package/md_cg/test_issue827_weaken_dedup.py +361 -0
  131. package/md_cg/test_issue829_calibration_positive_gate.py +473 -0
  132. package/md_cg/test_kernel_iface_contracts.py +18 -0
  133. package/md_cg/test_md_cg_no_hive_dep.py +238 -17
  134. package/md_cg/test_n204_n205_n226_n227_n228_n229_exit_gates.py +545 -542
  135. package/md_cg/test_n209_verify_write_face_gates.py +5 -2
  136. package/md_cg/test_neg_condition_hits.py +336 -333
  137. package/md_cg/test_probe_cd_shipped.py +385 -0
  138. package/md_cg/test_recent_tail.py +93 -0
  139. package/md_cg/test_run_tests_entry_gates.py +110 -0
  140. package/md_cg/test_scopeselect.py +70 -0
  141. package/md_cg/test_scopeselect_clock.py +70 -0
  142. package/md_cg/test_scrub_source_line_false_positive.py +64 -0
  143. package/md_cg/test_session_attribution_consistency.py +835 -813
  144. package/md_cg/test_state_extract_shipped.py +162 -0
  145. package/md_cg/test_state_ledger_chain.py +207 -0
  146. package/md_cg/test_state_ledger_e2e.py +145 -0
  147. package/md_cg/test_structboost.py +102 -0
  148. package/md_cg/test_timeboost.py +83 -0
  149. package/md_cg/test_timeboost_relativedate.py +71 -0
  150. package/md_cg/test_topicagg.py +129 -0
  151. package/md_cg/test_twophase.py +286 -286
  152. package/md_cg/timeboost.py +223 -0
  153. package/md_cg/tokens.py +55 -7
  154. package/md_cg/topicagg.py +142 -0
  155. package/md_cg/twophase.py +15 -6
  156. package/md_cg/whitebox_kb/wisdom/code_compose.py +33 -5
  157. package/md_cg/whitebox_kb/wisdom/compose_engine.py +5 -1
  158. package/md_cg/whitebox_kb/wisdom/role_compose.py +271 -268
  159. package/md_cg/writepipe.py +35 -3
  160. package/package.json +4 -1
  161. package/scripts/workspace_index.py +2 -0
  162. package/src/hooks.ts +182 -34
  163. package/src/index.ts +19 -3
  164. package/src/lib/console_web.mjs +30 -3
  165. package/src/lib/hook_audit.ts +15 -5
  166. package/src/lib/mdcg_client.ts +8 -4
  167. package/src/lib/prompt_safety.ts +73 -1
  168. package/src/lib/token_store.ts +5 -2
  169. package/src/tools.ts +5 -3
  170. package/ui/minimal/integrity.json +2 -2
  171. package/ui/minimal/payload/build-ui.mjs +41 -2
  172. package/ui/minimal/payload/lib/client.js +29 -3
  173. package/ui/minimal/payload/lib/lib/console_web.mjs +30 -3
  174. package/ui/minimal/payload/src/lib/console_web.mjs +30 -3
  175. package/ui/minimal/payload/web/app.js +3 -1
  176. package/ui/minimal/payload/web/client.js +30 -3
  177. package/web/app.js +3 -1
  178. package/web/client.js +30 -3
  179. package/zcode/AGENTS.md +1 -1
@@ -0,0 +1,339 @@
1
+ # -*- coding: utf-8 -*-
2
+ """桶作用域召回(bucketscope)——日桶 / 会话桶的**召回侧**落地。
3
+
4
+ 两类桶的定死口径(实测主库 34,218 节点,2026-11-09):
5
+ · **日桶** = `created_at` 的本地日期,`day:YYYY-MM-DD`。83 个桶;查询 5ms(索引)vs 2–17s(全扫)。
6
+ · **会话桶** = `frontmatter.session`,`session:<sid>`。631 个桶;**72% 是 `unattributed`**——
7
+ 这类节点的会话身份不可知,**只按日桶走**,绝不与他人合并(合并会把无关内容塞进同一上下文)。
8
+
9
+ 两条边界(都是实测教训):
10
+ 1. **日桶不能当「那天只写了这些」**:主库 2026-11-08 有 14,217 条(多日批量写入挤到同一天),
11
+ 所以日桶召回必须**再按查询词过滤**,不能整桶直接塞给模型。
12
+ 2. 会话桶同理:最大真实会话 3,026 条,整桶塞不进去。
13
+
14
+ 故本模块只做**受控召回**:桶内按查询词面筛选 → 取前 cap 条 → 以 0 分追加(不抢名次)。
15
+ """
16
+ from __future__ import annotations
17
+
18
+ import os
19
+ import re
20
+
21
+ from . import bucketindex as _bi
22
+
23
+ #: 单次桶召回上限(与 recent_tail 同一量级,避免撑爆预算)。
24
+ CAP = 8
25
+
26
+
27
+ def _terms(q):
28
+ """查询切词:先按标点/空白切段,再对**中文长段补 2 元**。
29
+
30
+ 为什么要补二元:中文查询通常没有空格(「江北回声那回到了哪里」是**一个**段),
31
+ 整段做子串匹配几乎不可能命中任何节点 ⇒ 桶内 `_pick` 恒返回空 ⇒ 上层回落
32
+ 「最近 14 天」。补 2 元后「江北」「回声」都能匹配;排序仍按命中数降序,
33
+ 长段本身也算一个词,故整段命中者权重更高。
34
+ """
35
+ out, seen = [], set()
36
+ for seg in re.split(r"[\s,,、;;。!?::()()\[\]【】]+", q or ""):
37
+ seg = seg.lower()
38
+ if len(seg) < 2:
39
+ continue
40
+ if seg not in seen:
41
+ seen.add(seg)
42
+ out.append(seg)
43
+ if len(seg) >= 4 and re.search(r"[\u4e00-\u9fff]", seg):
44
+ for i in range(len(seg) - 1):
45
+ bg = seg[i:i + 2]
46
+ if bg not in seen:
47
+ seen.add(bg)
48
+ out.append(bg)
49
+ return out
50
+
51
+
52
+ #: 桶内扫描预算:桶可能上万条(主库 2026-11-08 有 14,217 条),逐条读盘会拖到秒级;
53
+ #: 实测 14k 桶全扫 2.9s、预算 800 条约 0.2s。故按**id 序**取前 MAX_SCAN 条(确定性、可复算),
54
+ #: 代价是桶内覆盖不全——这是「快」与「全」的显式取舍,不是静默降级。
55
+ MAX_SCAN = 800
56
+
57
+
58
+ def _pick(store, ids, terms, cap, validity=False, now=None, fallback_recency=False,
59
+ fill_rest=False):
60
+ """桶内按词面命中筛选(确定性):命中词数降序 → **时间降序** → id;取前 cap。
61
+
62
+ 【#842】排序键必须**在截断之前**含时间:早先只按 (-hits, nid) 排再截 cap,
63
+ 9 条同等命中 + cap=8 时被 id 序丢掉的那条可能正是最新的(后置的时间排序
64
+ 已经拿不回来)。现在把 created_at 提到截断前,截断即保留「同分里最新的」。
65
+
66
+ 【#843】validity=True 时必须在此处就过滤过期节点:桶补召回是**在候选检索
67
+ 之后**追加的,若不在源头过滤,调用方显式要求的「排除过期」会被补召回推翻。
68
+ """
69
+ scored = []
70
+ for nid in list(ids)[:MAX_SCAN]:
71
+ node = store.get(nid)
72
+ if not node:
73
+ continue
74
+ if validity and _is_expired(node, now):
75
+ continue
76
+ text = str(node.get("content") or "").lower()
77
+ hits = sum(1 for t in terms if t in text)
78
+ if hits:
79
+ scored.append((-hits, _recency_key(node), nid, node))
80
+ scored.sort()
81
+ out = [(node, 0.0, {"reason": "bucket_scope"}, [])
82
+ for _h, _r, _nid, node in scored[:cap]]
83
+ # 【外测 2026-11-11 卡点 1】命中抑制:只要有一条命中,早先就把**同桶其余节点全丢**,
84
+ # 于是「锚点在桶里但没命中 query 词」的条目直接消失(P16/d03:桶 [114,115,116],
85
+ # 只命中 114 ⇒ 116 被丢)。现改为:命中优先,**按时间序补足到 cap**。
86
+ if fill_rest and len(out) < cap:
87
+ chosen = {n.get("id") for n, _s, _q, _p in out}
88
+ rest = [n for n in (store.get(i) for i in list(ids)[:MAX_SCAN])
89
+ if n and n.get("id") not in chosen]
90
+ rest.sort(key=lambda n: (_recency_key(n), str(n.get("id"))))
91
+ for n in rest[:cap - len(out)]:
92
+ out.append((n, 0.0, {"reason": "bucket_scope",
93
+ "fallback": ("recency" if not scored else "fill")}, []))
94
+ return out
95
+
96
+
97
+ def day_scope(store, query, day, cap=CAP, validity=False, now=None):
98
+ """日桶召回:day 桶内按查询词筛选。索引缺失返回 None(调用方据此区分「无桶」与「空桶」)。"""
99
+ ids = _bi.day_ids(store, day)
100
+ if not ids:
101
+ return None
102
+ return _pick(store, ids, _terms(query), cap, validity=validity, now=now)
103
+
104
+
105
+ def session_scope(store, query, session, cap=CAP, validity=False, now=None):
106
+ """会话桶召回;`unattributed` 明确拒绝(会话身份不可知,不冒充)。"""
107
+ sid = str(session or "")
108
+ if not sid or sid == _bi.UNATTRIBUTED:
109
+ return None
110
+ ids = _bi.session_ids(store, sid)
111
+ if not ids:
112
+ return None
113
+ return _pick(store, ids, _terms(query), cap, validity=validity, now=now)
114
+
115
+
116
+ #: 通用默认:桶作用域**默认开启**(日期 + 会话是记忆的基本坐标,不是可选优化)。
117
+ DEFAULT_ON = True
118
+ #: 桶召回默认条数(会话桶 + 日桶合计)。
119
+ DEFAULT_CAP = 8
120
+
121
+ #: **会话桶独占条数**(2026-11-10 外测 v2.1):早先会话桶与日桶共享 cap=8,
122
+ #: 会话桶先取满 ⇒ 日桶一条都挤不进去(P17 的锚点日已被选中却 0 条进 pack)。
123
+ #: 现按会话 4 + 日 4 分配:日桶保证拿到 DEFAULT_CAP - SESSION_QUOTA 个名额。
124
+ SESSION_QUOTA = 4
125
+ #: 早段意图下的会话配额(把名额让给日桶)
126
+ SESSION_QUOTA_EARLY = 2
127
+ #: 日桶回溯天数上限(从最近一天往回数,够覆盖「前几天说的」)。
128
+ DAY_LOOKBACK = 14
129
+
130
+ #: 「早段意图」标记:问的是很早以前的事,回落「最近优先」必然错过。
131
+ #: 依据(2026-11-10 外测):10 个稳定失败全是这类——「江北回声那回」「我最早说…」「我最初说…」。
132
+ _EARLY_RE = re.compile(r"最早|最初|第一次|头一次|头回|初次|当初|以前|从前|很久以前|"
133
+ r"上次|上一回|上回|那回|那次|当年|那时候")
134
+
135
+ #: 早段意图下向前扫多少个日桶(成本上界:每桶 ≤ MAX_SCAN 条读盘)。
136
+ EARLY_SCAN_DAYS = 16
137
+
138
+ #: 早段意图的日桶集合:**最早几天 + 全时段均匀采样**。
139
+ #: 教训(2026-11-10 外测 v2.0):只扫「最早 12 天」会漏掉「库内最后一天」这种锚点
140
+ #: (P16 锚点 2026-05-03 恰是最后一天、P20 锚点在第 18 天),故改为全时段覆盖。
141
+ def _early_days(all_days, limit=EARLY_SCAN_DAYS, head=4, tail=4):
142
+ """两端交替有序池(兼容入口;新代码请用 `_early_candidates`,它把证据也纳入池生成)。"""
143
+ n = len(all_days)
144
+ if n <= limit:
145
+ return list(all_days)
146
+ ends = _two_ended(all_days, head=head, tail=tail)
147
+ chosen = set(ends)
148
+ rest = max(0, limit - len(ends))
149
+ step = max(1, n // max(1, rest)) if rest else n
150
+ uniform = [d for d in all_days[::step] if d not in chosen][:rest]
151
+ return ends + uniform
152
+
153
+
154
+ def _evidence_scores(store, terms):
155
+ """全库每个日桶的**证据分**(= 查询词在该桶词表里出现的个数),毫秒级不读正文。"""
156
+ table = _bi._load(store, _bi.TERMS_FILE) or {}
157
+ scores = {}
158
+ for bucket, bag in table.items():
159
+ key = str(bucket)
160
+ if not key.startswith("day:"):
161
+ continue
162
+ d = key[4:]
163
+ scores[d] = sum(1 for t in terms if t in bag)
164
+ return scores
165
+
166
+
167
+ def _order_by_evidence(store, days, terms):
168
+ """把给定候选日按证据强弱排序(同分保持入参顺序)。"""
169
+ scores = _evidence_scores(store, terms)
170
+ pos = {d: i for i, d in enumerate(days)}
171
+ return sorted(days, key=lambda d: (-scores.get(d, 0), pos[d]))
172
+
173
+
174
+ def _early_candidates(store, all_days, terms, limit=EARLY_SCAN_DAYS, head=4, tail=4):
175
+ """早段意图的**候选池生成**:证据优先,两端兜底,均匀补齐。
176
+
177
+ 三条并列病因(外测 2026-11-11 #858 判别):
178
+ ① 名额:锚点日在池内排第 5,而日名额只有 4;
179
+ ② 桶内闸:该日词表与 query 无交集 ⇒ `_pick` 不补足(该桶是两端交替选进来的);
180
+ ③ 池生成:均匀采样漏掉**有强证据**的那天(P20:04-30)⇒ 证据排序在池外无从发力。
181
+ 本函数治第 ③ 条:**先在全库按证据挑**,再补两端,最后均匀补齐。
182
+ """
183
+ n = len(all_days)
184
+ if n <= limit:
185
+ return list(all_days)
186
+ scores = _evidence_scores(store, terms)
187
+ ends = _two_ended(all_days, head=head, tail=tail)
188
+ end_rank = {d: i for i, d in enumerate(ends)}
189
+ # 证据优先;同分时按「两端交替」序(末端与首端先占位),其余按日期
190
+ ordered = sorted(all_days, key=lambda d: (-scores.get(d, 0),
191
+ end_rank.get(d, 10 ** 6), d))
192
+ out = ordered[:limit]
193
+ # 两端必须仍在池里(证据为 0 的末端日靠这条兜底)
194
+ for d in ends:
195
+ if d not in out:
196
+ out.append(d)
197
+ return out[:limit + len(ends)]
198
+
199
+
200
+ def _two_ended(all_days, head=4, tail=4):
201
+ """两端交替有序列表:[末1,首1,末2,首2,…,末k,首k]。"""
202
+ heads = list(all_days[:head])
203
+ tails = list(all_days[-tail:])
204
+ order = []
205
+ for i in range(max(len(heads), len(tails))):
206
+ if i < len(tails):
207
+ order.append(tails[-1 - i])
208
+ if i < len(heads):
209
+ order.append(heads[i])
210
+ return order
211
+
212
+
213
+ def _early_intent(query):
214
+ """query 是否表达了「早段意图」(要很早的记忆,而不是最近的)。"""
215
+ return bool(_EARLY_RE.search(str(query or "")))
216
+
217
+
218
+ def _is_expired(node, now=None):
219
+ """节点是否已过期(valid_until 早于 now)。
220
+
221
+ 判据与检索面一致:只认 frontmatter.valid_until(ISO 日期/时间),解析失败不算过期
222
+ (宁放过不错杀——过期是显式字段,不是推断)。
223
+ """
224
+ fm = node.get("frontmatter") or {}
225
+ until = fm.get("valid_until") or fm.get("validUntil")
226
+ if not until:
227
+ return False
228
+ t = now if now is not None else time.time()
229
+ raw = str(until).strip()
230
+ for fmt in ("%Y-%m-%dT%H:%M:%S", "%Y-%m-%d %H:%M:%S", "%Y-%m-%d"):
231
+ try:
232
+ return time.mktime(time.strptime(raw[:19], fmt)) < float(t)
233
+ except ValueError:
234
+ continue
235
+ return False
236
+
237
+
238
+ def _recency_key(node):
239
+ """桶内排序键:created_at 降序(最近的在前)。"""
240
+ fm = node.get("frontmatter") or {}
241
+ try:
242
+ return -float(fm.get("created_at") or fm.get("valid_from") or 0)
243
+ except (TypeError, ValueError):
244
+ return 0.0
245
+
246
+
247
+ def scope_recall(store, query, session, now, cap=DEFAULT_CAP, validity=False):
248
+ """**默认口径**的桶作用域召回:会话桶优先 → 日桶按日期倒排 → 都按时间降序。
249
+
250
+ 使用者口径(2026-11-09):
251
+ 「日期、会话是默认就应该开启的;会话优先使用会话桶,然后按日期倒排,
252
+ 优先取自己最近的。条件才是筛选方式。」
253
+
254
+ 故本函数与「重排」无关——它只**补召回**:返回的每条 score=0.0,
255
+ 不抢正排名次,但一定进入候选(调用方可按预算追加)。
256
+ """
257
+ terms = _terms(query)
258
+ # 【外测 v2.0 建议 3】把解析出的**绝对日期**回吐给调用方(metadata.resolved_dates):
259
+ # 作答侧据此直接引用日期,不必自己把「上周五」换算成具体日子(B 类 7 题的根因之一)。
260
+ resolved_dates = []
261
+ try:
262
+ from . import timeboost as _tbd # noqa: PLC0415
263
+ resolved_dates = sorted(_tbd.query_dates(query, now))
264
+ except Exception: # noqa: BLE001 —— 解析失败不影响召回
265
+ resolved_dates = []
266
+ out = []
267
+ seen = set()
268
+ # ① 会话桶优先(自己的最近记录)
269
+ sid = str(session or "")
270
+ if sid and sid != _bi.UNATTRIBUTED:
271
+ ids = _bi.session_ids(store, sid)
272
+ # 【外测 #858 病因①】早段意图问的是「很久以前」,本会话桶的相关性天然更低:
273
+ # 会话配额从 4 降到 2,把 6 个名额留给日桶(P16 的锚点日在池内排第 5,4 个名额挤不进)。
274
+ _sess_quota = SESSION_QUOTA_EARLY if _early_intent(query) else SESSION_QUOTA
275
+ picked = _pick(store, ids, terms, min(cap, _sess_quota), validity=validity, now=now)
276
+ for node, score, qual, prov in picked:
277
+ qual = dict(qual)
278
+ qual["bucket"] = "session"
279
+ out.append((node, score, qual, prov))
280
+ seen.add(node.get("id"))
281
+ # ② 桶顺序:**范围选择优先**(状态追踪式共现推断),否则按日期倒排。
282
+ # MDCG_SCOPE_SELECT=1(缺省)→ 用 scopeselect 排序;=0 回落「最近优先」。
283
+ # 依据:bench 实测共现推断时间桶 94.20%(字面无关),词面消融仅 90.76%。
284
+ _ordered_days = None
285
+ _declared_days = False # 这些日桶是否来自「声明日期 / 共现推断」
286
+ _early_hit = False
287
+ try:
288
+ if os.environ.get("MDCG_SCOPE_SELECT", "1") != "0":
289
+ from . import scopeselect as _ss # noqa: PLC0415
290
+ _picked = _ss.pick_scopes(store, query, session=session,
291
+ limit=DAY_LOOKBACK, now=now)
292
+ _ordered_days = [key for kind, key, _s in _picked if kind == "day"]
293
+ _declared_days = bool(_ordered_days)
294
+ except Exception: # noqa: BLE001 —— 推断失败回落最近优先
295
+ _ordered_days = None
296
+ if not _ordered_days:
297
+ # 【2026-11-10 外测反馈】「最早 / 最初 / 上次 / 那回」这类**时间副词**查询,
298
+ # query 里没有可解析的绝对日期 ⇒ 选桶为空 ⇒ 若一路回落「最近 14 天」,
299
+ # 3–5 个月前的锚点必然被排除(实测 10/17 稳定失败属此类)。
300
+ # 故:早段意图优先扫**最早的日子**,其余仍回落最近优先。
301
+ _all_days = sorted(_bi.all_days(store))
302
+ _early_hit = bool(_early_intent(query) and _all_days)
303
+ if _early_hit:
304
+ _ordered_days = _early_candidates(store, _all_days, terms)
305
+ else:
306
+ _ordered_days = _all_days[::-1][:DAY_LOOKBACK]
307
+ # ② 日桶按上述顺序取
308
+ if len(out) < cap:
309
+ days = list(_ordered_days or [])
310
+ # 【外测 2026-11-11】跨日桶**轮转分配**:早先每天拿走「剩余全部名额」,
311
+ # 第一个桶条目够多就把 cap 吃光(实测 8 桶 cap=4 只覆盖到 2 桶,末日永不入选)。
312
+ # 现每轮每天最多取 per_day 条,保证「任何一天拿第 2 条之前,所有天都先拿到第 1 条」。
313
+ per_day = max(1, cap // max(1, len(days))) if days else cap
314
+ # 判据:桶是**按理由选进来**的(声明日期 / 早段意图 / 共现)⇒ 桶内补足;
315
+ # 若只是「最近 N 天」兜底,桶本身不携带查询证据,仍严格按词面筛选(避免塞无关条目)。
316
+ _fill = bool(_declared_days or _early_hit)
317
+ for day in days:
318
+ if len(out) >= cap:
319
+ break
320
+ picked = _pick(store, _bi.day_ids(store, day), terms,
321
+ min(per_day, cap - len(out)),
322
+ validity=validity, now=now, fallback_recency=True,
323
+ fill_rest=_fill)
324
+ for node, score, qual, prov in picked:
325
+ if node.get("id") in seen:
326
+ continue
327
+ qual = dict(qual)
328
+ qual["bucket"] = "day:%s" % day
329
+ if resolved_dates:
330
+ qual["resolved_dates"] = resolved_dates
331
+ out.append((node, score, qual, prov))
332
+ seen.add(node.get("id"))
333
+ # ③ 排序:**会话桶整体优先**,再按时间降序;日桶同样按时间降序。
334
+ # 使用者口径「会话优先使用会话桶,然后按日期倒排」——只按时间排会让日桶插到会话前面(实测发生过)。
335
+ def _key(it):
336
+ bucket = ((it[2] or {}).get("bucket") or "")
337
+ return (0 if bucket == "session" else 1, _recency_key(it[0]))
338
+ out.sort(key=_key)
339
+ return out[:cap]
@@ -263,8 +263,8 @@ CONST_ROWS = [
263
263
  ('MAX_CP', 'CALIB', 'calib', 'const', True, '1114111', 'int', '见 default(上限/条数)', '', '', 'const_assign', 'scripts/gen_id_charset_blocks.py', 'MAX_CP', 'uncalibrated', '无(未见标定注记)', 'unknown', ''),
264
264
  ('ACTIVE_WINDOW_S', 'TUNE', 'tune', 'const', True, '300', 'int', '见 default(超时/时间窗)', '', '', 'const_assign', 'scripts/mdcg_verify_render_meta.py', 'ACTIVE_WINDOW_S', 'uncalibrated', '无(未见标定注记)', 'unknown', '源码注释:「新鲜写入」窗口:重建后落入此窗的旧 render 即活跃污染证据'),
265
265
  ('SNIPPET_MAX', 'CALIB', 'calib', 'const', True, '30', 'int', '见 default(上限/条数)', '', '', 'const_assign', 'scripts/probe_b_state_events_v4.py', 'SNIPPET_MAX', 'uncalibrated', '无(未见标定注记)', 'unknown', '源码注释:证据短证硬上限(≤30 字)'),
266
- ('BIG_LIMIT', 'TUNE', 'tune', 'const', True, '1000000000', 'int', '见 default(上限/条数)', '', '', 'const_assign', 'scripts/probe_c_ledger_integrity.py', 'BIG_LIMIT', 'uncalibrated', '无(未见标定注记)', 'unknown', ''),
267
- ('ANCHOR_WINDOW', 'TUNE', 'tune', 'const', False, '[1.0, 100000000000.0]', 'list', '见 default(枚举/表)', '', '', 'const_assign', 'scripts/probe_d_retirement_discipline.py', 'ANCHOR_WINDOW', 'uncalibrated', '无(未见标定注记)', 'unknown', ''),
266
+ ('BIG_LIMIT', 'TUNE', 'tune', 'const', True, '1000000000', 'int', '见 default(上限/条数)', '', '', 'const_assign', 'md_cg/probe_c_ledger_integrity.py', 'BIG_LIMIT', 'uncalibrated', '无(未见标定注记)', 'unknown', ''),
267
+ ('ANCHOR_WINDOW', 'TUNE', 'tune', 'const', False, '[1.0, 100000000000.0]', 'list', '见 default(枚举/表)', '', '', 'const_assign', 'md_cg/probe_d_retirement_discipline.py', 'ANCHOR_WINDOW', 'uncalibrated', '无(未见标定注记)', 'unknown', ''),
268
268
  ('FULL_WINDOW_HI', 'TUNE', 'tune', 'const', True, '9999999999.0', 'float', '见 default(超时/时间窗)', '', '', 'const_assign', 'scripts/retr_time_coverage.py', 'FULL_WINDOW_HI', 'uncalibrated', '无(未见标定注记)', 'unknown', ''),
269
269
  ('MAX_FILES', 'CALIB', 'calib', 'const', True, '500', 'int', '见 default(上限/条数)', '', '', 'const_assign', 'scripts/review_reconcile.py', 'MAX_FILES', 'uncalibrated', '无(未见标定注记)', 'unknown', ''),
270
270
  ('MAX_ITEMS', 'CALIB', 'calib', 'const', True, '2000', 'int', '见 default(上限/条数)', '', '', 'const_assign', 'scripts/review_reconcile.py', 'MAX_ITEMS', 'uncalibrated', '无(未见标定注记)', 'unknown', ''),
@@ -349,6 +349,21 @@ ENV_ROWS = [
349
349
  ('MDCG_PREFILTER', 'TUNE', 'tune', 'env', True, "'0'", 'str', 'value', 'MDCG_PREFILTER', 'value', 'env_literal', 'md_cg/mdcos.py', '', 'uncalibrated', '2026-10-21 手补:PR 内新增开关(未设=开/缺省即字面量)', 'unknown', ''),
350
350
  ('MDCG_PREFILTER_MIN', 'TUNE', 'tune', 'env', True, "'50'", 'str', 'value', 'MDCG_PREFILTER_MIN', 'value', 'env_literal', 'md_cg/mdcos.py', '', 'uncalibrated', '2026-10-21 手补:PR 内新增开关(未设=开/缺省即字面量)', 'unknown', ''),
351
351
  ('MDCG_RECONCILE_BG', 'TUNE', 'tune', 'env', True, "'1'", 'str', 'value', 'MDCG_RECONCILE_BG', 'value', 'env_literal', 'md_cg/mcp_server.py', '', 'uncalibrated', '2026-10-21 手补:PR 内新增开关', 'unknown', ''),
352
+ ('MDCG_RECENT_TAIL', 'TUNE', 'tune', 'env', True, "'6'", 'str', 'value', 'MDCG_RECENT_TAIL', 'value', 'env_literal', 'md_cg/mdcos.py', '', 'uncalibrated', '2026-11-07:search_rrf 的 cutoff 邻近保障(Ornn8 测评建议②),0=关'),
353
+ ('MDCG_TOPIC_AGG', 'TUNE', 'tune', 'env', True, "'1'", 'str', 'value', 'MDCG_TOPIC_AGG', 'value', 'env_literal', 'md_cg/mdcos.py', '', 'uncalibrated', '2026-11-07:跨轮状态聚合(Ornn8 测评建议①),0=关'),
354
+ ('MDCG_DET_RERANK', 'TUNE', 'tune', 'env', True, "'0'", 'str', 'value', 'MDCG_DET_RERANK', 'value', 'env_literal', 'md_cg/mdcos.py', '', 'uncalibrated', '2026-11-07:确定性四信号重排(向量只给参考),0=关'),
355
+ ('MDCG_DET_MODE', 'TUNE', 'tune', 'env', True, "'tiebreak'", 'str', 'value', 'MDCG_DET_MODE', 'value', 'env_literal', 'md_cg/mdcos.py', '', 'uncalibrated', '2026-11-08:确定性重排模式 tiebreak|window|replace(缺省 tiebreak)'),
356
+ ('MDCG_TOPIC_AGG_TAGFALLBACK', 'TUNE', 'tune', 'env', True, "'0'", 'str', 'value', 'MDCG_TOPIC_AGG_TAGFALLBACK', 'value', 'env_literal', 'md_cg/topicagg.py', '', 'uncalibrated', '2026-11-08:topicagg 的 tag 召回兜底(缺省关;开启需接受泛词噪声)'),
357
+ ('MDCG_STRUCT_BOOST', 'TUNE', 'tune', 'env', True, "'0'", 'str', 'value', 'MDCG_STRUCT_BOOST', 'value', 'env_literal', 'md_cg/mdcos.py', '', 'uncalibrated', '2026-11-09:结构信号重排(章节路由+元素覆盖),0=关;需入库期 ch:N 盖章'),
358
+ ('MDCG_TIME_BOOST', 'TUNE', 'tune', 'env', True, "'0'", 'str', 'value', 'MDCG_TIME_BOOST', 'value', 'env_literal', 'md_cg/mdcos.py', '', 'uncalibrated', '2026-11-09:时间桶重排(主库章节=写入日期),0=关;需调用方显式给 now'),
359
+ ('MDCG_BUCKET_SCOPE', 'TUNE', 'tune', 'env', True, "'1'", 'str', 'value', 'MDCG_BUCKET_SCOPE', 'value', 'env_literal', 'md_cg/mdcos.py', '', 'uncalibrated', '2026-11-09:日/会话桶作用域召回(通用默认开,0=关),会话优先+日倒排'),
360
+
361
+
362
+ ('MDCG_SCOPE_SELECT', 'TUNE', 'tune', 'env', True, "'1'", 'str', 'value', 'MDCG_SCOPE_SELECT', 'value', 'env_literal', 'md_cg/bucketscope.py', '', 'uncalibrated', '2026-11-09:范围选择(槽位共现推断时间桶),0=回落最近优先'),
363
+ ('MDCG_SCOPESEL_COOC', 'TUNE', 'tune', 'env', True, "'1'", 'str', 'value', 'MDCG_SCOPESEL_COOC', 'value', 'env_literal', 'md_cg/scopeselect.py', '', 'uncalibrated', '2026-11-09:共现打分路(默认开——bench 生产侧 97.83%,idf 对照 35.33%)'),
364
+ ('MDCG_BUCKET_AUTO', 'TUNE', 'tune', 'env', True, "'1'", 'str', 'value', 'MDCG_BUCKET_AUTO', 'value', 'env_literal', 'md_cg/bucketindex.py', '', 'uncalibrated', '2026-11-09:桶索引每日自动重建(info 触发),0=关'),
365
+ ('MDCG_BUCKET_INTERVAL_H', 'TUNE', 'tune', 'env', True, "'20'", 'str', 'value', 'MDCG_BUCKET_INTERVAL_H', 'value', 'env_literal', 'md_cg/bucketindex.py', '', 'uncalibrated', '2026-11-09:重建间隔小时(缺省 20;0=强制)'),
366
+
352
367
  ('MDCG_MIN_SCORE', 'TUNE', 'tune', 'env', True, "'0'", 'str', 'value', 'MDCG_MIN_SCORE', 'value', 'env_literal', 'md_cg/mdcg.py', '', 'uncalibrated', '2026-10-23 手补:issue #637 可配最低分(0=关)', 'unknown', ''),
353
368
  ('MDCG_RECALL_MIN_SCORE', 'TUNE', 'tune', 'env', True, "'-1'", 'str', 'value', 'MDCG_RECALL_MIN_SCORE', 'value', 'env_literal', 'md_cg/mdcos.py', '', 'uncalibrated', '2026-10-23 手补:issue #637 召回面最低分(-1=关)', 'unknown', ''),
354
369
  ]
@@ -402,7 +417,6 @@ ENV_CATEGORY_ROWS = [
402
417
  ('LINGSHU_KB', 'env', 'dsh-memory-main', '目录路径', ''),
403
418
  ('LINGSHU_PYTHON / LINGSHU_REPO', 'env', 'dsh-memory-main', '解释器路径 / 目录', ''),
404
419
  ('MAGICK_HOME / ImageMagick_HOME', 'env', 'dsh-memory-main', '目录路径', '经 *_ENV 常量间接读取,朴素字面量扫描抓不到'),
405
- ('MDCG_ACTOR', 'env', 'dsh-memory-main', '字符串', '★ 两处默认值不同(同一 env 名,默认分叉)'),
406
420
  ('MDCG_ALLOW_ROOT_ISSUE', 'env', 'dsh-memory-main', '开关(1/true/yes/on 放行管理级令牌自铸)', '★ #543 签发授权闸的运维逃生口;缺省未设=关(tokens.py:471)'),
407
421
  ('MDCG_ARCHIVE_ROOT', 'env', 'dsh-memory-main', '目录路径', ''),
408
422
  ('MDCG_AUX_ROOT', 'env', 'dsh-memory-main', '目录路径', ''),
@@ -410,16 +424,13 @@ ENV_CATEGORY_ROWS = [
410
424
  ('MDCG_CAN_ADMIN', 'env', 'dsh-memory-main', 'bool 字符串 {1,true,True}', ''),
411
425
  ('MDCG_CAN_WRITE', 'env', 'dsh-memory-main', 'bool 字符串', ''),
412
426
  ('MDCG_CHILD_CWD', 'env', 'dsh-memory-main', '目录路径', '只在 TS 面定义/传递'),
413
- ('MDCG_CLEARANCE', 'env', 'dsh-memory-main', '枚举 {public,internal,restricted,private,secret}', ''),
414
427
  ('MDCG_CODE_TEST_CMD / MDCG_CODE_TEST_TIMEOUT', 'env', 'dsh-memory-main', '命令串 / 秒(int)', 'CODE_TEST_CMD 属权限面(执行外部命令)'),
415
428
  ('MDCG_DATA_ROOT', 'env', 'dsh-memory-main', '目录路径', ''),
416
429
  ('MDCG_DSH_SESSIONS_ROOT', 'env', 'dsh-memory-main', '目录路径', ''),
417
- ('MDCG_EXPORT_ROOT', 'env', 'dsh-memory-main', 'path 列表(os.pathsep 分隔)', '部署期写路径白名单'),
418
430
  ('MDCG_FRESHNESS_NOW', 'env', 'dsh-memory-main', 'float(可复算冻结口)', '供两侧探针逐位比对'),
419
431
  ('MDCG_GATE_SANDBOX_DIR', 'env', 'dsh-memory-main', '目录路径', ''),
420
432
  ('MDCG_HARNESS / MDCG_UNIT', 'env', 'dsh-memory-main', '字符串', '服务端身份面(同样有 MDCG_NODE_ID/MDCG_SWARM_DIR 归因面)'),
421
433
  ('MDCG_HIVE_JOBS', 'env', 'dsh-memory-main', '目录路径', ''),
422
- ('MDCG_INGEST_ROOT', 'env', 'dsh-memory-main', 'path 列表(os.pathsep 分隔)', '部署期写路径白名单'),
423
434
  ('MDCG_JUDGMENT_MANIFEST', 'env', 'dsh-memory-main', '文件路径', ''),
424
435
  ('MDCG_LEGACY_ENV_ADMIN', 'env', 'dsh-memory-main', 'bool 字符串', ''),
425
436
  ('MDCG_LEGACY_ENV_AUTH', 'env', 'dsh-memory-main', 'bool 字符串 {1,true,True}', ''),
@@ -429,7 +440,6 @@ ENV_CATEGORY_ROWS = [
429
440
  ('MDCG_LLM_KEY / MDCG_LLM_BASE / MDCG_LLM_MODEL', 'env', 'dsh-memory-main', '密钥 / URL / 模型名', ''),
430
441
  ('MDCG_QMD_LLM_BASE / MDCG_QMD_LLM_MODEL / MDCG_QMD_LLM_KEY', 'env', 'dsh-memory-main', 'URL / 模型名 / 密钥(敏感)', 'QMD 的 LLM 接入(md_cg/qmd_llm.py);三者缺一即 LLMUnavailable,不回落 MDCG_LLM_KEY / DEEPSEEK_API_KEY'),
431
442
  ('MDCG_MASTER_KEY', 'env', 'dsh-memory-main', '密钥(敏感,hex)', '敏感值不回显'),
432
- ('MDCG_MCP_SURFACE', 'env', 'dsh-memory-main', '枚举 {kernel, full}', 'kernel 只暴露 cg/stg 两基元'),
433
443
  ('MDCG_POLICY_FILE', 'env', 'dsh-memory-main', '文件路径', ''),
434
444
  ('MDCG_PYTHON', 'env', 'dsh-memory-main', '解释器路径', '★ 动态键读取,朴素 grep 抓不到'),
435
445
  ('MDCG_REACH_INDEX', 'env', 'dsh-memory-main', '文件路径', ''),
@@ -439,10 +449,11 @@ ENV_CATEGORY_ROWS = [
439
449
  ('MDCG_SESSION', 'env', 'dsh-memory-main', '字符串', 'TS 面已改为「请求声明归因」优先(见 src/tools.ts 注)'),
440
450
  ('MDCG_SIGNERS_FILE / MDCG_SIGNER_KEY_FILE / MDCG_SIGNER_MODULE', 'env', 'dsh-memory-main', '文件路径 / 文件路径 / 模块名', ''),
441
451
  ('MDCG_STATE_ROOT', 'env', 'dsh-memory-main', '目录路径', ''),
452
+ ('MDCG_STATE_EXTRACT', 'env', 'dsh-memory-main', '开关', 'L2 状态台账抽取开关:缺省开,置 0 关闭(md_cg/state_extract.py:76;recent·add 接线)'),
453
+ ('MDCG_STATE_SUBJECT', 'env', 'dsh-memory-main', '字符串', 'L2 台账抽取主体覆盖,缺省「我」(md_cg/state_extract.py:108)'),
442
454
  ('MDCG_SUSTAIN_DIR', 'env', 'dsh-memory-main', '目录路径', ''),
443
455
  ('MDCG_SUSTAIN_NAME', 'env', 'dsh-memory-main', '字符串', ''),
444
456
  ('MDCG_SWARM_DIR', 'env', 'dsh-memory-main', '目录路径', ''),
445
- ('MDCG_TENANT', 'env', 'dsh-memory-main', '租户名', "★ 注意:src/index.ts 另有 MDCG_TENANT 默认 'default'(插件 config.mdcg.tenant)"),
446
457
  ('MDCG_TENANT_REGISTRY', 'env', 'dsh-memory-main', '文件路径', ''),
447
458
  ('MDCG_TESTLOG_DIR', 'env', 'dsh-memory-main', '目录路径', ''),
448
459
  ('MDCG_THEORY_FILE / MDCG_THEORY', 'env', 'dsh-memory-main', '文件路径', ''),
@@ -476,9 +487,7 @@ ENV_CATEGORY_ROWS = [
476
487
  ('GITHUB_ACTIONS / PR_TITLE', 'env', 'lingshu', '字符串', ''),
477
488
  ('LINGSHU_COMPONENT_ROOT', 'env', 'lingshu', '路径', ''),
478
489
  ('LINGSHU_DATA_ROOT', 'env', 'lingshu', '路径', ''),
479
- ('MDCG_BRAIN_PYTHONPATH', 'env', 'lingshu', '路径列表', ''),
480
490
  ('MDCG_LEGACY_ENV_AUTH / MDCG_CAN_ADMIN / MDCG_LEGACY_ENV_ADMIN', 'env', 'lingshu', 'bool 语义', ''),
481
- ('MDCG_ROOT', 'env', 'lingshu', '路径', ''),
482
491
  ('PYTHONUTF8 / PYTHONIOENCODING', 'env', 'lingshu', '字符串', ''),
483
492
  ('MDCG_REPLAY_NEG_MAX_COVERAGE', 'env', 'dsh-memory-main', 'float(正数;非法/非正回落默认)', 'issue #278:consolidate neg_separated 的词面覆盖率阈值(默认 0.5)'),
484
493
  ]
@@ -26,12 +26,17 @@ RULES = {
26
26
  'MDCG_AUTO_SCRUB': {'bool_like': True, 'type': 'bool'},
27
27
  'MDCG_AUTO_TIDY': {'bool_like': True, 'type': 'bool'},
28
28
  'MDCG_BM25': {'bool_like': True, 'true_literals': ['1'], 'type': 'bool'},
29
+ 'MDCG_BUCKET_AUTO': {'type': 'str'},
30
+ 'MDCG_BUCKET_INTERVAL_H': {'type': 'str'},
29
31
  'MDCG_BUCKET_MIN_SIM': {'max': 1.0, 'min': 0.0, 'type': 'float'},
32
+ 'MDCG_BUCKET_SCOPE': {'type': 'str'},
30
33
  'MDCG_BUCKET_TOPK': {'type': 'str'},
31
34
  'MDCG_CHAIN_TYPES': {'type': 'str'},
32
35
  'MDCG_CLEARANCE': {'enum': ['public', 'internal', 'restricted', 'private', 'secret'],
33
36
  'type': 'str'},
34
37
  'MDCG_CN_GRAMS': {'bool_like': True, 'true_literals': ['1'], 'type': 'bool'},
