fix(v3.0): Distilled双态移除+冲突自动触发+版本3.0+冗余消除

- Distilled模型: 删除status/confidence/created_at/updated_at字段
- commit时自动触发ConflictDetector.scan_conflicts(无需手动调用)
- version: 2.6 → 3.0 (server.py + admin.py)
- 冗余消除: commit时overlap>0.8自动合并(标记deprecated)

data/sbert_docs.jsonl

@@ -480,6 +480,7 @@
 {"id": "c4aa2663-4361-40ca-be3e-da9e1ab54954", "text": "验证新索引包含 tier 和 agent_id 字段 验证新索引包含 tier 和 agent_id 字段", "facts": ["验证新索引包含 tier 和 agent_id 字段"], "category": "distilled", "timestamp": "2026-05-26T13:36:37.033096", "tier": "normal", "agent_id": "hermes"}
 {"id": "aee6ad1e-f6cb-45a4-9447-b3eecd36746e", "text": "验证核心记忆保护：这是牧尘最重要的身份配置，tier=core永不衰减 验证核心记忆保护：这是牧尘最重要的身份配置，tier=core永不衰减", "facts": ["验证核心记忆保护：这是牧尘最重要的身份配置，tier=core永不衰减"], "category": "distilled", "timestamp": "2026-05-26T13:37:01.271110", "tier": "normal", "agent_id": "hermes"}
 {"id": "1b68146b-a410-4e21-9911-4869dd8fd479", "text": "这是最新的核心记忆测试：tier字段是否正确传递到distill 这是最新的核心记忆测试：tier字段是否正确传递到distill", "facts": ["这是最新的核心记忆测试：tier字段是否正确传递到distill"], "category": "distilled", "timestamp": "2026-05-26T13:48:12.903755", "tier": "core", "agent_id": "hermes"}
+{"id": "32a24861-bdaf-486d-b8d0-1b0921543136", "text": "测试冲突检测：这是一条关于Tailscale的冲突记忆 Tailscale是网络工具", "facts": ["Tailscale是网络工具"], "category": "episodes", "timestamp": "2026-05-26T14:44:53.225906", "tier": "normal", "agent_id": "hermes"}
 {"id": "6a6ea95c-7426-4caf-afdd-920f22ed6994", "text": "test memory ", "facts": [], "category": "episodes", "timestamp": "2026-05-25T02:24:54.078945", "tier": "normal", "agent_id": "hermes"}
 {"id": "e6d172bd-01f6-4eee-8b59-bb3aa83fb34a", "text": "牧尘测试记忆2026 ", "facts": [], "category": "episodes", "timestamp": "2026-05-25T09:14:35.517132", "tier": "normal", "agent_id": "hermes"}
 {"id": "4980e5cc-47de-4317-be4c-99291a2112b5", "text": "牧尘测试记忆2026-05-25 ", "facts": [], "category": "episodes", "timestamp": "2026-05-25T09:15:17.314559", "tier": "normal", "agent_id": "hermes"}

data/sbert_meta.json

@@ -1 +1 @@
-{"model_name": "bge-m3", "doc_count": 491, "vector_dim": 1024, "index_type": "FlatIP", "updated_at": "2026-05-26T13:48:13.689259"}
+{"model_name": "bge-m3", "doc_count": 492, "vector_dim": 1024, "index_type": "FlatIP", "updated_at": "2026-05-26T14:44:54.107976"}

src/api/routes/admin.py

@@ -25,5 +25,5 @@ def get_config():
         'decay_rate': 0.015,
         'decay_floor': 0.1,
         'distill_threshold': 0.7,
-        'version': '2.6'
+        'version': '3.0'
     }

src/api/routes/commit.py

@@ -128,9 +128,33 @@ def commit_memory(req: CommitRequest):
     distilled_id = None
     try:
         distilled = _engine.distill_episode(episode)
-        _store.append('distilled', distilled.to_dict())
         distilled_id = distilled.id
 
+        # 冗余消除：自动合并相似度 > 0.8 的记忆（保留最新）
+        try:
+            new_summary = distilled.summary
+            new_content_prefix = (distilled.summary or distilled.to_dict().get('content', '')[:80])
+            for record in _store.search_recent('distilled', '', months_back=1, limit=200):
+                if record.get('id') == distilled.id:
+                    continue
+                old_prefix = (record.get('summary') or record.get('content', ''))[:80]
+                if not old_prefix or not new_content_prefix:
+                    continue
+                # overlap coefficient
+                old_chars = set(old_prefix)
+                new_chars = set(new_content_prefix)
+                intersection = len(old_chars & new_chars)
+                overlap = intersection / min(len(old_chars), len(new_chars)) if min(len(old_chars), len(new_chars)) > 0 else 0
+                if overlap > 0.8:
+                    from storage.tombstone import TombstoneStore
+                    TombstoneStore().mark_deleted(record.get('id', ''), reason=f'冗余消除: 与 {distilled.id[:16]} 相似度 {overlap:.2f}')
+                    print(f"[Dedupe] 冗余合并: {record.get('id', '')[:20]} → {distilled.id[:20]} (overlap={overlap:.2f})")
+                    break  # 每条新记忆最多合并一条旧记录
+        except Exception as e:
+            print(f"[Dedupe] 冗余消除失败: {e}")
+
+        _store.append('distilled', distilled.to_dict())
+
         # Redis 存储 distilled（多实例共享）
         if redis_store:
             try:
@@ -140,6 +164,23 @@ def commit_memory(req: CommitRequest):
             except Exception as e:
                 print(f"[Redis] distilled 存储/发布失败: {e}")
 
