优化加载后删除节点的逻辑；和数据库查询性能

docs/node_state_definition.md

@@ -37,6 +37,10 @@
 | `ABNORMAL` | 数据损坏或核心绑定异常 | 否 |
 | `DEPENDENCY_ERROR` | 依赖不可满足 | 否 |
 
+bootstrap 特殊语义：
+- 如果节点在持久化恢复后存在绑定，但 bootstrap datasource refresh 未返回最新数据，则节点进入 `S17` 语义并被 `bootstrap_blocked` 隔离。
+- 如果节点仅残留在 persistence 中、没有绑定，且 datasource refresh 也未返回，则节点按 stale zombie 清理，不再进入 bind。
+
 ## 3. create / update 语义
 
 ### create
@@ -64,6 +68,11 @@
 - 状态落地：`Decision.to_state` 对应 YAML `states[Sxx]`。
 - 不变量：每次落地后执行 `runtime.check_invariants()`。
 
+bootstrap 额外输入：
+- `is_create_zombie`
+- `bootstrap_source_present`
+- `has_binding_record`
+
 ## 6. 关键一致性约束
 
 - `UNCHECKED` 不允许带动作（`INV-1_UNCHECKED_ACTION_NONE`）。

docs/state_machine.md

@@ -21,6 +21,7 @@
 - `S12C/S12U/S12D`：按动作拆分的执行失败状态
 - `S13`：对侧创建等待态（`PEER_CREATING/NONE/PENDING`）
 - `S14`：UPDATE 运行时降级为 `NONE` 并跳过
+- `S17`：bootstrap 加载异常隔离（存在绑定，但 datasource refresh 未返回最新节点）
 
 ## 3. 事件到代码映射
 
@@ -51,6 +52,7 @@
 ## 5. context 字段（以 YAML 为准）
 
 ### bind 相关
+- `bootstrap_source_present`
 - `has_binding_record`
 - `core_binding_result` / `peer_core_binding_result` (`NORMAL|WARNING|ABNORMAL`)
 - `data_present`
@@ -75,6 +77,7 @@
 ## 6. 语义边界
 - strategy 负责业务语义计算（如依赖是否满足、自动绑定结果、差异检测）。
 - 状态机只负责迁移判定与不变量校验，不直接访问数据源。
+- bootstrap 阶段的数据源刷新由 pipeline 完成，状态机只消费 `bootstrap_source_present / has_binding_record / is_create_zombie` 这些已计算上下文。
 
 ## 7. 不变量
 - `INV-1_UNCHECKED_ACTION_NONE`

docs/sync_flow.md

@@ -13,23 +13,34 @@
 
 ### 2.1 加载时恢复
 - 在 `sync_state_machine/common/collection.py::load_from_persistence()` 执行。
-- 语义：仅恢复持久化状态（不在此阶段做状态归并）。
+- 语义：仅恢复持久化状态，并记录本轮 `sync_log` 的 `persistence` 恢复来源；不在此阶段做状态归并。
+- 恢复后保留 `data/origin_data`，等待 bootstrap 阶段的 datasource 刷新覆盖最新数据。
 
-### 2.2 加载后 cleanup
+### 2.2 bootstrap datasource refresh
-- pipeline 直接调用 `run_phase2_cleanup()`。
+- pipeline 在持久化恢复后、`E01` 之前，按已恢复到内存的 `node_type` 做一次 datasource 刷新。
+- 实际实现：`sync_state_machine/pipeline/full_sync_pipeline.py::_bootstrap_refresh_latest_data()`。
+- 语义：
+	- 如果 datasource 命中节点，则用最新数据覆盖持久化快照，并清除 `_loaded_from_persistence` 标记。
+	- 如果 datasource 未命中节点，则节点保留持久化快照，供 `E01` 判断它是 stale zombie 还是 bootstrap 异常隔离对象。
+
+### 2.3 加载后 cleanup
+- pipeline 在 datasource refresh 完成后调用 `run_phase2_cleanup()`。
 - 实际实现：`sync_state_machine/sync_system/strategy_ops/reset_ops.py::run_phase2_cleanup()`。
 - 核心事件：`E01`（`S16 -> S00/S15`）。
-- `S16` 表示“持久化加载输入态”：本轮从持久化恢复出的节点集合（可包含上轮遗留节点，也可包含上轮已稳定节点）。
+- `S16` 表示“bootstrap 输入态”：持久化恢复出的节点，叠加 datasource refresh 后仍待 `E01` 判定的节点集合。
 - 判定：
 	- `is_create_zombie=true` -> `S15`，节点删除（并尝试解绑）；
-	- `is_create_zombie=false` -> `S00`，节点归并为 `UNCHECKED/NONE/PENDING` 并清空 `error`。
+	- `bootstrap_source_present=false && has_binding_record=false` -> `S15`，说明该节点仅残留在 persistence 中且未绑定，按 stale zombie 清理；
+	- `bootstrap_source_present=false && has_binding_record=true` -> `S17`，说明该节点存在绑定，但 datasource refresh 未返回，进入 bootstrap 加载异常隔离；
+	- 其余情况 -> `S00`，节点归并为 `UNCHECKED/NONE/PENDING` 并清空 `error`。
 