38
+ 'MDCG_DET_MODE': {'type': 'str'},
39
+ 'MDCG_DET_RERANK': {'type': 'str'},
35
40
  'MDCG_D_META': {'bool_like': True, 'type': 'bool'},
36
41
  'MDCG_EN_ATOMS': {'enum': ['norm', 'raw'], 'type': 'str'},
37
42
  'MDCG_EXPLORE_BUDGET_MAX': {'type': 'str'},
@@ -67,11 +72,14 @@ RULES = {
67
72
  'MDCG_REACH_TTL': {'min': 0.0, 'type': 'float'},
68
73
  'MDCG_READ_CACHE': {'bool_like': True, 'true_literals': ['1'], 'type': 'bool'},
69
74
  'MDCG_RECALL_MIN_SCORE': {'type': 'str'},
75
+ 'MDCG_RECENT_TAIL': {'type': 'str'},
70
76
  'MDCG_RECONCILE': {'bool_like': True, 'type': 'bool'},
71
77
  'MDCG_RECONCILE_BG': {'type': 'str'},
72
78
  'MDCG_RELOAD_MIN_INTERVAL': {'type': 'str'},
73
79
  'MDCG_RETRIEVAL_PIPELINE': {'bool_like': True, 'true_literals': ['1'], 'type': 'bool'},
74
80
  'MDCG_S7_FRESHNESS': {'bool_like': True, 'true_literals': ['1'], 'type': 'bool'},
81
+ 'MDCG_SCOPESEL_COOC': {'type': 'str'},
82
+ 'MDCG_SCOPE_SELECT': {'type': 'str'},
75
83
  'MDCG_SCORE_MODE': {'enum': ['legacy', 'jaccard'], 'type': 'str'},
76
84
  'MDCG_SEMANTIC': {'bool_like': True, 'true_literals': ['1'], 'type': 'bool'},
77
85
  'MDCG_SLEEP': {'bool_like': True, 'type': 'bool'},
@@ -85,12 +93,16 @@ RULES = {
85
93
  'MDCG_SPREAD_GAIN': {'max': 1.0, 'min': 0.0, 'type': 'float'},
86
94
  'MDCG_SPREAD_HOPS': {'min': 1, 'type': 'int'},
87
95
  'MDCG_STATUS_HEAD': {'bool_like': True, 'true_literals': ['1'], 'type': 'bool'},
96
+ 'MDCG_STRUCT_BOOST': {'type': 'str'},
88
97
  'MDCG_SUSTAIN': {'bool_like': True, 'type': 'bool'},
89
98
  'MDCG_SUSTAIN_AUTOHEAL': {'bool_like': True, 'type': 'bool'},
90
99
  'MDCG_SUSTAIN_BEAT_THREAD': {'bool_like': True, 'type': 'bool'},
91
100
  'MDCG_TEMPORAL_GAMMA': {'min': 0.0, 'type': 'float'},
92
101
  'MDCG_TENANT': {'type': 'str'},
102
+ 'MDCG_TIME_BOOST': {'type': 'str'},
93
103
  'MDCG_TOOL_FACE': {'bool_like': True, 'true_literals': ['1'], 'type': 'bool'},
104
+ 'MDCG_TOPIC_AGG': {'type': 'str'},
105
+ 'MDCG_TOPIC_AGG_TAGFALLBACK': {'type': 'str'},
94
106
  'MDCG_UNIFY_QUERY': {'bool_like': True, 'true_literals': ['1'], 'type': 'bool'},
95
107
  'MDCG_WARM_SEARCH': {'type': 'str'},
96
108
  'MDCG_WRITELIMIT': {'bool_like': True, 'type': 'bool'},
@@ -231,6 +243,8 @@ RULES = {
231
243
  'dsh-memory-main::md_cg/predict.py::PREFERENCE_THRESHOLD': {'type': 'float'},
232
244
  'dsh-memory-main::md_cg/predict.py::SEMANTIC_MIN_SIM': {'type': 'float'},
233
245
  'dsh-memory-main::md_cg/predict.py::SEMANTIC_TOP_K': {'type': 'int'},
246
+ 'dsh-memory-main::md_cg/probe_c_ledger_integrity.py::BIG_LIMIT': {'type': 'int'},
247
+ 'dsh-memory-main::md_cg/probe_d_retirement_discipline.py::ANCHOR_WINDOW': {'type': 'str'},
234
248
  'dsh-memory-main::md_cg/progressive.py::MIN_DISCRIM': {'type': 'float'},
235
249
  'dsh-memory-main::md_cg/provenance.py::DEFAULT_FIND_LIMIT': {'type': 'int'},
236
250
  'dsh-memory-main::md_cg/provenance.py::MAX_FIND_LIMIT': {'type': 'int'},
@@ -322,8 +336,6 @@ RULES = {
322
336
  'dsh-memory-main::scripts/gen_id_charset_blocks.py::MAX_CP': {'type': 'int'},
323
337
  'dsh-memory-main::scripts/mdcg_verify_render_meta.py::ACTIVE_WINDOW_S': {'type': 'int'},
324
338
  'dsh-memory-main::scripts/probe_b_state_events_v4.py::SNIPPET_MAX': {'type': 'int'},
325
- 'dsh-memory-main::scripts/probe_c_ledger_integrity.py::BIG_LIMIT': {'type': 'int'},
326
- 'dsh-memory-main::scripts/probe_d_retirement_discipline.py::ANCHOR_WINDOW': {'type': 'str'},
327
339
  'dsh-memory-main::scripts/retr_time_coverage.py::FULL_WINDOW_HI': {'type': 'float'},
328
340
  'dsh-memory-main::scripts/review_reconcile.py::MAX_FILES': {'type': 'int'},
329
341
  'dsh-memory-main::scripts/review_reconcile.py::MAX_ITEMS': {'type': 'int'},
@@ -0,0 +1,137 @@
1
+ # -*- coding: utf-8 -*-
2
+ """确定性重排(detrerank)——检索结论的**唯一裁决口**。
3
+
4
+ 裁定(structural: mem_ruling_vector_ref_deterministic):
5
+ 向量/语义相似度 **只提供参考**(候选扩展),结论必须由**词法·时间·空间·交互**
6
+ 四类确定性信号重排,且**必须由确定性规则裁决**:固定常量 + 可数信号,
7
+ 同输入同输出(可对拍、可回放)。
8
+
9
+ 为什么这样切:相似度回答「像不像」,因果观察三元组(时间/空间/交互)回答
10
+ 「是不是真的发生过」,词法回答「原文对不对得上」。三者性质不同,
11
+ 用一个浮点分数混合会让「像」压过「真」——这正是 #753/#790 那类 fail-open 的同源病。
12
+
13
+ 确定性约束(本模块的三条硬纪律):
14
+ 1. **无模型分**:函数体不得出现 embedding / 相似度 / 模型调用;
15
+ 2. **时钟显式**:`now` 必须由调用方传入(不内部取 time.time()),否则不可回放;
16
+ 3. **全序**:分数相同按 id 字典序兜底,保证任意两次结果逐位相同。
17
+ """
18
+ from __future__ import annotations
19
+
20
+ import re
21
+
22
+ #: 四信号权重(固定常量;改值等于改裁决口径,必须走评审 + 读数)。
23
+ WEIGHTS = {
24
+ "lexical": 4, # 词法最重:原文对得上才算证据
25
+ "interaction": 3, # 交互次之:真的发生过作用
26
+ "spatial": 2, # 空间可达:在同域/同链路里
27
+ "temporal": 1, # 时间只做轻微加权(新旧不代表对错)
28
+ }
29
+
30
+ #: 时间衰减半衰期(天)——与 temporal 路同源口径,固定常量,不随环境变。
31
+ TEMPORAL_HALF_LIFE_D = 30.0
32
+
33
+ #: window 模式:只在融合前 WINDOW_SIZE 名内允许有界上浮(防弱条目顶上第一)。
34
+ WINDOW_SIZE = 8
35
+
36
+
37
+ def _terms(text) -> list:
38
+ return [t.lower() for t in re.split(r"[\s,,、;;。!?::()()\[\]【】]+", text or "")
39
+ if len(t) >= 2]
40
+
41
+
42
+ def _snippet(node, limit: int = 1200) -> str:
43
+ parts = [str(node.get("content") or "")[:limit]]
44
+ fm = node.get("frontmatter") or {}
45
+ for key in ("tags", "title", "conditions"):
46
+ val = fm.get(key)
47
+ if val:
48
+ parts.append(str(val))
49
+ return " ".join(parts).lower()
50
+
51
+
52
+ def signals(node, query_terms, now, session=None) -> dict:
53
+ """四信号取值(全部确定性可数)。"""
54
+ text = _snippet(node)
55
+ # 词法:query 词项在原文/标签里的命中数(整数)
56
+ lexical = sum(1 for t in query_terms if t in text)
57
+ fm = node.get("frontmatter") or {}
58
+ # 交互:出边数 + 入标记(edges 是「真的发生过作用」的痕迹)
59
+ edges = fm.get("edges")
60
+ out_edges = len(edges) if isinstance(edges, list) else 0
61
+ refs = fm.get("derived_from") or fm.get("alias_of") or []
62
+ refs_n = len(refs) if isinstance(refs, (list, tuple)) else (1 if refs else 0)
63
+ interaction = out_edges + refs_n
64
+ # 空间:同会话(同域)+ 同层(结构可达)——两个可数布尔
65
+ same_session = 1 if (session and str(fm.get("session") or "") == str(session)) else 0
66
+ layer = str(fm.get("layer") or "")
67
+ spatial = same_session + (1 if layer in ("knowledge", "structural") else 0)
68
+ # 时间:按半衰期衰减到 [0,1] 后取整到 3 位(确定性,无时钟读取)
69
+ try:
70
+ created = float(fm.get("created_at") or 0)
71
+ except (TypeError, ValueError):
72
+ created = 0.0
73
+ if created > 0 and now:
74
+ age_days = max(0.0, (float(now) - created) / 86400.0)
75
+ temporal = round(0.5 ** (age_days / TEMPORAL_HALF_LIFE_D), 3)
76
+ else:
77
+ temporal = 0.0
78
+ return {"lexical": lexical, "interaction": interaction,
79
+ "spatial": spatial, "temporal": temporal}
80
+
81
+
82
+ def score(sig: dict) -> float:
83
+ """加权和(整数权重 × 确定性信号),保留 6 位。"""
84
+ total = 0.0
85
+ for key, weight in WEIGHTS.items():
86
+ total += float(weight) * float(sig.get(key) or 0)
87
+ return round(total, 6)
88
+
89
+
90
+ def rerank(results, query, now, session=None):
91
+ """对 [(node, fused_score, qual, prov), ...] 做确定性重排。
92
+
93
+ 返回新列表(不改入参):按 (det_score desc, id asc) 全序。
94
+ `now` 必须由调用方给(时钟显式,可回放)。
95
+ """
96
+ terms = _terms(query)
97
+ scored = []
98
+ for item in results or ():
99
+ node = item[0]
100
+ sig = signals(node, terms, now, session=session)
101
+ scored.append((score(sig), str(node.get("id") or ""), sig, item))
102
+ scored.sort(key=lambda x: (-x[0], x[1]))
103
+ return [(item, det, sig) for det, _nid, sig, item in scored]
104
+
105
+
106
+ def rerank_v2(results, query, now, mode="tiebreak", session=None):
107
+ """v2:**保留 RRF 为主序**,四信号只做有界修正。
108
+
109
+ bench 读数(hive-memory-bench · e2e-simple/蜂巢指南 38 题 · 块级词面覆盖):
110
+ v1(replace,按 det 总分重排):cov@1 **0.625 → 0.313**(g1 28 → 11)。
111
+ 原因:词法计数奖励「长块/词多」,与「这段是否解答了问题」无关;
112
+ 把 RRF 的候选质量丢掉,等于用一个更粗的信号覆盖更细的信号。
113
+
114
+ 三种模式:
115
+ · `tiebreak`(**缺省**):fused 相同才看四信号——零风险,只在并列时用证据说话;
116
+ · `window`:融合前 WINDOW_SIZE 名内、词法 ≥ 2 的条目允许上浮(有界);
117
+ · `replace`:v1 行为,保留以便复现上述读数,**不建议**。
118
+ """
119
+ terms = _terms(query)
120
+ scored = []
121
+ for idx, item in enumerate(results or ()):
122
+ node = item[0]
123
+ sig = signals(node, terms, now, session=session)
124
+ fused = float(item[1]) if len(item) > 1 and isinstance(item[1], (int, float)) else 0.0
125
+ scored.append({"idx": idx, "id": str(node.get("id") or ""), "sig": sig,
126
+ "det": score(sig), "fused": fused, "item": item})
127
+ if mode == "replace":
128
+ scored.sort(key=lambda x: (-x["det"], x["id"]))
129
+ elif mode == "window":
130
+ scored.sort(key=lambda x: (-x["fused"], x["idx"]))
131
+ head, tail = scored[:WINDOW_SIZE], scored[WINDOW_SIZE:]
132
+ head.sort(key=lambda x: (0 if x["sig"]["lexical"] >= 2 else 1,
133
+ -x["det"], -x["fused"], x["id"]))
134
+ scored = head + tail
135
+ else: # tiebreak(缺省,零风险)
136
+ scored.sort(key=lambda x: (-x["fused"], -x["det"], x["id"]))
137
+ return [(x["item"], x["det"], x["sig"]) for x in scored]