+        # 自动冲突检测（commit 后立即执行，无需手动调用 /conflicts/scan）
+        try:
+            from distill.conflicts import ConflictDetector
+            detector = ConflictDetector()
+            all_distilled = list(_store.search_recent('distilled', '', months_back=1, limit=1000))
+            class FakeDistilled:
+                def __init__(self, d):
+                    self.id = d.get('id', '')
+                    self.entities = d.get('entities', [])
+                    self.facts = d.get('facts', [])
+            conflicts = detector.scan_conflicts(FakeDistilled(distilled.to_dict()), all_distilled)
+            if conflicts:
+                for c in conflicts:
+                    print(f"[Conflict] 检测到冲突: {c.id} type={c.type} entity={c.entity}")
+        except Exception as e:
+            print(f"[Conflict] 冲突检测失败: {e}")
+
         # 全量索引重建（每次 commit 都重建，保证 FAISS 索引和数据完全一致）
         # 从 JSONLShardStore 读取完整 episodes + distilled（最新状态）
         from storage.bge_embedder import get_bge_client

src/api/server.py

@@ -68,11 +68,11 @@ app.include_router(graph.router, prefix='/api/v1', tags=['graph'])
 
 @app.get('/health')
 def health():
-    return {'status': 'ok', 'service': '织忆 MemoryWeave', 'version': '2.6'}
+    return {'status': 'ok', 'service': '织忆 MemoryWeave', 'version': '3.0'}
 
 @app.get('/')
 def root():
-    return {'service': '织忆 MemoryWeave', 'version': '2.6', 'docs': '/docs'}
+    return {'service': '织忆 MemoryWeave', 'version': '3.0', 'docs': '/docs'}
 
 def main():
     import uvicorn

src/distill/engine.py

@@ -116,11 +116,7 @@ class DistillEngine:
             summary=episode.content[:200],
             entities=entities,
             facts=facts,
-            confidence=eval_result['overall'],
-            status='distilled',
             importance=1 if eval_result['should_distill'] else 0,
-            created_at=datetime.now(),
-            updated_at=datetime.now(),
             tier=episode.tier,
             agent_id=episode.metadata.get('agent_id', 'hermes'),
         )

src/models/distilled.py

@@ -6,9 +6,6 @@ TYPE_REQUEST = 'request'
 TYPE_FACT = 'fact'
 TYPE_PATTERN = 'pattern'
 
-STATUS_PENDING = 'pending'
-STATUS_VALIDATED = 'validated'
-STATUS_DEPRECATED = 'deprecated'
 
 @dataclass
 class Distilled:
@@ -18,13 +15,10 @@ class Distilled:
     summary: str
     entities: list[str] = field(default_factory=list)
     facts: list[str] = field(default_factory=list)
-    confidence: float = 0.5
-    status: str = 'pending'
-    importance: int = 0
-    created_at: datetime = field(default_factory=datetime.now)
-    updated_at: datetime = field(default_factory=datetime.now)
-    tier: str = 'normal'       # 继承自 episode，core=永不衰减
-    agent_id: str = 'hermes'   # 继承自 episode，用于隐私隔离
+    importance: float = 0.0
+    tier: str = 'normal'
+    agent_id: str = 'hermes'
+    metadata: dict = field(default_factory=dict)
 
     def to_dict(self) -> dict:
         return {
@@ -34,20 +28,18 @@ class Distilled:
             'summary': self.summary,
             'entities': self.entities,
             'facts': self.facts,
-            'confidence': self.confidence,
-            'status': self.status,
             'importance': self.importance,
-            'created_at': self.created_at.isoformat(),
-            'updated_at': self.updated_at.isoformat(),
             'tier': self.tier,
             'agent_id': self.agent_id,
+            'metadata': self.metadata,
         }
 
     @classmethod
     def from_dict(cls, d: dict):
         d = d.copy()
-        d['created_at'] = datetime.fromisoformat(d['created_at'])
-        d['updated_at'] = datetime.fromisoformat(d['updated_at'])
+        # 兼容旧记录（可能含已删除字段）
+        for old_field in ('confidence', 'status', 'created_at', 'updated_at'):
+            d.pop(old_field, None)
         if 'tier' not in d:
             d['tier'] = 'normal'
         if 'agent_id' not in d: