@modusensus/dsh-mneme 0.8.14 → 0.8.16

package/lib/dream/sleep.js

@@ -1,819 +1,819 @@
-import { STR, langOf } from "../lang.js";
-// System-level sleep (v0.4.0): an idle-triggered, LLM-assisted deep pass over
-// the memory store. Four independent, fail-safe phases:
-//   1. conflict resolution — high-similarity same-type pairs are either parked
-//      for review (freeze mode) or adjudicated by the LLM (winner kept / loser
-//      archived), reusing the dream conflict machinery. Strictness-graded.
-//      v0.8.1 (issue #170): cross-scope pairs are ALWAYS parked for human
-//      ownership review with a dedicated reason — never auto-adjudicated.
-//   2. archival demotion — memories unreferenced past sleepArchiveDays shrink
-//      to a one-line summary with the full body moved to _full_content; past
-//      sleepCompressDays they are archived outright.
-//   3. pattern discovery — the LLM scans the most recent memories and mints
-//      type=pattern entries carrying evidence id references.
-//   4. relation completion — orphan entities (zero relations) get implied
-//      relations completed from memory co-occurrence.
-// Each phase is wrapped so one failure never aborts the others, and a missing
-// LLM route / semantic embedder only skips the phases that need it. A run is
-// abortable via an AbortController signal (user activity) — phases check the
-// signal between batches so a running cycle yields promptly.
-import { randomUUID, createHash } from "node:crypto";
-import { validateDecisions, applyDecisions, ACTIONS } from "./decisions.js";
-import { findPotentialConflicts, cosineSimilarity } from "./clustering.js";
-import { scopeKeyOf } from "../scope.js";
-import { buildReceipt, describeStreamFailure, resolveDreamEffort, withEffortFallback, maintainIndexAfterDream } from "../dream.js";
-import { computeHeat } from "../heat.js";
-
-// Issue #126 review（Copilot）：默认档（sleepActionSet="conflict"）允许的动作 = 全局
-// ACTIONS 去掉 #126 新增的两个。从 ACTIONS 派生而非硬编码，将来再扩动作集时自动跟随。
-// 只换 prompt 挡不住模型自发输出新动作，必须由校验器按档位收口。
-const LEGACY_ACTIONS = [...ACTIONS].filter((a) => a !== "supersede" && a !== "differentiate");
-
-const SUMMARY_MAX = 120;
-// Conflict similarity threshold per strictness level (v0.4.0):
-//   gentle     only high-confidence pairs (0.92) — first-time users
-//   normal     standard dream-level (0.85) — default
-//   aggressive low-confidence pairs too (0.75) — bloated stores
-const CONFLICT_THRESHOLDS = { gentle: 0.92, normal: 0.85, aggressive: 0.75 };
-
-function parseJsonArray(text) {
-  if (typeof text !== "string") return undefined;
-  const start = text.indexOf("[");
-  const end = text.lastIndexOf("]");
-  if (start === -1 || end <= start) return undefined;
-  try {
-    const parsed = JSON.parse(text.slice(start, end + 1));
-    return Array.isArray(parsed) ? parsed : undefined;
-  } catch {
-    return undefined;
-  }
-}
-
-/**
- * Same stream consumption contract as dream.js.
- *
- * `onUsage`（optional, issue #250）receives any usage chunk for token accounting.
- * dream.js 那份一直有它；sleep 这份副本没有，导致 conflict / pattern 两条链路的
- * token 从未进 llm_audit_logs（面板因此只看到 autoDream 与 autoSummarize）。
- */
-async function streamText(ctx, options, onUsage, onStreamError) {
-  if (!ctx?.llm?.stream) return undefined;
-  let text = "";
-  for await (const chunk of ctx.llm.stream(options)) {
-    if (chunk.type === "text-delta" && typeof chunk.text === "string") text += chunk.text;
-    // DSH 的 StreamChunk 契约把用量嵌在 chunk.usage（TokenUsage：inputTokens /
-    // outputTokens / totalTokens）；兼容直接平铺在 chunk 上的旧形态。归一放在
-    // 这里，调用侧的 reporter 只需认平铺形状——与 dream.js:239 同一处理。
-    if (chunk.type === "usage" && typeof onUsage === "function") onUsage(chunk.usage ?? chunk);
-    if (chunk.type === "finish" && (chunk.reason?.kind === "error" || chunk.reason?.kind === "aborted")) {
-      // Same rc.1 error-as-finish-chunk behavior as dream.js — surface the
-      // cause instead of discarding it.
-      if (typeof onStreamError === "function") {
-        try { onStreamError(chunk.reason); } catch { /* diagnostics only */ }
-      }
-      return undefined;
-    }
-  }
-  return text;
-}
-
-/**
- * Issue #250：给 sleep 的一条 LLM 调用记一笔 llm_audit_logs（token / 时长 / 状态 /
- * 触发源）。与 dream.js 的 runAuditedLlm 同口径但各自独立一份——dream 那两个
- * checker 的调用形状绑定着它自己的排程细节，硬抽共享抽象只会收敛回 dream 的形态。
- *
- * 记账是 best-effort：写审计行失败只 warn，绝不反噬 sleep 本体（CONTRIBUTING 的
- * fail-safe 硬约定）。`body(reportUsage)` 负责真实消费流，usage chunk 经
- * reportUsage 上报。
- *
- * 审计诚实性：流被中止/报错时 body 返回 undefined，记 status='error'；流式成功但
- * 输出不可用时 `spec.auditError` 给出原因，同样记 error——否则同一轮调用会在
- * 「phase 报 failed」与「audit 报 success」之间自相矛盾（CONTRIBUTING 的审计诚实性
- * 硬约定）。`auditError` 必须是**纯判据函数**：调用方的控制流不能依赖它的副作用，
- * 因为审计关闭时这里整段早退、钩子根本不会被调用（dream 正是那样踩的，见 PR 正文）。
- */
-async function auditedSleepLlm(ctx, service, config, logger, spec, body) {
-  if (config?.llmAudit?.enabled === false || typeof service?.saveLlmAudit !== "function") {
-    return body(() => {});
-  }
-  const startedAt = Date.now();
-  const timestamp = new Date(startedAt).toISOString();
-  let inputTokens = 0;
-  let outputTokens = 0;
-  let status = "success";
-  let errorMessage = null;
-  try {
-    const result = await body((usage) => {
-      if (!usage) return;
-      const i = usage.input_tokens ?? usage.inputTokens ?? usage.prompt_tokens ?? usage.promptTokens;
-      const o = usage.output_tokens ?? usage.outputTokens ?? usage.completion_tokens ?? usage.completionTokens;
-      if (Number.isFinite(i)) inputTokens = i;
-      if (Number.isFinite(o)) outputTokens = o;
-    });
-    if (result === undefined) {
-      status = "error";
-      const streamErr = typeof spec.streamError === "function" ? String(spec.streamError() ?? "") : "";
-      errorMessage = streamErr ? `llm stream aborted or errored (${streamErr})` : "llm stream aborted or errored";
-    } else if (typeof spec.auditError === "function") {
-      // 流式成功但输出不可用，仍然是一次失败的调用——记 error，不与调用方的
-      // failed 返回值打架。
-      const message = spec.auditError(result);
-      if (message) {
-        status = "error";
-        errorMessage = message;
-      }
-    }
-    return result;
-  } catch (error) {
-    status = "error";
-    errorMessage = String(error?.message ?? error);
-    throw error;
-  } finally {
-    try {
-      service.saveLlmAudit({
-        timestamp,
-        trigger_source: spec.triggerSource,
-        operation_type: spec.operationType,
-        model_id: spec.modelId,
-        input_tokens: inputTokens,
-        output_tokens: outputTokens,
-        total_tokens: inputTokens + outputTokens,
-        cost_usd: 0,
-        duration_ms: Date.now() - startedAt,
-        status,
-        error_message: errorMessage,
-        related_memory_ids: spec.relatedMemoryIds ?? []
-      });
-    } catch (auditError) {
-      logger?.warn?.(`dsh-mneme sleep: llm audit write failed: ${String(auditError)}`);
-    }
-  }
-}
-
-/** LLM route (Issue #25): explicit sleepProvider/Model wins, then the dream
- *  route as a shared explicit fallback, then the agent default model. Sleep
- *  can pin a cheaper model for its bulk passes without disturbing the dream
- *  route. Explicit config first — otherwise the config routes are dead code
- *  whenever agentDefaultModel resolves (see resolveRoute in dream.js). */
-function resolveSleepRoute(ctx, config, logger) {
-  if (config.sleepProvider && config.sleepModel) return { provider: config.sleepProvider, model: config.sleepModel };
-  if (config.dreamProvider && config.dreamModel) return { provider: config.dreamProvider, model: config.dreamModel };
-  try {
-    const sel = ctx?.agentDefaultModel?.currentSelection?.();
-    if (sel?.provider && sel?.model) return { provider: sel.provider, model: sel.model };
-  } catch { /* fall through to warn */ }
-  logger?.warn?.("dsh-mneme sleep: no llm route available");
-  return undefined;
-}
-
-function makeSummary(m) {
-  const text = (m.content ?? "").trim();
-  if (!text) return (m.title ?? "").trim();
-  return text.length <= SUMMARY_MAX ? text : `${text.slice(0, SUMMARY_MAX)}…`;
-}
-
-// ---------------------------------------------------------------- phases
-
-/**
- * Phase 1 — conflict resolution. Needs a semantic embedder + vector index.
- * In freeze mode (conflictFreezeEnabled) conflicting pairs are parked in
- * conflict_pending for human review (no LLM). Otherwise the LLM adjudicates:
- * each pair → winner kept / loser archived. Returns a per-run summary.
- */
-async function phaseConflicts(ctx, service, config, logger, runId, semantic = null, signal = null) {
-  const language = langOf(config);
-  const embedder = semantic?.embedder;
-  const vectorIndex = semantic?.vectorIndex;
-  if (!embedder || !vectorIndex || typeof embedder.embed !== "function") {
-    return { status: "skipped", reason: "no semantic embedder" };
-  }
-  const strictness = config.sleepConflictStrictness ?? "normal";
-  const threshold = CONFLICT_THRESHOLDS[strictness] ?? CONFLICT_THRESHOLDS.normal;
-  const memories = service.all().filter((m) => !m.archived && !m.forgotten && m.type !== "summary" && m.type !== "document");
-  if (memories.length < 2) return { status: "skipped", reason: "too few memories" };
-  if (signal?.aborted) return { status: "aborted", reason: "user activity" };
-
-  // Backfill + collect vectors for every eligible memory (best effort).
-  const vectors = new Array(memories.length);
-  const missing = [];
-  for (let i = 0; i < memories.length; i++) {
-    const cached = vectorIndex.getEmbedding?.(memories[i].id);
-    if (cached) vectors[i] = cached;
-    else missing.push(i);
-  }
-  if (missing.length) {
-    try {
-      const texts = missing.map((i) => [memories[i].title, memories[i].content].filter(Boolean).join("\n"));
-      const rows = await embedder.embed(texts);
-      missing.forEach((mi, j) => {
-        if (rows[j]?.length) {
-          vectors[mi] = rows[j];
-          vectorIndex.saveEmbedding?.(memories[mi].id, rows[j]);
-        }
-      });
-    } catch (error) {
-      logger?.warn?.(`dsh-mneme sleep: conflict vector backfill failed: ${String(error)}`);
-    }
-  }
-  const usable = [];
-  for (let i = 0; i < memories.length; i++) {
-    if (vectors[i]?.length) usable.push(i);
-  }
-  if (usable.length < 2) return { status: "skipped", reason: "no usable vectors" };
-  const usableMemories = usable.map((i) => memories[i]);
-  const usableVectors = usable.map((i) => vectors[i]);
-
-  // Issue #126：动作集档位既要决定 prompt，也要决定**校验器允许的动作集合**（见下方
-  // allowedActions），并且在候选构造阶段就要知道——full 档需要补跨类型对。因此在这里
-  // 解析，而不是等到拼 prompt 时。
-  const actionSet = config.sleepActionSet ?? "conflict";
-  const pairs = findPotentialConflicts(usableMemories, usableVectors, threshold);
-  if (actionSet === "full") {
-    // Issue #126 review（Copilot）：findPotentialConflicts 只看同类型，而 full 档的
-    // prompt 明确宣传 differentiate 用于互补型——互补本来就常跨类型（同一事实的项目侧
-    // 与偏好侧）。不补这一遍，那条路径永远选不到。merge 的类型守卫不受影响：它拦在
-    // validateDecisions 里（allowCrossTypeMerge），不在候选生成。
-    for (let i = 0; i < usableMemories.length; i++) {
-      for (let j = i + 1; j < usableMemories.length; j++) {
-        if (usableMemories[i].type === usableMemories[j].type) continue;
-        const sim = cosineSimilarity(usableVectors[i], usableVectors[j]);
-        if (sim > threshold) pairs.push({ a: usableMemories[i], b: usableMemories[j], similarity: sim });
-      }
-    }
-  }
-  if (pairs.length === 0) return { status: "skipped", reason: "no conflicts found" };
-
-  // Dedupe: each memory participates in at most one pair, highest similarity
-  // first — overlapping pairs would violate validateDecisions' "one claim".
-  pairs.sort((a, b) => b.similarity - a.similarity);
-  const used = new Set();
-  const selected = [];
-  for (const p of pairs) {
-    if (used.has(p.a.id) || used.has(p.b.id)) continue;
-    used.add(p.a.id);
-    used.add(p.b.id);
-    selected.push(p);
-  }
-
-  // v0.8.1（issue #170 第 2 步）：跨 scope 相似对分流。同一内容落在两个归属下
-  // （任一维标注键不等即算，NULL=未标注=全局也参与比较——「全局副本 + 专属副本
-  // 并存」正是收窄/提升裁决的典型候选）不能自动裁决：归档败者可能销毁该内容在
-  // 某个归属下的唯一副本，LLM 也无从替用户决定归属。一律停车到冲突待确认队列
-  // （专属 reason），同 scope 对维持既有路径（freeze 停车 / LLM 裁决）。
-  const isCrossScopePair = (p) =>
-    scopeKeyOf(p.a.agent_scope) !== scopeKeyOf(p.b.agent_scope)
-    || scopeKeyOf(p.a.workspace_scope) !== scopeKeyOf(p.b.workspace_scope);
-  const crossPairs = selected.filter(isCrossScopePair);
-  const samePairs = selected.filter((p) => !isCrossScopePair(p));
-  const parkPair = (p, reason) => {
-    try {
-      // 返回 undefined = 该对已被人工裁决且两侧内容未变（复核项 4）——不算停车。
-      return service.saveConflictPending({ run_id: runId, memory_a: p.a.id, memory_b: p.b.id, reason }) != null;
-    } catch (error) {
-      logger?.warn?.(`dsh-mneme sleep: failed to park conflict ${p.a.id}/${p.b.id}: ${String(error)}`);
-      return false;
-    }
-  };
-  const scopeReason = (p) => STR.scopeCandidateReason[language](p.similarity.toFixed(2));
-  const simReason = (p) => STR.similarityReason[language](p.similarity.toFixed(2));
-
-  // Freeze mode: park pairs for manual review, no LLM required. Cross-scope
-  // pairs carry the dedicated scope-candidate reason so the queue reads them
-  // as ownership decisions rather than plain duplicates.
-  if (config.conflictFreezeEnabled === true) {
-    let frozen = 0;
-    for (const p of selected) {
-      if (parkPair(p, isCrossScopePair(p) ? scopeReason(p) : simReason(p))) frozen++;
-    }
-    return { status: frozen > 0 ? "ok" : "noop", frozen, scopeCandidates: crossPairs.length, pairs: selected.length };
-  }
-
-  // 非 freeze：跨 scope 对先停车（绝不进 LLM 裁决），同 scope 对照常走 LLM。
-  let scopeParked = 0;
-  for (const p of crossPairs) {
-    if (parkPair(p, scopeReason(p))) scopeParked++;
-  }
-  if (samePairs.length === 0) {
-    return { status: scopeParked > 0 ? "ok" : "noop", frozen: 0, scopeCandidates: scopeParked, pairs: selected.length };
-  }
-
-  // LLM adjudication（仅同 scope 对）。
-  const route = resolveSleepRoute(ctx, config, logger);
-  if (!route) return { status: "skipped", reason: "no llm route", scopeCandidates: scopeParked, pairs: selected.length };
-  const snapshot = new Map();
-  for (const p of samePairs) {
-    snapshot.set(p.a.id, p.a);
-    snapshot.set(p.b.id, p.b);
-  }
-  const listText = samePairs.map((p) =>
-    STR.candidateConflicts[language](p)
-  ).join("\n\n");
-  const sleepEffort = await resolveDreamEffort(ctx, route, config.sleepReasoningEffort, logger);
-  let conflictStreamFailure = "";
-  // actionSet 已在候选构造前解析（跨类型候选需要它），这里只按档位选 prompt：
-  // 默认档保持只有 conflict/keep 的窄 prompt，"full" 才开放六分支。
-  const conflictPrompt = actionSet === "full" ? STR.prompts.conflictFull[language] : STR.prompts.conflict[language];
-  const runConflict = (withEffort) => {
-    conflictStreamFailure = "";
-    return auditedSleepLlm(ctx, service, config, logger, {
-      triggerSource: "sleep",
-      operationType: "sleep_conflict",
-      modelId: `${route.provider}:${route.model}`,
-      // 候选对的双方 id：面板的活动流用 related_memory_ids 解析「沉淀条数」。
-      relatedMemoryIds: samePairs.flatMap((p) => [p.a.id, p.b.id]),
-      streamError: () => conflictStreamFailure,
-      // 纯判据：输出里没有 JSON 数组 ⇒ 这轮是失败的调用，审计记 error。**不要**把
-      // 解析结果存进闭包给下面的控制流用——审计关闭时 auditedSleepLlm 整段早退、
-      // 这个钩子不会被调用，控制流会跟着失效（dream 的既有坑）。代价是审计开启时
-      // 多解析一次，换来两条路径彻底解耦。
-      auditError: (text) => (parseJsonArray(text) ? null : "no json array in llm output")
-    }, (reportUsage) => streamText(ctx, {
-    provider: route.provider,
-    model: route.model,
-    purpose: "sleep-conflict",
-    maxTokens: config.sleepMaxTokens ?? 2048,
-    ...(withEffort && sleepEffort ? { reasoningEffort: sleepEffort } : {}),
-    messages: [
-      { role: "system", content: [{ type: "text", text: conflictPrompt }], source: { kind: "plugin:dsh-mneme", plugin: "dsh-mneme" } },
-      { role: "user", content: [{ type: "text", text: listText }], source: { kind: "plugin:dsh-mneme", plugin: "dsh-mneme" } }
-    ]
-  }, reportUsage, (reason) => { conflictStreamFailure = describeStreamFailure(reason); }));
-  };
-  const text = await withEffortFallback(ctx, sleepEffort, () => runConflict(true), () => runConflict(false), () => conflictStreamFailure);
-  if (text === undefined) {
-    if (conflictStreamFailure) ctx.logger?.warn?.(`dsh-mneme sleep: conflict stream aborted or errored (${conflictStreamFailure})`);
-    return { status: "failed", error: "llm failed" };
-  }
-  const decisions = parseJsonArray(text);
-  if (!decisions) return { status: "failed", error: "invalid decisions json" };
-  // validateDecisions 要求每个 snapshot id 恰好被 claim 一次。v0.4.4 起它本身
-  // 就会为未覆盖的 id 自动补 keep（dreamImplicitKeep 默认开启），这里保留显式
-  // 预填作为防御性双保险——漏判读作"未裁决冲突"而非"冲突阶段整体失败"。
-  const covered = new Set();
-  for (const d of decisions) {
-    // Issue #126 review（Copilot）：supersede 同样用 winner/loser 定位（没有 ids），
-    // 此前只认 conflict → 它的两个目标都没进 covered → 被补成 implicit keep → 与
-    // supersede 的 claim 冲突：严格模式整单拒绝，宽容模式把合成的 keep 记成 skipped。
-    if (d?.action === "conflict" || d?.action === "supersede") {
-      if (typeof d?.winner === "string") covered.add(d.winner);
-      if (typeof d?.loser === "string") covered.add(d.loser);
-    } else if (Array.isArray(d?.ids)) {
-      for (const id of d.ids) covered.add(id);
-    }
-  }
-  for (const id of snapshot.keys()) {
-    if (!covered.has(id)) decisions.push({ action: "keep", ids: [id] });
-  }
-  // Issue #126：sleep 是多对批量裁决，严格模式"一票否决"太脆（一对非法就丢掉整轮
-  // 成果）→ 与 dream 对齐透传 skipInvalid：非法的那一对只跳过自己，合法子集照常
-  // 应用，run 记 degraded。update 上限与保护期同样取 dream 的同一组配置，保证两个
-  // 模块对"什么算合法"的判据一致。
-  const { ok, errors, skipped } = validateDecisions(decisions, snapshot, {
-    maxUpdatePerRun: config.reflectionUpdateMaxPerRun,
-    minAgeHours: config.reflectionUpdateMinAgeHours,
-    skipInvalid: config.dreamSkipInvalid !== false,
-    allowCrossTypeMerge: config.allowCrossTypeMerge === true,
-    // issue #339 / E8：guarded 类型 merge 护栏（opt-in，dreamMergeGuard），
-    // 与 dream 主链路同一判据。
-    mergeGuard: config.dreamMergeGuard === true,
-    // Issue #126 review（Copilot）：默认档必须真的只放行旧动作集——只换 prompt 挡不住
-    // 模型自发输出 supersede / differentiate，那样"opt-in 零行为变化"就不成立。
-    allowedActions: actionSet === "full" ? null : LEGACY_ACTIONS
-  });
-  if (!ok) return { status: "failed", error: `invalid decisions: ${errors.join("; ")}`, skipped };
-  const { applied, failures, conflicts } = applyDecisions(decisions, service, logger, snapshot, config);
-  // Issue #126 review（Copilot）：事务内 service.update 会跳过 scheduleEmbed（txDepth>0），
-  // 而 sleep 路径此前没有 dream 那样的 post-apply 索引维护——differentiate 的差异注记与
-  // supersede 的归档都反映不到缓存向量，"注记进 embedding 防再判重复"就是空承诺。
-  if (applied > 0 && semantic?.embedder && semantic?.vectorIndex) {
-    try {
-      await maintainIndexAfterDream(decisions, service, semantic);
-    } catch (error) {
-      logger?.warn?.(`dsh-mneme sleep: index maintenance failed: ${String(error)}`);
-    }
-  }
-  // Issue #126 review（Copilot）：有决策被跳过时不能报 ok——与 dream 的 degraded 契约
-  // 对齐（"合法子集已应用"是 degraded 而非 ok），否则消费方分不清整轮与残轮。
-  const degraded = skipped.length > 0;
-  return {
-    status: applied > 0 ? (degraded ? "degraded" : "ok") : failures.length ? "failed" : "noop",
-    pairs: selected.length,
-    applied,
-    failures,
-    conflicts,
-    // Issue #126：#104 同向——被跳过的逐对明细要进审计行，否则事后无法判断
-    // 是哪一对、因何被跳过（此前只进 logger.warn，splice 后就地消失）。
-    skipped
-  };
-}
-
-/**
- * Phase 2 — archival demotion. No LLM: tiering is time-based, summaries are
- * truncations, and the full body is preserved in _full_content so nothing is
- * lost. Deterministic and cheap, so it runs even with no LLM route.
- */
-function phaseDemotion(service, config, logger, runId, signal = null) {
-  const archiveDays = config.sleepArchiveDays ?? 30;
-  const compressDays = config.sleepCompressDays ?? 90;
-  const archiveCut = Date.now() - archiveDays * 86400000;
-  const compressCut = Date.now() - compressDays * 86400000;
-  const demoted = [];
-  const archived = [];
-  for (const m of service.all()) {
-    if (signal?.aborted) break;
-    if (m.archived || m.forgotten) continue;
-    const ref = m.last_accessed_at ?? m.created_at;
-    if (!ref) continue;
-    const t = new Date(ref).getTime();
-    if (Number.isNaN(t)) continue;
-    // v0.7.0 热联合判定（仅 heatEnabled 时启用；默认关则退回纯时间分层，
-    // 与 v0.7.12 行为一致）：时间窗之外再加两道保护闸——热度低于
-    // sleepHeatThreshold 且 importance<5 才允许降级。λ=0 的免疫类型 heat 恒
-    // 1.0 天然豁免（preference/pattern/summary 永不因 sleep 降级）；importance
-    // ≥5 的紧要记忆无论多冷都保留。`冷但重要` 与 `热但低值` 均不满足条件。
-    const heatOn = config.heatEnabled !== false;
-    if (heatOn) {
-      const heat = computeHeat(m, Date.now(), config);
-      const heatProtected = heat >= (config.sleepHeatThreshold ?? 0.05);
-      const important = (m.importance ?? 0) >= 5;
-      if (heatProtected || important) continue;
-    }
-    if (t < compressCut) {
-      service.setArchived(m.id, true);
-      archived.push(m.id);
-    } else if (t < archiveCut) {
-      // minRefTimeMs re-checks freshness inside demoteToSummary's transaction:
-      // a recall touch landing after this snapshot must not demote the memory.
-      service.demoteToSummary(m.id, makeSummary(m), { minRefTimeMs: archiveCut });
-      demoted.push(m.id);
-    }
-  }
-  return {
-    status: demoted.length || archived.length ? "ok" : "noop",
-    demoted,
-    archived
-  };
-}
-
-/**
- * Phase 3 — pattern discovery. The LLM scans the most recent memories and
- * mints type=pattern entries (create actions) with evidence references.
- * The empty snapshot is intentional: create claims no existing id, so the
- * "every id claimed" invariant is trivially satisfied for pure-create lists.
- */
-async function phasePatterns(ctx, service, config, logger, runId, signal = null) {
-  const language = langOf(config);
-  const route = resolveSleepRoute(ctx, config, logger);
-  if (!route) return { status: "skipped", reason: "no llm route" };
-  const limit = config.sleepPatternMinMemories ?? 100;
-  // #230：document 在查询层排除（excludeTypes）而不是截断后过滤——LIMIT 200
-  // 先生效的话，指针行一多就会把普通记忆挤出扫描窗，池子饿到门槛以下。
-  const memories = service
-    .list({ limit: 200, includeForgotten: false, excludeTypes: ["document"] })
-    .filter((m) => !m.archived && m.type !== "summary" && m.type !== "pattern")
-    .sort((a, b) => (a.updated_at < b.updated_at ? 1 : -1))
-    .slice(0, limit);
-  if (memories.length === 0) return { status: "skipped", reason: "no memories to scan" };
-  if (signal?.aborted) return { status: "aborted", reason: "user activity" };
-  const listText = memories
-    .map((m) => `id=${m.id} | type=${m.type} | importance=${m.importance} | updated=${m.updated_at} | title=${m.title} | content=${m.content}`)
-    .join("\n");
-  const maxPatterns = config.sleepMaxPatternPerRun ?? 3;
-  const sleepEffort = await resolveDreamEffort(ctx, route, config.sleepReasoningEffort, logger);
-  let patternStreamFailure = "";
-  const runPattern = (withEffort) => {
-    patternStreamFailure = "";
-    return auditedSleepLlm(ctx, service, config, logger, {
-      triggerSource: "sleep",
-      operationType: "sleep_pattern",
-      modelId: `${route.provider}:${route.model}`,
-      // 本轮扫描的记忆 id：与 dream 的 relatedMemoryIds 同义，供面板解析沉淀条数。
-      relatedMemoryIds: memories.map((m) => m.id),
-      streamError: () => patternStreamFailure,
-      // 纯判据，同 conflict 阶段：不把解析结果存进闭包供控制流使用。
-      auditError: (text) => (parseJsonArray(text) ? null : "no json array in llm output")
-    }, (reportUsage) => streamText(ctx, {
-    provider: route.provider,
-    model: route.model,
-    purpose: "sleep-pattern",
-    maxTokens: config.sleepMaxTokens ?? 2048,
-    ...(withEffort && sleepEffort ? { reasoningEffort: sleepEffort } : {}),
-    messages: [
-      { role: "system", content: [{ type: "text", text: STR.prompts.pattern[language].replace("N", String(maxPatterns)) }], source: { kind: "plugin:dsh-mneme", plugin: "dsh-mneme" } },
-      { role: "user", content: [{ type: "text", text: listText }], source: { kind: "plugin:dsh-mneme", plugin: "dsh-mneme" } }
-    ]
-  }, reportUsage, (reason) => { patternStreamFailure = describeStreamFailure(reason); }));
-  };
-  const text = await withEffortFallback(ctx, sleepEffort, () => runPattern(true), () => runPattern(false), () => patternStreamFailure);
-  if (text === undefined) {
-    if (patternStreamFailure) ctx.logger?.warn?.(`dsh-mneme sleep: pattern stream aborted or errored (${patternStreamFailure})`);
-    return { status: "failed", error: "llm failed" };
-  }
-  const decisions = parseJsonArray(text);
-  if (!decisions || decisions.length === 0) return { status: "skipped", reason: "no patterns found" };
-  // Evidence ids are provenance refs; an LLM-fabricated id would mint a dead
-  // ev:tag pointing nowhere. Intersect evidence with the scanned set so only
-  // real memory references survive.
-  const scannedIds = new Set(memories.map((m) => m.id));
-  for (const d of decisions) {
-    if (d?.action === "create" && Array.isArray(d.evidence)) {
-      d.evidence = d.evidence.filter((id) => typeof id === "string" && scannedIds.has(id));
-    }
-  }
-  const snapshot = new Map();
-  const { ok, errors } = validateDecisions(decisions, snapshot, {
-    maxCreatePerRun: maxPatterns
-  });
-  if (!ok) return { status: "failed", error: `invalid decisions: ${errors.join("; ")}` };
-  const { applied, failures, conflicts } = applyDecisions(decisions, service, logger, snapshot, config);
-  return {
-    status: applied > 0 ? "ok" : "noop",
-    scanned: memories.length,
-    applied,
-    failures,
-    conflicts
-  };
-}
-
-/**
- * Phase 4 — entity relation completion. Detects orphan entities (zero
- * relations) and completes implied relations from memory co-occurrence:
- * entities named in the same memory → related_to; container kinds
- * (project/module) → part_of; tech-ish pairs → depends_on. Deterministic,
- * no LLM — cheap, so it runs even without a route. saveRelation is
- * bookkeeping (no write hook), so it never re-triggers the scheduler.
