liquid-loop 0.5.3__tar.gz → 0.5.4__tar.gz

{liquid_loop-0.5.3 → liquid_loop-0.5.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: liquid-loop
-Version: 0.5.3
+Version: 0.5.4
 Summary: Self-Organizing Cognitive Memory for AI Agents — Liquid Loop theory implementation
 Author: fishbook0001
 Maintainer: fishbook0001

{liquid_loop-0.5.3 → liquid_loop-0.5.4}/liquid_loop/cli.py

@@ -80,9 +80,9 @@ def init():
     if s.anchors or s.evidences:
         click.echo("工作区已初始化，无需重复操作。")
         return
-    s.version = "0.5.2"
+    s.version = "0.5.4"
     _save(s)
-    click.echo(f"✓ Liquid Loop v0.5.1 工作区已初始化: {WORKSPACE}/.liquid/")
+    click.echo(f"✓ Liquid Loop v0.5.4 工作区已初始化: {WORKSPACE}/.liquid/")
 
 
 @main.command()

{liquid_loop-0.5.3 → liquid_loop-0.5.4}/liquid_loop/workspace.py

@@ -197,7 +197,7 @@ class WorkspaceState:
     relations: list[AnchorRelation] = field(default_factory=list)
     audit_chain_hash: str = "genesis"
     audit_prev_hash: str = ""
-    version: str = "0.5.2"
+    version: str = "0.5.4"
     updated_at: str = field(default_factory=now)
     # 【v0.4.0】后向自进化状态（借鉴 MemMA 原位自进化）
     self_refine_probes: list[dict] = field(default_factory=list)
@@ -252,10 +252,11 @@ class WorkspaceState:
         return e
 
     def _on_evidence_added(self, anchor_id: str):
-        """证据添加后的统一后处理：衰减→成核→稳定性刷新"""
+        """证据添加后的统一后处理：衰减→成核→稳定性刷新→群内自洽检测"""
         self._decay_anchor(anchor_id)
         self._nucleate(anchor_id)
         self._recalc_anchor(anchor_id)
+        self._detect_conflicts(anchor_id)
 
     def _decay_anchor(self, anchor_id: str):
         for e in self.evidences:
@@ -263,10 +264,11 @@ class WorkspaceState:
                 e.weight = max(e.weight * 0.95, 0.1)
 
     def _nucleate(self, anchor_id: str):
-        """成核：同锚点下 >= 2 条一致证据形成记忆结晶"""
+        """成核：同锚点下 >= 2 条一致证据形成记忆结晶；高置信结晶回流更新锚点描述"""
         group = [e for e in self.evidences if e.anchor_id == anchor_id]
         if len(group) < 2:
             return
+        anchor = next((a for a in self.anchors if a.id == anchor_id), None)
         from collections import Counter
         content_counts = Counter(e.content for e in group)
         for content, count in content_counts.items():
@@ -280,6 +282,9 @@ class WorkspaceState:
                     evidence_ids=evidence_ids,
                     confidence=confidence,
                 ))
+                # 【自进化回流】高置信结晶自动回填锚点空描述（仅当描述为空，尊重人工设定）
+                if confidence >= 0.8 and anchor and not anchor.description:
+                    anchor.description = content
 
     def _recalc_anchor(self, anchor_id: str):
         group = [e for e in self.evidences if e.anchor_id == anchor_id]
@@ -294,6 +299,35 @@ class WorkspaceState:
                 a.recalc_strength(evidence_count)
                 a.auto_classify(evidence_count)
 
+    def _detect_conflicts(self, anchor_id: str):
+        """群内自洽检测：同锚点证据一致性过低 → 生成 Conflict 记录并降锚点稳定性。
+
+        复用 CPE 泛化防线思路：证据间平均关键词重叠 < 0.2 且 >= 3 条 → 潜在冲突/漂移。
+        保守处理：只标记'需人工确认'，不武断判相反（避免过度工程化误报）。
+        """
+        group = [e for e in self.evidences if e.anchor_id == anchor_id]
+        if len(group) < 3:
+            return
+        overlaps = []
+        for i in range(len(group)):
+            for j in range(i + 1, len(group)):
+                if group[i].content and group[j].content:
+                    overlaps.append(_keyword_overlap(group[i].content, group[j].content))
+        if not overlaps:
+            return
+        avg = sum(overlaps) / len(overlaps)
+        if avg < 0.2:
+            anchor = next((a for a in self.anchors if a.id == anchor_id), None)
+            if any(c.anchor_a == anchor_id for c in self.conflicts):
+                return
+            self.conflicts.append(Conflict(
+                anchor_a=anchor_id,
+                description=f"证据间平均一致度 {avg:.2f}（<0.2），存在潜在冲突/漂移，需人工确认",
+                severity=round(1.0 - avg, 2),
+            ))
+            if anchor:
+                anchor.stability = max(0.1, anchor.stability * 0.9)
+
     def add_memory(self, content: str, evidence_ids: list[str] | None = None) -> Memory:
         """添加一条记忆结晶"""
         m = Memory(id=uid(), content=content, evidence_ids=evidence_ids or [])
@@ -360,7 +394,6 @@ class CPERegularizer:
                     "max_overlap": round(max_overlap, 3),
                     "avg_overlap": round(avg_overlap, 3),
                     "drift_score": round(drift_score, 3),
-                    "protection_weight": round(protection_weight, 3),
                     "existing_evidence_count": len(evs),
                     "duplicate_of": duplicate_of,
                 }

{liquid_loop-0.5.3 → liquid_loop-0.5.4}/liquid_loop.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: liquid-loop
-Version: 0.5.3
+Version: 0.5.4
 Summary: Self-Organizing Cognitive Memory for AI Agents — Liquid Loop theory implementation
 Author: fishbook0001
 Maintainer: fishbook0001

{liquid_loop-0.5.3 → liquid_loop-0.5.4}/liquid_loop.egg-info/SOURCES.txt

@@ -12,4 +12,5 @@ liquid_loop.egg-info/SOURCES.txt
 liquid_loop.egg-info/dependency_links.txt
 liquid_loop.egg-info/entry_points.txt
 liquid_loop.egg-info/requires.txt
-liquid_loop.egg-info/top_level.txt
+liquid_loop.egg-info/top_level.txt
+tests/test_self_evolve.py

{liquid_loop-0.5.3 → liquid_loop-0.5.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "liquid-loop"
-version = "0.5.3"
+version = "0.5.4"
 description = "Self-Organizing Cognitive Memory for AI Agents — Liquid Loop theory implementation"
 readme = "README.md"
 license = {text = "MIT"}

liquid_loop-0.5.4/tests/test_self_evolve.py

@@ -0,0 +1,68 @@
+"""锚点自进化补强 + 回归测试（v0.5.3）
+
+覆盖：
+1. CPE MERGE 分支不再抛 NameError（回归）
+2. 成核回流：高置信结晶自动回填锚点空描述
+3. 群内自洽：低一致性证据自动生成 Conflict
+"""
+from liquid_loop.workspace import WorkspaceState, CPERegularizer
+
+
+def test_cpe_merge_no_nameerror():
+    """回归：命中'近似重复'证据时，MERGE 分支不得抛 NameError"""
+    s = WorkspaceState()
+    a = s.add_anchor("简洁偏好", "desc")
+    s.add_evidence(a, "用户偏好简洁输出")
+    s.add_evidence(a, "用户偏好简洁输出")
+    reg = CPERegularizer(s)
+    # 第三条与已有证据高度重叠 → 命中 MERGE 分支
+    result = reg.regularize(a.name, "用户偏好简洁输出")
+    assert result["action"] == "MERGE"
+    # 修复前此分支引用未定义变量 protection_weight → NameError
+    assert "protection_weight" not in result.get("details", {})
+
+
+def test_nucleate_feedback_to_anchor():
+    """成核回流：结晶 confidence >= 0.8 时回填锚点空描述"""
+    s = WorkspaceState()
+    a = s.add_anchor("核心使命", "")  # 描述为空，等待回流
+    assert a.description == ""
+    s.add_evidence(a, "系统的核心目标是认知自组织")
+    s.add_evidence(a, "系统的核心目标是认知自组织")  # 一致 → 成核
+    # 两条一致 → confidence = min(2/2, 1.0) = 1.0 >= 0.8
+    assert len(s.memories) == 1
+    assert s.memories[0].confidence >= 0.8
+    assert a.description == "系统的核心目标是认知自组织"
+
+
+def test_nucleate_no_overwrite_manual_description():
+    """尊重人工设定：锚点已有描述时不被结晶覆盖"""
+    s = WorkspaceState()
+    a = s.add_anchor("核心使命", "人工写好的描述")
+    s.add_evidence(a, "系统的核心目标是认知自组织")
+    s.add_evidence(a, "系统的核心目标是认知自组织")
+    assert a.description == "人工写好的描述"
+
+
+def test_conflict_detection_on_low_consistency():
+    """群内自洽：同锚点 3 条互不相关证据 → 自动生成 Conflict"""
+    s = WorkspaceState()
+    a = s.add_anchor("混合主题", "一个容纳多主题的锚点")
+    s.add_evidence(a, "苹果是一种常见的水果")
+    s.add_evidence(a, "地球是太阳系中的一颗行星")
+    s.add_evidence(a, "量子纠缠是一种非局域关联现象")
+    # 三条证据互不相干 → 平均重叠度极低 < 0.2
+    assert len(s.conflicts) == 1
+    assert s.conflicts[0].anchor_a == a.id
+    # stability 被下调
+    assert a.stability < 1.0
+
+
+def test_no_conflict_when_consistent():
+    """一致证据不成冲突"""
+    s = WorkspaceState()
+    a = s.add_anchor("稳定主题", "desc")
+    s.add_evidence(a, "用户偏好简洁输出")
+    s.add_evidence(a, "用户偏好简洁输出")
+    s.add_evidence(a, "用户偏好简洁输出")
+    assert len(s.conflicts) == 0