- */
-function inferRelationType(a, b) {
-  if ((a.type === "project" || a.type === "module") && a.type !== b.type) return "part_of";
-  if ((b.type === "project" || b.type === "module") && b.type !== a.type) return "part_of";
-  if (/npm|plugin|api|sdk|lib|framework|package|deps?|build/i.test(`${a.name} ${b.name}`)) return "depends_on";
-  return "related_to";
-}
-
-function phaseRelations(service, config, logger, runId, signal = null) {
-  const entities = service.listEntities({ limit: 1000 }) ?? [];
-  if (entities.length < 2) return { status: "skipped", reason: "too few entities" };
-  const orphans = entities.filter((e) => (service.getRelations(e.id) ?? []).length === 0);
-  if (orphans.length === 0) return { status: "skipped", reason: "no orphan entities" };
-  const memories = service.all().filter((m) => !m.archived && !m.forgotten);
-  const seen = new Set();
-  const related = [];
-  const MAX_RELATIONS_PER_ORPHAN = 3;
-  for (const o of orphans) {
-    if (signal?.aborted) break;
-    let made = 0;
-    for (const m of memories) {
-      if (signal?.aborted || made >= MAX_RELATIONS_PER_ORPHAN) break;
-      const text = `${m.title ?? ""} ${m.content ?? ""}`;
-      if (!text.includes(o.name)) continue;
-      for (const other of entities) {
-        if (other.id === o.id || other.name === o.name) continue;
-        const key = [o.id, other.id].sort().join("|");
-        if (seen.has(key)) continue;
-        if (!text.includes(other.name)) continue;
-        const relationType = inferRelationType(o, other);
-        try {
-          service.saveRelation({ from_entity: o.id, to_entity: other.id, relation_type: relationType, memory_id: m.id, metadata: { source: "sleep_relation_completion" } });
-          seen.add(key);
-          related.push({ from: o.id, to: other.id, type: relationType });
-          made++;
-        } catch (error) {
-          logger?.warn?.(`dsh-mneme sleep: relation ${o.id}/${other.id} failed: ${String(error)}`);
-        }
-      }
-    }
-  }
-  return {
-    status: related.length > 0 ? "ok" : "noop",
-    orphanCount: orphans.length,
-    related
-  };
-}
-
-// ---------------------------------------------------------------- run
-
-function deriveStatus(phases) {
-  const list = Object.values(phases);
-  if (list.length === 0) return "noop";
-  const anyError = list.some((p) => p.status === "failed" || p.status === "error");
-  // Issue #126 review（Copilot）：phase 自身可以是 degraded（合法子集已应用但有条目被
-  // 跳过），它同样算"有产出"——否则整轮会被误判成 noop，把真实变更报成空轮。
-  const anyWork = list.some((p) => p.status === "ok" || p.status === "degraded");
-  const anyDegraded = list.some((p) => p.status === "degraded");
-  if (anyDegraded) return "degraded";
-  if (anyWork && anyError) return "degraded";
-  if (anyError) return "failed";
-  if (anyWork) return "ok";
-  return "noop";
-}
-
-/**
- * Run one full sleep cycle. Best-effort across all phases; writes a
- * run_type='sleep' audit row (same dream_runs table) so sleep activity is
- * observable alongside consolidation runs.
- */
-export async function runSleep(ctx, service, config, logger, semantic = null, signal = null) {
-  const runId = randomUUID();
-  // Issue #89：开跑时刻——审计行 created_at 的基准（= 冷却种子），run 耗时
-  // （LLM 各 phase）不应计入重启后的冷却窗口。
-  const sleepStartedAt = Date.now();
-  const phases = {};
-  const attempt = async (name, fn) => {
-    if (signal?.aborted) return; // user resumed activity — stop before next phase
-    try {
-      phases[name] = await fn();
-    } catch (error) {
-      phases[name] = { status: "failed", error: error?.message ?? String(error) };
-      logger?.warn?.(`dsh-mneme sleep: ${name} phase failed: ${error?.message ?? error}`);
-    }
-  };
-  await attempt("conflicts", () => phaseConflicts(ctx, service, config, logger, runId, semantic, signal));
-  await attempt("demotion", () => phaseDemotion(service, config, logger, runId, signal));
-  await attempt("patterns", () => phasePatterns(ctx, service, config, logger, runId, signal));
-  await attempt("relations", () => phaseRelations(service, config, logger, runId, signal));
-
-  const status = deriveStatus(phases);
-  const route = resolveSleepRoute(ctx, config, logger);
-  const totalApplied = Object.values(phases).reduce((n, p) => n + (Number.isInteger(p?.applied) ? p.applied : 0), 0);
-  // Issue #126（与 #104 同向）：把各 phase 的跳过明细汇总成一行审计数据，带 phase
-  // 名（同一次 run 里 conflicts 与 patterns 都可能产出 skipped）。全空时写 NULL，
-  // 不落空数组——审计行只记真实发生过的跳过。
-  const skippedDetail = Object.entries(phases).flatMap(([phase, p]) =>
-    Array.isArray(p?.skipped) ? p.skipped.map((s) => ({ phase, ...s })) : []
-  );
-  const snapshotHash = createHash("sha256").update(JSON.stringify(phases)).digest("hex");
-  const receipt = buildReceipt({
-    runId,
-    status,
-    snapshotHash,
-    inputCount: 0,
-    applied: totalApplied,
-    summaryStored: false
-  });
-  try {
-    service.saveDreamRun({
-      id: runId,
-      // Issue #89：审计行记开跑时刻而非完成时刻——run 耗时（含 LLM 各 phase）
-      // 不该计入重启后的 sleep 冷却种子。
-      created_at: new Date(sleepStartedAt).toISOString(),
-      status,
-      provider: route?.provider,
-      model: route?.model,
-      snapshot_hash: snapshotHash,
-      input_count: 0,
-      input: null,
-      decisions: phases,
-      outcome: phases,
-      applied: totalApplied,
-      summary_stored: false,
-      receipt,
-      policy_epoch: config.policyEpoch ?? 0,
-      run_type: "sleep",
-      skipped: skippedDetail.length ? skippedDetail : undefined
-    });
-  } catch (error) {
-    logger?.warn?.(`dsh-mneme sleep: failed to record audit run: ${String(error)}`);
-  }
-  return { ok: status === "ok" || status === "degraded", status, runId, phases, receipt };
-}
-
-// ---------------------------------------------------------------- scheduler
-
-/**
- * Idle-triggered scheduler. DSH plugins have no resident cron, so a sleep run
- * fires when: sleep is enabled, the store has been quiet for sleepIdleMinutes,
- * and the previous run is older than sleepMinIntervalHours. noteWrite() is
- * called on every store write and (re)arms an idle timer that re-checks at the
- * exact moment the idle window elapses — no polling, no cron.
- *
- * A `now` clock can be injected for tests; it defaults to Date.now.
- */
-export function createSleepScheduler({
-  service,
-  config,
-  logger,
-  onRun = null,
-  now = () => Date.now(),
-  setTimeoutFn = setTimeout,
-  clearTimeoutFn = clearTimeout,
-  // Issue #89 同族：冷却时刻由调用方从 dream_runs 审计表恢复（run_type='sleep'）——
-  // 内存变量进程重启即归零，冷却闸对新实例放行 → 重启后立即进入 sleep 冷却盲区。
-  // 测试注入 0（默认）即保持旧行为。
-  lastRunAtSeed = 0
-}) {
-  let lastWriteAt = now();
-  let lastRunAt = lastRunAtSeed;
-  let running = false;
-  let disposed = false;
-  let idleTimer = null;
-  let sleepAbort = null;
-
-  function armIdleTimer() {
-    if (disposed || idleTimer) return;
-    if (config.sleepModeEnabled !== true) return;
-    const idleMs = (config.sleepIdleMinutes ?? 5) * 60000;
-    const delay = Math.max(0, idleMs - (now() - lastWriteAt)) + 1000;
-    idleTimer = setTimeoutFn(async () => {
-      idleTimer = null;
-      // Issue #187：false = 撞上 CD（或 run 失败）——此前这里直接丢弃，"CD 到期"
-      // 成了没有监听者的时刻，要等下一次写入才救回来。改为按剩余窗口重排。
-      const ran = await maybeSchedule();
-      if (!ran) armNextWindow();
-    }, delay);
-    idleTimer.unref?.();
-  }
-
-  // Issue #187：撞 CD / run 失败后的补挂表——按「剩余 CD / 剩余静默」里更晚的
-  // 到点时刻重排一次，回调里重复同样的判断所以不会空转。跑成了一轮就不重排：
-  // 安静的库没有要处理的新东西，触发权交还给下一次写入的 noteWrite。
-  function armNextWindow() {
-    if (disposed || idleTimer) return;
-    if (config.sleepModeEnabled !== true) return;
-    // lastRunAt === 0 表示从未跑过（此路径不该出现，防御性返回）。
-    if (lastRunAt <= 0) return;
-    const idleMs = (config.sleepIdleMinutes ?? 5) * 60000;
-    const cdMs = (config.sleepMinIntervalHours ?? 8) * 3600000;
-    const delay = Math.max(
-      lastRunAt + cdMs - now(),
-      lastWriteAt + idleMs - now(),
-      0
-    ) + 1000;
-    idleTimer = setTimeoutFn(async () => {
-      idleTimer = null;
-      const ran = await maybeSchedule();
-      if (!ran) armNextWindow();
-    }, delay);
-    idleTimer.unref?.();
-  }
-
-  function shouldRun(at = now()) {
-    if (disposed || running) return false;
-    if (config.sleepModeEnabled !== true) return false;
-    if (at - lastWriteAt < (config.sleepIdleMinutes ?? 5) * 60000) return false;
-    // lastRunAt === 0 means never ran — the min-interval check must not block
-    // the very first cycle (a real run stamps a nonzero timestamp).
-    if (lastRunAt > 0 && at - lastRunAt < (config.sleepMinIntervalHours ?? 8) * 3600000) return false;
-    return true;
-  }
-
-  /** Called on writes: resets the idle clock and re-arms the fire timer. The
-   *  pending timer is cleared first — a stale timer armed against the old idle
-   *  window would otherwise fire early, fail shouldRun, and leave nothing armed
-   *  for the next window (a missed trigger until the next write).
-   *
-   *  While a sleep run is executing (running=true) the in-flight AbortController
-   *  is NOT aborted: the run's own writes (demoteToSummary / setArchived ride
-   *  the normal write-hook path) would otherwise self-abort the cycle. External
-   *  activity during the run still resets the idle clock here, so no new cycle
-   *  fires until the store is quiet again. */
-  function noteWrite() {
-    lastWriteAt = now();
-    if (!running && sleepAbort) {
-      sleepAbort.abort(); // user resumed activity — interrupt an idle run
-    }
-    if (idleTimer) {
-      clearTimeoutFn(idleTimer);
-      idleTimer = null;
-    }
-    armIdleTimer();
-  }
-
-  async function maybeSchedule() {
-    if (!shouldRun()) return false;
-    running = true;
-    const abort = new AbortController();
-    sleepAbort = abort;
-    try {
-      lastRunAt = now();
-      const result = await service.enqueue(() =>
-        onRun ? onRun(abort.signal) : Promise.resolve({ ok: true, skipped: true })
-      );
-      return !!(result && result.ok);
-    } catch (error) {
-      logger?.warn?.(`dsh-mneme sleep: run failed: ${error?.message ?? error}`);
-      return false;
-    } finally {
-      sleepAbort = null;
-      running = false;
-    }
-  }
-
-  async function dispose() {
-    disposed = true;
-    if (idleTimer) {
-      clearTimeoutFn(idleTimer);
-      idleTimer = null;
-    }
-    if (sleepAbort) {
-      sleepAbort.abort();
-      sleepAbort = null;
-    }
-  }
-
-  // Issue #187（同族缺陷）：构造即挂表——重启后只要没发生过写入，此前永远没有
-  // 闹钟，安静多久都不会触发。lastWriteAt 已在构造时设为当前时刻，shouldRun 的
-  // 静默条件自然生效：重启后静默满 idleMinutes 才可能起跑。
-  armIdleTimer();
-
-  return { noteWrite, maybeSchedule, shouldRun, dispose };
-}
+import { STR, langOf } from "../lang.js";
+// System-level sleep (v0.4.0): an idle-triggered, LLM-assisted deep pass over
+// the memory store. Four independent, fail-safe phases:
+//   1. conflict resolution — high-similarity same-type pairs are either parked
+//      for review (freeze mode) or adjudicated by the LLM (winner kept / loser
+//      archived), reusing the dream conflict machinery. Strictness-graded.
+//      v0.8.1 (issue #170): cross-scope pairs are ALWAYS parked for human
+//      ownership review with a dedicated reason — never auto-adjudicated.
+//   2. archival demotion — memories unreferenced past sleepArchiveDays shrink
+//      to a one-line summary with the full body moved to _full_content; past
+//      sleepCompressDays they are archived outright.
+//   3. pattern discovery — the LLM scans the most recent memories and mints
+//      type=pattern entries carrying evidence id references.
+//   4. relation completion — orphan entities (zero relations) get implied
+//      relations completed from memory co-occurrence.
+// Each phase is wrapped so one failure never aborts the others, and a missing
+// LLM route / semantic embedder only skips the phases that need it. A run is
+// abortable via an AbortController signal (user activity) — phases check the
+// signal between batches so a running cycle yields promptly.
+import { randomUUID, createHash } from "node:crypto";
+import { validateDecisions, applyDecisions, ACTIONS } from "./decisions.js";
+import { findPotentialConflicts, cosineSimilarity } from "./clustering.js";
+import { scopeKeyOf } from "../scope.js";
+import { buildReceipt, describeStreamFailure, resolveDreamEffort, withEffortFallback, maintainIndexAfterDream } from "../dream.js";
+import { computeHeat } from "../heat.js";
+
+// Issue #126 review（Copilot）：默认档（sleepActionSet="conflict"）允许的动作 = 全局
+// ACTIONS 去掉 #126 新增的两个。从 ACTIONS 派生而非硬编码，将来再扩动作集时自动跟随。
+// 只换 prompt 挡不住模型自发输出新动作，必须由校验器按档位收口。
+const LEGACY_ACTIONS = [...ACTIONS].filter((a) => a !== "supersede" && a !== "differentiate");
+
+const SUMMARY_MAX = 120;
+// Conflict similarity threshold per strictness level (v0.4.0):
+//   gentle     only high-confidence pairs (0.92) — first-time users
+//   normal     standard dream-level (0.85) — default
+//   aggressive low-confidence pairs too (0.75) — bloated stores
+const CONFLICT_THRESHOLDS = { gentle: 0.92, normal: 0.85, aggressive: 0.75 };
+
+function parseJsonArray(text) {
+  if (typeof text !== "string") return undefined;
+  const start = text.indexOf("[");
+  const end = text.lastIndexOf("]");
+  if (start === -1 || end <= start) return undefined;
+  try {
+    const parsed = JSON.parse(text.slice(start, end + 1));
+    return Array.isArray(parsed) ? parsed : undefined;
+  } catch {
+    return undefined;
+  }
+}
+
+/**
+ * Same stream consumption contract as dream.js.
+ *
+ * `onUsage`（optional, issue #250）receives any usage chunk for token accounting.
+ * dream.js 那份一直有它；sleep 这份副本没有，导致 conflict / pattern 两条链路的
+ * token 从未进 llm_audit_logs（面板因此只看到 autoDream 与 autoSummarize）。
+ */
+async function streamText(ctx, options, onUsage, onStreamError) {
+  if (!ctx?.llm?.stream) return undefined;
+  let text = "";
+  for await (const chunk of ctx.llm.stream(options)) {
+    if (chunk.type === "text-delta" && typeof chunk.text === "string") text += chunk.text;
+    // DSH 的 StreamChunk 契约把用量嵌在 chunk.usage（TokenUsage：inputTokens /
+    // outputTokens / totalTokens）；兼容直接平铺在 chunk 上的旧形态。归一放在
+    // 这里，调用侧的 reporter 只需认平铺形状——与 dream.js:239 同一处理。
+    if (chunk.type === "usage" && typeof onUsage === "function") onUsage(chunk.usage ?? chunk);
+    if (chunk.type === "finish" && (chunk.reason?.kind === "error" || chunk.reason?.kind === "aborted")) {
+      // Same rc.1 error-as-finish-chunk behavior as dream.js — surface the
+      // cause instead of discarding it.
+      if (typeof onStreamError === "function") {
+        try { onStreamError(chunk.reason); } catch { /* diagnostics only */ }
+      }
+      return undefined;
+    }
+  }
+  return text;
+}
+
+/**
+ * Issue #250：给 sleep 的一条 LLM 调用记一笔 llm_audit_logs（token / 时长 / 状态 /
+ * 触发源）。与 dream.js 的 runAuditedLlm 同口径但各自独立一份——dream 那两个
+ * checker 的调用形状绑定着它自己的排程细节，硬抽共享抽象只会收敛回 dream 的形态。
+ *
+ * 记账是 best-effort：写审计行失败只 warn，绝不反噬 sleep 本体（CONTRIBUTING 的
+ * fail-safe 硬约定）。`body(reportUsage)` 负责真实消费流，usage chunk 经
+ * reportUsage 上报。
+ *
+ * 审计诚实性：流被中止/报错时 body 返回 undefined，记 status='error'；流式成功但
+ * 输出不可用时 `spec.auditError` 给出原因，同样记 error——否则同一轮调用会在
+ * 「phase 报 failed」与「audit 报 success」之间自相矛盾（CONTRIBUTING 的审计诚实性
+ * 硬约定）。`auditError` 必须是**纯判据函数**：调用方的控制流不能依赖它的副作用，
+ * 因为审计关闭时这里整段早退、钩子根本不会被调用（dream 正是那样踩的，见 PR 正文）。
+ */
+async function auditedSleepLlm(ctx, service, config, logger, spec, body) {
+  if (config?.llmAudit?.enabled === false || typeof service?.saveLlmAudit !== "function") {
+    return body(() => {});
+  }
+  const startedAt = Date.now();
+  const timestamp = new Date(startedAt).toISOString();
+  let inputTokens = 0;
+  let outputTokens = 0;
+  let status = "success";
+  let errorMessage = null;
+  try {
+    const result = await body((usage) => {
+      if (!usage) return;
+      const i = usage.input_tokens ?? usage.inputTokens ?? usage.prompt_tokens ?? usage.promptTokens;
+      const o = usage.output_tokens ?? usage.outputTokens ?? usage.completion_tokens ?? usage.completionTokens;
+      if (Number.isFinite(i)) inputTokens = i;
+      if (Number.isFinite(o)) outputTokens = o;
+    });
+    if (result === undefined) {
+      status = "error";
+      const streamErr = typeof spec.streamError === "function" ? String(spec.streamError() ?? "") : "";
+      errorMessage = streamErr ? `llm stream aborted or errored (${streamErr})` : "llm stream aborted or errored";
+    } else if (typeof spec.auditError === "function") {
+      // 流式成功但输出不可用，仍然是一次失败的调用——记 error，不与调用方的
+      // failed 返回值打架。
+      const message = spec.auditError(result);
+      if (message) {
+        status = "error";
+        errorMessage = message;
+      }
+    }
+    return result;
+  } catch (error) {
+    status = "error";
+    errorMessage = String(error?.message ?? error);
+    throw error;
+  } finally {
+    try {
+      service.saveLlmAudit({
+        timestamp,
+        trigger_source: spec.triggerSource,
+        operation_type: spec.operationType,
+        model_id: spec.modelId,
+        input_tokens: inputTokens,
+        output_tokens: outputTokens,
+        total_tokens: inputTokens + outputTokens,
+        cost_usd: 0,
+        duration_ms: Date.now() - startedAt,
+        status,
+        error_message: errorMessage,
+        related_memory_ids: spec.relatedMemoryIds ?? []
+      });
+    } catch (auditError) {
+      logger?.warn?.(`dsh-mneme sleep: llm audit write failed: ${String(auditError)}`);
+    }
+  }
+}
+
+/** LLM route (Issue #25): explicit sleepProvider/Model wins, then the dream
+ *  route as a shared explicit fallback, then the agent default model. Sleep
+ *  can pin a cheaper model for its bulk passes without disturbing the dream
+ *  route. Explicit config first — otherwise the config routes are dead code
+ *  whenever agentDefaultModel resolves (see resolveRoute in dream.js). */
+function resolveSleepRoute(ctx, config, logger) {
+  if (config.sleepProvider && config.sleepModel) return { provider: config.sleepProvider, model: config.sleepModel };
+  if (config.dreamProvider && config.dreamModel) return { provider: config.dreamProvider, model: config.dreamModel };
+  try {
+    const sel = ctx?.agentDefaultModel?.currentSelection?.();
+    if (sel?.provider && sel?.model) return { provider: sel.provider, model: sel.model };
+  } catch { /* fall through to warn */ }
+  logger?.warn?.("dsh-mneme sleep: no llm route available");
+  return undefined;
+}
+
+function makeSummary(m) {
+  const text = (m.content ?? "").trim();
+  if (!text) return (m.title ?? "").trim();
+  return text.length <= SUMMARY_MAX ? text : `${text.slice(0, SUMMARY_MAX)}…`;
+}
+
+// ---------------------------------------------------------------- phases
+
+/**
+ * Phase 1 — conflict resolution. Needs a semantic embedder + vector index.
+ * In freeze mode (conflictFreezeEnabled) conflicting pairs are parked in
+ * conflict_pending for human review (no LLM). Otherwise the LLM adjudicates:
+ * each pair → winner kept / loser archived. Returns a per-run summary.
+ */
+async function phaseConflicts(ctx, service, config, logger, runId, semantic = null, signal = null) {
+  const language = langOf(config);
+  const embedder = semantic?.embedder;
+  const vectorIndex = semantic?.vectorIndex;
+  if (!embedder || !vectorIndex || typeof embedder.embed !== "function") {
+    return { status: "skipped", reason: "no semantic embedder" };
+  }
+  const strictness = config.sleepConflictStrictness ?? "normal";
+  const threshold = CONFLICT_THRESHOLDS[strictness] ?? CONFLICT_THRESHOLDS.normal;
+  const memories = service.all().filter((m) => !m.archived && !m.forgotten && m.type !== "summary" && m.type !== "document");
+  if (memories.length < 2) return { status: "skipped", reason: "too few memories" };
+  if (signal?.aborted) return { status: "aborted", reason: "user activity" };
+
+  // Backfill + collect vectors for every eligible memory (best effort).
+  const vectors = new Array(memories.length);
+  const missing = [];
+  for (let i = 0; i < memories.length; i++) {
+    const cached = vectorIndex.getEmbedding?.(memories[i].id);
+    if (cached) vectors[i] = cached;
+    else missing.push(i);
+  }
+  if (missing.length) {
+    try {
+      const texts = missing.map((i) => [memories[i].title, memories[i].content].filter(Boolean).join("\n"));
+      const rows = await embedder.embed(texts);
+      missing.forEach((mi, j) => {
+        if (rows[j]?.length) {
+          vectors[mi] = rows[j];
+          vectorIndex.saveEmbedding?.(memories[mi].id, rows[j]);
+        }
+      });
+    } catch (error) {
+      logger?.warn?.(`dsh-mneme sleep: conflict vector backfill failed: ${String(error)}`);
+    }
+  }
+  const usable = [];
+  for (let i = 0; i < memories.length; i++) {
+    if (vectors[i]?.length) usable.push(i);
+  }
+  if (usable.length < 2) return { status: "skipped", reason: "no usable vectors" };
+  const usableMemories = usable.map((i) => memories[i]);
+  const usableVectors = usable.map((i) => vectors[i]);
+
+  // Issue #126：动作集档位既要决定 prompt，也要决定**校验器允许的动作集合**（见下方
+  // allowedActions），并且在候选构造阶段就要知道——full 档需要补跨类型对。因此在这里
+  // 解析，而不是等到拼 prompt 时。
+  const actionSet = config.sleepActionSet ?? "conflict";
+  const pairs = findPotentialConflicts(usableMemories, usableVectors, threshold);
+  if (actionSet === "full") {
+    // Issue #126 review（Copilot）：findPotentialConflicts 只看同类型，而 full 档的
+    // prompt 明确宣传 differentiate 用于互补型——互补本来就常跨类型（同一事实的项目侧
+    // 与偏好侧）。不补这一遍，那条路径永远选不到。merge 的类型守卫不受影响：它拦在
+    // validateDecisions 里（allowCrossTypeMerge），不在候选生成。
+    for (let i = 0; i < usableMemories.length; i++) {
+      for (let j = i + 1; j < usableMemories.length; j++) {
+        if (usableMemories[i].type === usableMemories[j].type) continue;
+        const sim = cosineSimilarity(usableVectors[i], usableVectors[j]);
+        if (sim > threshold) pairs.push({ a: usableMemories[i], b: usableMemories[j], similarity: sim });
+      }
+    }
+  }
+  if (pairs.length === 0) return { status: "skipped", reason: "no conflicts found" };
+
+  // Dedupe: each memory participates in at most one pair, highest similarity
+  // first — overlapping pairs would violate validateDecisions' "one claim".
+  pairs.sort((a, b) => b.similarity - a.similarity);
+  const used = new Set();
+  const selected = [];
+  for (const p of pairs) {
+    if (used.has(p.a.id) || used.has(p.b.id)) continue;
+    used.add(p.a.id);
+    used.add(p.b.id);
+    selected.push(p);
+  }
+
+  // v0.8.1（issue #170 第 2 步）：跨 scope 相似对分流。同一内容落在两个归属下
+  // （任一维标注键不等即算，NULL=未标注=全局也参与比较——「全局副本 + 专属副本
+  // 并存」正是收窄/提升裁决的典型候选）不能自动裁决：归档败者可能销毁该内容在
+  // 某个归属下的唯一副本，LLM 也无从替用户决定归属。一律停车到冲突待确认队列
+  // （专属 reason），同 scope 对维持既有路径（freeze 停车 / LLM 裁决）。
+  const isCrossScopePair = (p) =>
+    scopeKeyOf(p.a.agent_scope) !== scopeKeyOf(p.b.agent_scope)
+    || scopeKeyOf(p.a.workspace_scope) !== scopeKeyOf(p.b.workspace_scope);
+  const crossPairs = selected.filter(isCrossScopePair);
+  const samePairs = selected.filter((p) => !isCrossScopePair(p));
+  const parkPair = (p, reason) => {
+    try {
+      // 返回 undefined = 该对已被人工裁决且两侧内容未变（复核项 4）——不算停车。
+      return service.saveConflictPending({ run_id: runId, memory_a: p.a.id, memory_b: p.b.id, reason }) != null;
+    } catch (error) {
+      logger?.warn?.(`dsh-mneme sleep: failed to park conflict ${p.a.id}/${p.b.id}: ${String(error)}`);
+      return false;
+    }
+  };
+  const scopeReason = (p) => STR.scopeCandidateReason[language](p.similarity.toFixed(2));
+  const simReason = (p) => STR.similarityReason[language](p.similarity.toFixed(2));
+
+  // Freeze mode: park pairs for manual review, no LLM required. Cross-scope
+  // pairs carry the dedicated scope-candidate reason so the queue reads them
+  // as ownership decisions rather than plain duplicates.
+  if (config.conflictFreezeEnabled === true) {
+    let frozen = 0;
+    for (const p of selected) {
+      if (parkPair(p, isCrossScopePair(p) ? scopeReason(p) : simReason(p))) frozen++;
+    }
+    return { status: frozen > 0 ? "ok" : "noop", frozen, scopeCandidates: crossPairs.length, pairs: selected.length };
+  }
+
+  // 非 freeze：跨 scope 对先停车（绝不进 LLM 裁决），同 scope 对照常走 LLM。
+  let scopeParked = 0;
+  for (const p of crossPairs) {
+    if (parkPair(p, scopeReason(p))) scopeParked++;
+  }
+  if (samePairs.length === 0) {
+    return { status: scopeParked > 0 ? "ok" : "noop", frozen: 0, scopeCandidates: scopeParked, pairs: selected.length };
+  }
+
+  // LLM adjudication（仅同 scope 对）。
+  const route = resolveSleepRoute(ctx, config, logger);
+  if (!route) return { status: "skipped", reason: "no llm route", scopeCandidates: scopeParked, pairs: selected.length };
+  const snapshot = new Map();
+  for (const p of samePairs) {
+    snapshot.set(p.a.id, p.a);
+    snapshot.set(p.b.id, p.b);
+  }
+  const listText = samePairs.map((p) =>
+    STR.candidateConflicts[language](p)
+  ).join("\n\n");
+  const sleepEffort = await resolveDreamEffort(ctx, route, config.sleepReasoningEffort, logger);
+  let conflictStreamFailure = "";
+  // actionSet 已在候选构造前解析（跨类型候选需要它），这里只按档位选 prompt：
+  // 默认档保持只有 conflict/keep 的窄 prompt，"full" 才开放六分支。
+  const conflictPrompt = actionSet === "full" ? STR.prompts.conflictFull[language] : STR.prompts.conflict[language];
+  const runConflict = (withEffort) => {
+    conflictStreamFailure = "";
+    return auditedSleepLlm(ctx, service, config, logger, {
+      triggerSource: "sleep",
+      operationType: "sleep_conflict",
+      modelId: `${route.provider}:${route.model}`,
+      // 候选对的双方 id：面板的活动流用 related_memory_ids 解析「沉淀条数」。
+      relatedMemoryIds: samePairs.flatMap((p) => [p.a.id, p.b.id]),
+      streamError: () => conflictStreamFailure,
+      // 纯判据：输出里没有 JSON 数组 ⇒ 这轮是失败的调用，审计记 error。**不要**把
+      // 解析结果存进闭包给下面的控制流用——审计关闭时 auditedSleepLlm 整段早退、
+      // 这个钩子不会被调用，控制流会跟着失效（dream 的既有坑）。代价是审计开启时
+      // 多解析一次，换来两条路径彻底解耦。
+      auditError: (text) => (parseJsonArray(text) ? null : "no json array in llm output")
+    }, (reportUsage) => streamText(ctx, {
+    provider: route.provider,
+    model: route.model,
+    purpose: "sleep-conflict",
+    maxTokens: config.sleepMaxTokens ?? 2048,
+    ...(withEffort && sleepEffort ? { reasoningEffort: sleepEffort } : {}),
+    messages: [
+      { role: "system", content: [{ type: "text", text: conflictPrompt }], source: { kind: "plugin:dsh-mneme", plugin: "dsh-mneme" } },
+      { role: "user", content: [{ type: "text", text: listText }], source: { kind: "plugin:dsh-mneme", plugin: "dsh-mneme" } }
+    ]
+  }, reportUsage, (reason) => { conflictStreamFailure = describeStreamFailure(reason); }));
+  };
+  const text = await withEffortFallback(ctx, sleepEffort, () => runConflict(true), () => runConflict(false), () => conflictStreamFailure);
+  if (text === undefined) {
+    if (conflictStreamFailure) ctx.logger?.warn?.(`dsh-mneme sleep: conflict stream aborted or errored (${conflictStreamFailure})`);
+    return { status: "failed", error: "llm failed" };
+  }
+  const decisions = parseJsonArray(text);
+  if (!decisions) return { status: "failed", error: "invalid decisions json" };
+  // validateDecisions 要求每个 snapshot id 恰好被 claim 一次。v0.4.4 起它本身
+  // 就会为未覆盖的 id 自动补 keep（dreamImplicitKeep 默认开启），这里保留显式
+  // 预填作为防御性双保险——漏判读作"未裁决冲突"而非"冲突阶段整体失败"。
+  const covered = new Set();
+  for (const d of decisions) {
+    // Issue #126 review（Copilot）：supersede 同样用 winner/loser 定位（没有 ids），
+    // 此前只认 conflict → 它的两个目标都没进 covered → 被补成 implicit keep → 与
+    // supersede 的 claim 冲突：严格模式整单拒绝，宽容模式把合成的 keep 记成 skipped。
+    if (d?.action === "conflict" || d?.action === "supersede") {
+      if (typeof d?.winner === "string") covered.add(d.winner);
+      if (typeof d?.loser === "string") covered.add(d.loser);
+    } else if (Array.isArray(d?.ids)) {
+      for (const id of d.ids) covered.add(id);
+    }
+  }
+  for (const id of snapshot.keys()) {
+    if (!covered.has(id)) decisions.push({ action: "keep", ids: [id] });
+  }
+  // Issue #126：sleep 是多对批量裁决，严格模式"一票否决"太脆（一对非法就丢掉整轮
+  // 成果）→ 与 dream 对齐透传 skipInvalid：非法的那一对只跳过自己，合法子集照常
+  // 应用，run 记 degraded。update 上限与保护期同样取 dream 的同一组配置，保证两个
+  // 模块对"什么算合法"的判据一致。
+  const { ok, errors, skipped } = validateDecisions(decisions, snapshot, {
+    maxUpdatePerRun: config.reflectionUpdateMaxPerRun,
+    minAgeHours: config.reflectionUpdateMinAgeHours,
+    skipInvalid: config.dreamSkipInvalid !== false,
+    allowCrossTypeMerge: config.allowCrossTypeMerge === true,
+    // issue #339 / E8：guarded 类型 merge 护栏（默认开，dreamMergeGuard），
+    // 与 dream 主链路同一判据。
+    mergeGuard: config.dreamMergeGuard === true,
+    // Issue #126 review（Copilot）：默认档必须真的只放行旧动作集——只换 prompt 挡不住
+    // 模型自发输出 supersede / differentiate，那样"opt-in 零行为变化"就不成立。
+    allowedActions: actionSet === "full" ? null : LEGACY_ACTIONS
+  });
+  if (!ok) return { status: "failed", error: `invalid decisions: ${errors.join("; ")}`, skipped };
+  const { applied, failures, conflicts } = applyDecisions(decisions, service, logger, snapshot, config);
+  // Issue #126 review（Copilot）：事务内 service.update 会跳过 scheduleEmbed（txDepth>0），
+  // 而 sleep 路径此前没有 dream 那样的 post-apply 索引维护——differentiate 的差异注记与
+  // supersede 的归档都反映不到缓存向量，"注记进 embedding 防再判重复"就是空承诺。
+  if (applied > 0 && semantic?.embedder && semantic?.vectorIndex) {
+    try {
+      await maintainIndexAfterDream(decisions, service, semantic);
+    } catch (error) {
+      logger?.warn?.(`dsh-mneme sleep: index maintenance failed: ${String(error)}`);
+    }
+  }
+  // Issue #126 review（Copilot）：有决策被跳过时不能报 ok——与 dream 的 degraded 契约
+  // 对齐（"合法子集已应用"是 degraded 而非 ok），否则消费方分不清整轮与残轮。
+  const degraded = skipped.length > 0;
+  return {
+    status: applied > 0 ? (degraded ? "degraded" : "ok") : failures.length ? "failed" : "noop",
+    pairs: selected.length,
+    applied,
+    failures,
+    conflicts,
+    // Issue #126：#104 同向——被跳过的逐对明细要进审计行，否则事后无法判断
+    // 是哪一对、因何被跳过（此前只进 logger.warn，splice 后就地消失）。
+    skipped
+  };
+}
+
+/**
+ * Phase 2 — archival demotion. No LLM: tiering is time-based, summaries are
+ * truncations, and the full body is preserved in _full_content so nothing is
+ * lost. Deterministic and cheap, so it runs even with no LLM route.
+ */
+function phaseDemotion(service, config, logger, runId, signal = null) {
+  const archiveDays = config.sleepArchiveDays ?? 30;
+  const compressDays = config.sleepCompressDays ?? 90;
+  const archiveCut = Date.now() - archiveDays * 86400000;
+  const compressCut = Date.now() - compressDays * 86400000;
+  const demoted = [];
+  const archived = [];
+  for (const m of service.all()) {
+    if (signal?.aborted) break;
+    if (m.archived || m.forgotten) continue;
+    const ref = m.last_accessed_at ?? m.created_at;
+    if (!ref) continue;
+    const t = new Date(ref).getTime();
+    if (Number.isNaN(t)) continue;
+    // v0.7.0 热联合判定（仅 heatEnabled 时启用；默认关则退回纯时间分层，
+    // 与 v0.7.12 行为一致）：时间窗之外再加两道保护闸——热度低于
+    // sleepHeatThreshold 且 importance<5 才允许降级。λ=0 的免疫类型 heat 恒
+    // 1.0 天然豁免（preference/pattern/summary 永不因 sleep 降级）；importance
+    // ≥5 的紧要记忆无论多冷都保留。`冷但重要` 与 `热但低值` 均不满足条件。
+    const heatOn = config.heatEnabled !== false;
+    if (heatOn) {
+      const heat = computeHeat(m, Date.now(), config);
+      const heatProtected = heat >= (config.sleepHeatThreshold ?? 0.05);
+      const important = (m.importance ?? 0) >= 5;
+      if (heatProtected || important) continue;
+    }
+    if (t < compressCut) {
+      service.setArchived(m.id, true);
+      archived.push(m.id);
+    } else if (t < archiveCut) {
+      // minRefTimeMs re-checks freshness inside demoteToSummary's transaction:
+      // a recall touch landing after this snapshot must not demote the memory.
+      service.demoteToSummary(m.id, makeSummary(m), { minRefTimeMs: archiveCut });
+      demoted.push(m.id);
+    }
+  }
+  return {
+    status: demoted.length || archived.length ? "ok" : "noop",
+    demoted,
+    archived
+  };
+}
+
+/**
+ * Phase 3 — pattern discovery. The LLM scans the most recent memories and
+ * mints type=pattern entries (create actions) with evidence references.
+ * The empty snapshot is intentional: create claims no existing id, so the
+ * "every id claimed" invariant is trivially satisfied for pure-create lists.
+ */
+async function phasePatterns(ctx, service, config, logger, runId, signal = null) {
+  const language = langOf(config);
+  const route = resolveSleepRoute(ctx, config, logger);
+  if (!route) return { status: "skipped", reason: "no llm route" };
+  const limit = config.sleepPatternMinMemories ?? 100;
+  // #230：document 在查询层排除（excludeTypes）而不是截断后过滤——LIMIT 200
+  // 先生效的话，指针行一多就会把普通记忆挤出扫描窗，池子饿到门槛以下。
+  const memories = service
+    .list({ limit: 200, includeForgotten: false, excludeTypes: ["document"] })
+    .filter((m) => !m.archived && m.type !== "summary" && m.type !== "pattern")
+    .sort((a, b) => (a.updated_at < b.updated_at ? 1 : -1))
+    .slice(0, limit);
+  if (memories.length === 0) return { status: "skipped", reason: "no memories to scan" };
+  if (signal?.aborted) return { status: "aborted", reason: "user activity" };
+  const listText = memories
+    .map((m) => `id=${m.id} | type=${m.type} | importance=${m.importance} | updated=${m.updated_at} | title=${m.title} | content=${m.content}`)
+    .join("\n");
+  const maxPatterns = config.sleepMaxPatternPerRun ?? 3;
+  const sleepEffort = await resolveDreamEffort(ctx, route, config.sleepReasoningEffort, logger);
+  let patternStreamFailure = "";
+  const runPattern = (withEffort) => {
+    patternStreamFailure = "";
+    return auditedSleepLlm(ctx, service, config, logger, {
+      triggerSource: "sleep",
+      operationType: "sleep_pattern",
+      modelId: `${route.provider}:${route.model}`,
+      // 本轮扫描的记忆 id：与 dream 的 relatedMemoryIds 同义，供面板解析沉淀条数。
+      relatedMemoryIds: memories.map((m) => m.id),
+      streamError: () => patternStreamFailure,
+      // 纯判据，同 conflict 阶段：不把解析结果存进闭包供控制流使用。
+      auditError: (text) => (parseJsonArray(text) ? null : "no json array in llm output")
+    }, (reportUsage) => streamText(ctx, {
+    provider: route.provider,
+    model: route.model,
+    purpose: "sleep-pattern",
+    maxTokens: config.sleepMaxTokens ?? 2048,
+    ...(withEffort && sleepEffort ? { reasoningEffort: sleepEffort } : {}),
+    messages: [
+      { role: "system", content: [{ type: "text", text: STR.prompts.pattern[language].replace("N", String(maxPatterns)) }], source: { kind: "plugin:dsh-mneme", plugin: "dsh-mneme" } },
+      { role: "user", content: [{ type: "text", text: listText }], source: { kind: "plugin:dsh-mneme", plugin: "dsh-mneme" } }
+    ]
+  }, reportUsage, (reason) => { patternStreamFailure = describeStreamFailure(reason); }));
+  };
+  const text = await withEffortFallback(ctx, sleepEffort, () => runPattern(true), () => runPattern(false), () => patternStreamFailure);
+  if (text === undefined) {
+    if (patternStreamFailure) ctx.logger?.warn?.(`dsh-mneme sleep: pattern stream aborted or errored (${patternStreamFailure})`);
+    return { status: "failed", error: "llm failed" };
+  }
+  const decisions = parseJsonArray(text);
+  if (!decisions || decisions.length === 0) return { status: "skipped", reason: "no patterns found" };
+  // Evidence ids are provenance refs; an LLM-fabricated id would mint a dead
+  // ev:tag pointing nowhere. Intersect evidence with the scanned set so only
+  // real memory references survive.
+  const scannedIds = new Set(memories.map((m) => m.id));
+  for (const d of decisions) {
+    if (d?.action === "create" && Array.isArray(d.evidence)) {
+      d.evidence = d.evidence.filter((id) => typeof id === "string" && scannedIds.has(id));
+    }
+  }
+  const snapshot = new Map();
+  const { ok, errors } = validateDecisions(decisions, snapshot, {
+    maxCreatePerRun: maxPatterns
+  });
+  if (!ok) return { status: "failed", error: `invalid decisions: ${errors.join("; ")}` };
+  const { applied, failures, conflicts } = applyDecisions(decisions, service, logger, snapshot, config);
+  return {
+    status: applied > 0 ? "ok" : "noop",
+    scanned: memories.length,
+    applied,
+    failures,
+    conflicts
+  };
+}
+
+/**
+ * Phase 4 — entity relation completion. Detects orphan entities (zero
+ * relations) and completes implied relations from memory co-occurrence:
+ * entities named in the same memory → related_to; container kinds
+ * (project/module) → part_of; tech-ish pairs → depends_on. Deterministic,
+ * no LLM — cheap, so it runs even without a route. saveRelation is
+ * bookkeeping (no write hook), so it never re-triggers the scheduler.
+ */
+function inferRelationType(a, b) {
+  if ((a.type === "project" || a.type === "module") && a.type !== b.type) return "part_of";
+  if ((b.type === "project" || b.type === "module") && b.type !== a.type) return "part_of";
+  if (/npm|plugin|api|sdk|lib|framework|package|deps?|build/i.test(`${a.name} ${b.name}`)) return "depends_on";
+  return "related_to";
+}
+
+function phaseRelations(service, config, logger, runId, signal = null) {
+  const entities = service.listEntities({ limit: 1000 }) ?? [];
+  if (entities.length < 2) return { status: "skipped", reason: "too few entities" };
+  const orphans = entities.filter((e) => (service.getRelations(e.id) ?? []).length === 0);
+  if (orphans.length === 0) return { status: "skipped", reason: "no orphan entities" };
+  const memories = service.all().filter((m) => !m.archived && !m.forgotten);
+  const seen = new Set();
+  const related = [];
+  const MAX_RELATIONS_PER_ORPHAN = 3;
+  for (const o of orphans) {
+    if (signal?.aborted) break;
+    let made = 0;
+    for (const m of memories) {
+      if (signal?.aborted || made >= MAX_RELATIONS_PER_ORPHAN) break;
+      const text = `${m.title ?? ""} ${m.content ?? ""}`;
+      if (!text.includes(o.name)) continue;
+      for (const other of entities) {
+        if (other.id === o.id || other.name === o.name) continue;
+        const key = [o.id, other.id].sort().join("|");
+        if (seen.has(key)) continue;
+        if (!text.includes(other.name)) continue;
+        const relationType = inferRelationType(o, other);
+        try {
+          service.saveRelation({ from_entity: o.id, to_entity: other.id, relation_type: relationType, memory_id: m.id, metadata: { source: "sleep_relation_completion" } });
+          seen.add(key);
+          related.push({ from: o.id, to: other.id, type: relationType });
+          made++;
+        } catch (error) {
+          logger?.warn?.(`dsh-mneme sleep: relation ${o.id}/${other.id} failed: ${String(error)}`);
+        }
+      }
+    }
+  }
+  return {
+    status: related.length > 0 ? "ok" : "noop",
+    orphanCount: orphans.length,
+    related
+  };
+}
+
+// ---------------------------------------------------------------- run
+
+function deriveStatus(phases) {
+  const list = Object.values(phases);
+  if (list.length === 0) return "noop";
+  const anyError = list.some((p) => p.status === "failed" || p.status === "error");
+  // Issue #126 review（Copilot）：phase 自身可以是 degraded（合法子集已应用但有条目被
+  // 跳过），它同样算"有产出"——否则整轮会被误判成 noop，把真实变更报成空轮。
+  const anyWork = list.some((p) => p.status === "ok" || p.status === "degraded");
+  const anyDegraded = list.some((p) => p.status === "degraded");
+  if (anyDegraded) return "degraded";
+  if (anyWork && anyError) return "degraded";
+  if (anyError) return "failed";
+  if (anyWork) return "ok";
+  return "noop";
+}
+
+/**
+ * Run one full sleep cycle. Best-effort across all phases; writes a
+ * run_type='sleep' audit row (same dream_runs table) so sleep activity is
+ * observable alongside consolidation runs.
+ */
+export async function runSleep(ctx, service, config, logger, semantic = null, signal = null) {
+  const runId = randomUUID();
+  // Issue #89：开跑时刻——审计行 created_at 的基准（= 冷却种子），run 耗时
+  // （LLM 各 phase）不应计入重启后的冷却窗口。
+  const sleepStartedAt = Date.now();
+  const phases = {};
+  const attempt = async (name, fn) => {
+    if (signal?.aborted) return; // user resumed activity — stop before next phase
+    try {
+      phases[name] = await fn();
+    } catch (error) {
+      phases[name] = { status: "failed", error: error?.message ?? String(error) };
+      logger?.warn?.(`dsh-mneme sleep: ${name} phase failed: ${error?.message ?? error}`);
+    }
+  };
+  await attempt("conflicts", () => phaseConflicts(ctx, service, config, logger, runId, semantic, signal));
+  await attempt("demotion", () => phaseDemotion(service, config, logger, runId, signal));
+  await attempt("patterns", () => phasePatterns(ctx, service, config, logger, runId, signal));
+  await attempt("relations", () => phaseRelations(service, config, logger, runId, signal));
+
+  const status = deriveStatus(phases);
+  const route = resolveSleepRoute(ctx, config, logger);
+  const totalApplied = Object.values(phases).reduce((n, p) => n + (Number.isInteger(p?.applied) ? p.applied : 0), 0);
+  // Issue #126（与 #104 同向）：把各 phase 的跳过明细汇总成一行审计数据，带 phase
+  // 名（同一次 run 里 conflicts 与 patterns 都可能产出 skipped）。全空时写 NULL，
+  // 不落空数组——审计行只记真实发生过的跳过。
+  const skippedDetail = Object.entries(phases).flatMap(([phase, p]) =>
+    Array.isArray(p?.skipped) ? p.skipped.map((s) => ({ phase, ...s })) : []
+  );
+  const snapshotHash = createHash("sha256").update(JSON.stringify(phases)).digest("hex");
+  const receipt = buildReceipt({
+    runId,
+    status,
+    snapshotHash,
+    inputCount: 0,
+    applied: totalApplied,
+    summaryStored: false
+  });
+  try {
+    service.saveDreamRun({
+      id: runId,
+      // Issue #89：审计行记开跑时刻而非完成时刻——run 耗时（含 LLM 各 phase）
+      // 不该计入重启后的 sleep 冷却种子。
+      created_at: new Date(sleepStartedAt).toISOString(),
+      status,
+      provider: route?.provider,
+      model: route?.model,
+      snapshot_hash: snapshotHash,
+      input_count: 0,
+      input: null,
+      decisions: phases,
+      outcome: phases,
+      applied: totalApplied,
+      summary_stored: false,
+      receipt,
+      policy_epoch: config.policyEpoch ?? 0,
+      run_type: "sleep",
+      skipped: skippedDetail.length ? skippedDetail : undefined
+    });
+  } catch (error) {
+    logger?.warn?.(`dsh-mneme sleep: failed to record audit run: ${String(error)}`);
+  }
+  return { ok: status === "ok" || status === "degraded", status, runId, phases, receipt };
+}
+
+// ---------------------------------------------------------------- scheduler
+
+/**
+ * Idle-triggered scheduler. DSH plugins have no resident cron, so a sleep run
+ * fires when: sleep is enabled, the store has been quiet for sleepIdleMinutes,
+ * and the previous run is older than sleepMinIntervalHours. noteWrite() is
+ * called on every store write and (re)arms an idle timer that re-checks at the
+ * exact moment the idle window elapses — no polling, no cron.
+ *
+ * A `now` clock can be injected for tests; it defaults to Date.now.
+ */
+export function createSleepScheduler({
+  service,
+  config,
+  logger,
+  onRun = null,
+  now = () => Date.now(),
+  setTimeoutFn = setTimeout,
+  clearTimeoutFn = clearTimeout,
+  // Issue #89 同族：冷却时刻由调用方从 dream_runs 审计表恢复（run_type='sleep'）——
+  // 内存变量进程重启即归零，冷却闸对新实例放行 → 重启后立即进入 sleep 冷却盲区。
+  // 测试注入 0（默认）即保持旧行为。
+  lastRunAtSeed = 0
+}) {
+  let lastWriteAt = now();
+  let lastRunAt = lastRunAtSeed;
+  let running = false;
+  let disposed = false;
+  let idleTimer = null;
+  let sleepAbort = null;
+
+  function armIdleTimer() {
+    if (disposed || idleTimer) return;
+    if (config.sleepModeEnabled !== true) return;
+    const idleMs = (config.sleepIdleMinutes ?? 5) * 60000;
+    const delay = Math.max(0, idleMs - (now() - lastWriteAt)) + 1000;
+    idleTimer = setTimeoutFn(async () => {
+      idleTimer = null;
+      // Issue #187：false = 撞上 CD（或 run 失败）——此前这里直接丢弃，"CD 到期"
+      // 成了没有监听者的时刻，要等下一次写入才救回来。改为按剩余窗口重排。
+      const ran = await maybeSchedule();
+      if (!ran) armNextWindow();
+    }, delay);
+    idleTimer.unref?.();
+  }
+
+  // Issue #187：撞 CD / run 失败后的补挂表——按「剩余 CD / 剩余静默」里更晚的
+  // 到点时刻重排一次，回调里重复同样的判断所以不会空转。跑成了一轮就不重排：
+  // 安静的库没有要处理的新东西，触发权交还给下一次写入的 noteWrite。
+  function armNextWindow() {
+    if (disposed || idleTimer) return;
+    if (config.sleepModeEnabled !== true) return;
+    // lastRunAt === 0 表示从未跑过（此路径不该出现，防御性返回）。
+    if (lastRunAt <= 0) return;
+    const idleMs = (config.sleepIdleMinutes ?? 5) * 60000;
+    const cdMs = (config.sleepMinIntervalHours ?? 8) * 3600000;
+    const delay = Math.max(
+      lastRunAt + cdMs - now(),
+      lastWriteAt + idleMs - now(),
+      0
+    ) + 1000;
+    idleTimer = setTimeoutFn(async () => {
+      idleTimer = null;
+      const ran = await maybeSchedule();
+      if (!ran) armNextWindow();
+    }, delay);
+    idleTimer.unref?.();
+  }
+
+  function shouldRun(at = now()) {
+    if (disposed || running) return false;
+    if (config.sleepModeEnabled !== true) return false;
+    if (at - lastWriteAt < (config.sleepIdleMinutes ?? 5) * 60000) return false;
+    // lastRunAt === 0 means never ran — the min-interval check must not block
+    // the very first cycle (a real run stamps a nonzero timestamp).
+    if (lastRunAt > 0 && at - lastRunAt < (config.sleepMinIntervalHours ?? 8) * 3600000) return false;
+    return true;
+  }
+
+  /** Called on writes: resets the idle clock and re-arms the fire timer. The
+   *  pending timer is cleared first — a stale timer armed against the old idle
+   *  window would otherwise fire early, fail shouldRun, and leave nothing armed
+   *  for the next window (a missed trigger until the next write).
+   *
+   *  While a sleep run is executing (running=true) the in-flight AbortController
+   *  is NOT aborted: the run's own writes (demoteToSummary / setArchived ride
+   *  the normal write-hook path) would otherwise self-abort the cycle. External
+   *  activity during the run still resets the idle clock here, so no new cycle
+   *  fires until the store is quiet again. */
+  function noteWrite() {
+    lastWriteAt = now();
+    if (!running && sleepAbort) {
+      sleepAbort.abort(); // user resumed activity — interrupt an idle run
+    }
+    if (idleTimer) {
+      clearTimeoutFn(idleTimer);
+      idleTimer = null;
+    }
+    armIdleTimer();
+  }
+
+  async function maybeSchedule() {
+    if (!shouldRun()) return false;
+    running = true;
+    const abort = new AbortController();
+    sleepAbort = abort;
+    try {
+      lastRunAt = now();
+      const result = await service.enqueue(() =>
+        onRun ? onRun(abort.signal) : Promise.resolve({ ok: true, skipped: true })
+      );
+      return !!(result && result.ok);
+    } catch (error) {
+      logger?.warn?.(`dsh-mneme sleep: run failed: ${error?.message ?? error}`);
+      return false;
+    } finally {
+      sleepAbort = null;
+      running = false;
+    }
+  }
+
+  async function dispose() {
+    disposed = true;
+    if (idleTimer) {
+      clearTimeoutFn(idleTimer);
+      idleTimer = null;
+    }
+    if (sleepAbort) {
+      sleepAbort.abort();
+      sleepAbort = null;
+    }
+  }
+
+  // Issue #187（同族缺陷）：构造即挂表——重启后只要没发生过写入，此前永远没有
+  // 闹钟，安静多久都不会触发。lastWriteAt 已在构造时设为当前时刻，shouldRun 的
+  // 静默条件自然生效：重启后静默满 idleMinutes 才可能起跑。
+  armIdleTimer();
+
+  return { noteWrite, maybeSchedule, shouldRun, dispose };
+}