 ### 2.2.1 `enable_persistence` 与 bootstrap 关系
-- `enable_persistence=true`：先 `load_from_persistence()`，再对加载出的节点执行 `E01`（`S16` 域内处理）。
+- `enable_persistence=true`：先 `load_from_persistence()`，再做 bootstrap datasource refresh，最后对输入节点执行 `E01`（`S16` 域内处理）。
 - `enable_persistence=false`：本轮没有持久化输入，`S16/E01` 不参与；随后 Stage1 新加载节点直接进入本轮 bind/create/update/sync 主流程。
 
-### 2.3 加载异常隔离（S17）
+### 2.4 加载异常隔离（S17）
 - 持久化中若出现非法枚举值（`action/status/binding_status`），节点标记为 `bootstrap_blocked`（`S17` 语义）。
+- 持久化恢复后若节点存在绑定，但 bootstrap datasource refresh 未返回该节点，也会进入 `S17` 语义并隔离。
 - 这类节点保留 `error`，用于人工排查。
 - 这类节点不会进入 `E01`，也不会参与 bind/create/update/sync 自动流程。
 
@@ -51,6 +62,8 @@
 - 有绑定记录：根据本端/对端有效性进入 `S01/S03/S04`。
 - 无绑定记录：进入 `S02`。
 
+注意：如果节点在 bootstrap 阶段已被隔离为 `S17` 语义（`bootstrap_blocked=true`），默认不会进入 bind 工作集。
+
 ### 依赖检查
 - `phase_dependency_check()` -> `check_dependencies_for_nodes()` 内触发 `E15`。
 - `data_present=true` 且 `dep_satisfied=true` -> 不迁移，节点保持在 `S02`，继续进入自动绑定判定。

sync_state_machine/common/collection.py

@@ -488,6 +488,7 @@ class DataCollection:
                     context=context_value,
                 )
                 node.context["_loaded_from_persistence"] = True
+                node.append_log("加载来源: persistence 恢复节点")
             except Exception:
                 import logging
                 logger = logging.getLogger(__name__)
@@ -623,6 +624,7 @@ class DataCollection:
                 context=context_value,
             )
             node.context["_loaded_from_persistence"] = True
+            node.append_log("加载来源: persistence 恢复节点")
             self._nodes[node.node_id] = node
 
             if node.data_id and node.data_id != "":

sync_state_machine/common/persistence_backends/sqlite_backend.py

@@ -112,6 +112,9 @@ class SQLitePersistenceBackend(PersistenceBackend):
         await self.conn.execute(
             "CREATE INDEX IF NOT EXISTS idx_nodes_collection_scope_data_id ON nodes (collection_id, scope, data_id)"
         )
+        await self.conn.execute(
+            "CREATE INDEX IF NOT EXISTS idx_nodes_scope_type_node_id ON nodes (scope, node_type, node_id)"
+        )
 
     async def _ensure_bindings_schema(self) -> None:
         assert self.conn is not None

sync_state_machine/config/node_state_machine.yaml

@@ -17,6 +17,7 @@ meta:
 # represented as separate invocations with different inputs.
 context:
   is_create_zombie: { type: bool, desc: "初始化时判定为 CREATE 僵尸：action=CREATE 且 (status=FAILED 或 data_id 为空)" }
+  bootstrap_source_present: { type: bool, desc: "bootstrap 阶段 datasource 刷新后，当前节点是否在最新数据源中命中" }
 
   has_binding_record: { type: bool, desc: "binding_manager 里存在绑定记录 (remote_id/local_id 非空)" }
   core_binding_result: { type: enum, values: [NORMAL, WARNING, ABNORMAL], desc: "SyncDecisionEngine.determine_core_status() 给本节点的结果（仅在 has_binding_record=true 时有效）" }
@@ -304,19 +305,30 @@ states:
       status: SKIPPED
       data_id_exist: true
 
+  S17:
+    binding_status: ABNORMAL
+    action: NONE
+    status: FAILED
+    data_id_exist: [false, true]
+    desc: |
+      bootstrap 加载异常隔离态
+      持久化恢复后存在绑定，但 datasource 刷新未命中最新数据
+      binding_status: ABNORMAL
+      action: NONE
+      status: FAILED
+      data_id_exist: false | true
+
 pseudo_states:
   S15:
     desc: "节点被清理删除（zombie cleanup）"
   S16:
     desc: "持久化加载输入态：本轮 E01 的输入节点集合（可包含上轮遗留与已稳定节点）"
-  S17:
-    desc: "加载异常隔离态：持久化枚举非法，保留错误供人工处理，不进入自动流程"
 
 stages:
   bootstrap:
-    desc: "启动/初始化（仅针对持久化加载出的节点）：僵尸清理与归并到 S00"
+    desc: "启动/初始化：持久化恢复后先做 datasource 刷新，再按 E01 清理僵尸或归并/隔离"
     entry_states: [S16]
-    exit_states: [S00, S15]
+    exit_states: [S00, S15, S17]
 
   bind:
     desc: "绑定判定：core matrix +（可选）依赖检查 +（可选）自动绑定；stage 仅约束边界，不要求内部顺序"
@@ -346,13 +358,23 @@ stages:
 transitions:
   E01:
     stage: bootstrap
-    desc: "初始化：删CREATE僵尸；其余重置后进入起点"
+    desc: "初始化：先依据 datasource 刷新结果判定 source 是否存在，再做僵尸清理/隔离/归并"
     from: [S16]
     outcomes:
       - when: { is_create_zombie: true }
         to: S15
-      - when: { is_create_zombie: false }
+      - when: { is_create_zombie: false, bootstrap_source_present: false, has_binding_record: false }
+        to: S15
+        emit:
+          note: "持久化孤儿节点未命中最新数据源，按僵尸清理"
+      - when: { is_create_zombie: false, bootstrap_source_present: false, has_binding_record: true }
+        to: S17
+        emit:
+          note: "存在绑定但最新数据源未返回节点，进入加载异常隔离"
+      - when: { is_create_zombie: false, bootstrap_source_present: true }
         to: S00
+        emit:
+          note: "bootstrap 已刷新到最新数据，归并到起点"
 
   # --- Bind phase1 core ---
   E10:

sync_state_machine/domain/construction/api_handler.py

@@ -101,7 +101,7 @@ class ConstructionTaskApiHandler(BaseApiHandler):
                 else:
                     results.append(TaskResult.in_progress(node_id=node.node_id, task_id=push_id))
             except Exception as e:
-                results.append(TaskResult.failed(node_id=node.node_id, error=str(e)))
+                results.append(TaskResult.failed(node_id=node.node_id, error=str(e), push_id=push_id))
         
         return results
     
@@ -135,6 +135,7 @@ class ConstructionTaskApiHandler(BaseApiHandler):
             
             node_updated = False
             error = None
+            push_failure_id = None
             schema_status_lines: List[str] = []
             
             # 1. 基础信息更新 (ConstructionUpdate)
@@ -186,6 +187,7 @@ class ConstructionTaskApiHandler(BaseApiHandler):
                     except Exception as e:
                         error = str(e)
                         base_update_error = str(e)
+                        push_failure_id = push_id
                 else:
                     base_update_error = "show_name changed but task_type != QT"
 
@@ -236,6 +238,7 @@ class ConstructionTaskApiHandler(BaseApiHandler):
                     except Exception as e:
                         error = str(e)
                         plan_update_error = str(e)
+                        push_failure_id = push_id
 
             schema_status_lines.append(
                 self.build_update_schema_status(
@@ -253,7 +256,7 @@ class ConstructionTaskApiHandler(BaseApiHandler):
             
             # 汇总结果
             if error:
-                results.append(TaskResult.failed(node_id=node.node_id, error=error))
+                results.append(TaskResult.failed(node_id=node.node_id, error=error, push_id=push_failure_id))
             elif node_updated:
                 results.append(TaskResult.success(node_id=node.node_id))
             else:
@@ -279,7 +282,7 @@ class ConstructionTaskApiHandler(BaseApiHandler):
                 await api_delete_construction_task(self.api_client, biz_id, push_id)
                 results.append(TaskResult.success(node_id=node.node_id))
             except Exception as e:
-                results.append(TaskResult.failed(node_id=node.node_id, error=str(e)))
+                results.append(TaskResult.failed(node_id=node.node_id, error=str(e), push_id=push_id))
         
         return results
     

sync_state_machine/domain/construction_task_detail/api_handler.py

@@ -130,7 +130,7 @@ class ConstructionTaskDetailApiHandler(BaseApiHandler):
                 else:
                     results.append(TaskResult.in_progress(node_id=node.node_id, task_id=push_id))
             except Exception as e:
-                results.append(TaskResult.failed(node_id=node.node_id, error=str(e)))
+                results.append(TaskResult.failed(node_id=node.node_id, error=str(e), push_id=push_id))
         
         return results
     
@@ -222,7 +222,7 @@ class ConstructionTaskDetailApiHandler(BaseApiHandler):
                 # 整组失败
                 error_msg = str(e)
                 for node, _ in group_items:
-                    results.append(TaskResult.failed(node_id=node.node_id, error=error_msg))
+                    results.append(TaskResult.failed(node_id=node.node_id, error=error_msg, push_id=push_id))
         
         return results
     
@@ -240,7 +240,7 @@ class ConstructionTaskDetailApiHandler(BaseApiHandler):
                 await api_delete_construction_task_detail(self.api_client, biz_id, push_id)
                 results.append(TaskResult.success(node_id=node.node_id))
             except Exception as e:
-                results.append(TaskResult.failed(node_id=node.node_id, error=str(e)))
+                results.append(TaskResult.failed(node_id=node.node_id, error=str(e), push_id=push_id))
         
         return results
     

sync_state_machine/engine/events.py

@@ -276,8 +276,14 @@ def e01_bootstrap(
     *,
     node: "SyncNode",
     is_create_zombie: bool,
+    bootstrap_source_present: bool = True,
+    has_binding_record: bool = False,
 ) -> Decision:
-    context = {"is_create_zombie": is_create_zombie}
+    context = {
+        "is_create_zombie": is_create_zombie,
+        "bootstrap_source_present": bootstrap_source_present,
+        "has_binding_record": has_binding_record,
+    }
     decision = _dispatch_required(
         runtime,
         node=node,
@@ -289,6 +295,15 @@ def e01_bootstrap(
         _apply_decision(runtime, node=node, decision=decision, fields={"binding_status", "action", "status"})
         node.error = None
         return decision
+    if decision.to_state == "S17":
+        _apply_decision(runtime, node=node, decision=decision, fields={"binding_status", "action", "status"})
+        reason_text = decision_note(decision) or "<no-reason>"
+        node.context["bootstrap_blocked"] = True
+        node.context["bootstrap_block_reason"] = "source_missing_after_persistence_refresh"
+        node.context["bootstrap_block_errors"] = [reason_text]
+        node.context["bootstrap_source_present"] = False
+        node.error = f"BOOTSTRAP_SOURCE_MISSING: {reason_text}"
+        return decision
     if decision.to_state == "S15":
         reason_text = decision_note(decision) or "<no-reason>"
         node.append_log(

sync_state_machine/pipeline/full_sync_pipeline.py

@@ -256,6 +256,12 @@ class FullSyncPipeline:
         self._prepare_datasources()
         self._logger.info("✅ Bootstrap: datasource prepared")
 
+        await self._bootstrap_refresh_latest_data()
+        self._logger.info("✅ Bootstrap: datasource refresh completed")
+
+        await self._run_bootstrap_reset_cleanup()
+        self._logger.info("✅ Bootstrap: reset cleanup completed")
+
         await self._apply_project_scope_cleanup()
 
     async def phase_bind(self) -> None:
@@ -359,7 +365,7 @@ class FullSyncPipeline:
         await _safe_close("persistence", self.persistence.close())
 
     async def _load_persistence_state(self) -> None:
-        """加载持久化数据并清理僵尸节点"""
+        """加载持久化数据到内存，等待 bootstrap datasource refresh 与 E01 cleanup。"""
         if not self.enable_persistence:
             return
 
@@ -377,9 +383,49 @@ class FullSyncPipeline:
             await self.binding_manager.load_from_persistence(node_type)
         self._logger.info("🔄 Persistence loaded (state preserved; pending E01 normalization)")
 
-        # 加载后 reset 清理：清理 CREATE 失败的僵尸绑定
+    def _resolve_bootstrap_refresh_node_types(self) -> List[str]:
+        local_types = {node.node_type for node in self.local_collection.filter()}
+        remote_types = {node.node_type for node in self.remote_collection.filter()}
+        candidate_types = local_types | remote_types
+        if not candidate_types:
+            return []
+
+        ordered: List[str] = []
+        for node_type in [*self.load_order, *self.bootstrap_binding_node_types, *sorted(candidate_types)]:
+            if node_type in candidate_types and node_type not in ordered:
+                ordered.append(node_type)
+        return ordered
+
+    async def _bootstrap_refresh_latest_data(self) -> None:
+        if not self.enable_persistence:
+            return
+
+        node_types = self._resolve_bootstrap_refresh_node_types()
+        if not node_types:
+            self._logger.info("⏭️  Bootstrap datasource refresh skipped: no persisted node types")
+            return
+
+        for node_type in node_types:
+            self._logger.info("🔄 Bootstrap refresh node_type=%s from datasource", node_type)
+            await self._load_node_type_from_datasource(
+                self.local_datasource,
+                datasource_role="local",
+                node_type=node_type,
+                reason="bootstrap persistence refresh",
+            )
+            await self._load_node_type_from_datasource(
+                self.remote_datasource,
+                datasource_role="remote",
+                node_type=node_type,
+                reason="bootstrap persistence refresh",
+            )
+
+    async def _run_bootstrap_reset_cleanup(self) -> None:
+        if not self.enable_persistence:
+            return
         if not self.strategies:
             raise RuntimeError("no strategies configured: cannot run bootstrap event E01")
+
         total_cleaned = await run_phase2_cleanup(
             node_types=self.node_types,
             local_collection=self.local_collection,
@@ -388,8 +434,7 @@ class FullSyncPipeline:
             runtime=self.sm_runtime,
         )
         if total_cleaned > 0:
-            self._logger.info(f"🧹 Reset cleanup: {total_cleaned} CREATE-failed zombies cleaned")
+            self._logger.info(f"🧹 Reset cleanup: {total_cleaned} bootstrap zombies cleaned")
-        self._logger.info("🔄 Post-load reset cleanup completed")
 
     async def _apply_preloaded_state(self) -> None:
         if self._preloaded_local_nodes:

sync_state_machine/sync_system/strategy_ops/reset_ops.py

@@ -30,6 +30,25 @@ async def _resolve_local_id_for_unbind(
     return None
 
 
+async def _has_binding_record(
+    *,
+    node_type: str,
+    any_side_node_id: str,
+    binding_manager: "BindingManager",
+) -> bool:
+    local_id = await _resolve_local_id_for_unbind(
+        node_type=node_type,
+        any_side_node_id=any_side_node_id,
+        binding_manager=binding_manager,
+    )
+    return local_id is not None
+
+
+def _bootstrap_source_present(node) -> bool:
+    context = node.context if isinstance(node.context, dict) else {}
+    return not bool(context.get("_loaded_from_persistence"))
+
+
 def _is_create_zombie(node) -> bool:
     if node.action != SyncAction.CREATE:
         return False
@@ -55,8 +74,27 @@ async def _cleanup_one_side(
             )
             continue
 
-        decision = e01_bootstrap(runtime, node=node, is_create_zombie=_is_create_zombie(node))
+        has_binding_record = await _has_binding_record(
+            node_type=node_type,
+            any_side_node_id=node.node_id,
+            binding_manager=binding_manager,
+        )
+        bootstrap_source_present = _bootstrap_source_present(node)
+
+        decision = e01_bootstrap(
+            runtime,
+            node=node,
+            is_create_zombie=_is_create_zombie(node),
+            bootstrap_source_present=bootstrap_source_present,
+            has_binding_record=has_binding_record,
+        )
         if decision.to_state != "S15":
+            if decision.to_state == "S17":
+                node.context.pop("_loaded_from_persistence", None)
+                logger.warning(
+                    f"[{node_type}] Bootstrap isolated node ({label}): "
+                    f"node_id={node.node_id}, reason={node.error}"
+                )
             continue
 
         local_id_for_unbind = await _resolve_local_id_for_unbind(
@@ -74,9 +112,12 @@ async def _cleanup_one_side(
         else:
             bind_note = "no_binding"
 
-        reason = "status=FAILED" if node.status == SyncStatus.FAILED else "data_id empty"
+        if _is_create_zombie(node):
+            reason = "status=FAILED" if node.status == SyncStatus.FAILED else "data_id empty"
+        else:
+            reason = "persistence_only_source_missing_without_binding"
         logger.info(
-            f"[{node_type}] Cleaning up CREATE zombie ({label}): "
+            f"[{node_type}] Cleaning up bootstrap zombie ({label}): "
             f"node_id={node.node_id}, reason={reason}, {bind_note}"
         )
 
@@ -118,7 +159,7 @@ async def run_phase2_cleanup(
         total_cleaned += cleaned_count
         if cleaned_count > 0:
             logger.warning(
-                f"[{node_type}] Cleaned up CREATE-failed zombies: "
+                f"[{node_type}] Cleaned up bootstrap zombies: "
                 f"local_deleted={len(cleaned_local)}, remote_deleted={len(cleaned_remote)}"
             )
 

tests/integration/test_mock_full_event_state_coverage.py

@@ -128,6 +128,16 @@ async def test_mock_based_full_event_and_state_coverage() -> None:
     d = e01_bootstrap(runtime, node=node_e01_del, is_create_zombie=True)
     record(node_e01_del, d)
 
+    node_e01_isolated = _new_node("e01_isolated", action=SyncAction.NONE, status=SyncStatus.PENDING)
+    d = e01_bootstrap(
+        runtime,
+        node=node_e01_isolated,
+        is_create_zombie=False,
+        bootstrap_source_present=False,
+        has_binding_record=True,
+    )
+    record(node_e01_isolated, d)
+
     # E10 merged outcomes
     for idx, (has_binding, local_res, peer_res) in enumerate(
         [
@@ -507,7 +517,8 @@ async def test_sync_log_cleared_on_initial_load(tmp_path: Path) -> None:
     await reloaded.load_from_persistence()
     loaded_node = reloaded.get("log-node")
     assert loaded_node is not None
-    assert loaded_node.sync_log is None
+    assert loaded_node.sync_log is not None
+    assert loaded_node.sync_log.count("加载来源: persistence 恢复节点") == 1
 
     in_memory_after_reload_report = REPORT_DIR / "in_memory_sync_log_after_reload.txt"
     in_memory_after_reload_report.write_text(
@@ -566,6 +577,8 @@ async def test_load_preserves_state_then_e01_normalizes_to_s00(tmp_path: Path) -
     assert loaded_node.status == SyncStatus.FAILED
     assert loaded_node.binding_status == BindingStatus.WARNING
     assert loaded_node.error == "last run failed"
+    assert loaded_node.sync_log is not None
+    assert "加载来源: persistence 恢复节点" in loaded_node.sync_log
 
     # E01 后归并到 S00，并清空历史 error
     runtime = StateMachineRuntime(StateMachineConfig.load(CFG_PATH))
@@ -579,6 +592,114 @@ async def test_load_preserves_state_then_e01_normalizes_to_s00(tmp_path: Path) -
     await persistence.close()
 
 
+@pytest.mark.asyncio
+async def test_persistence_only_unbound_node_is_cleaned_after_bootstrap_refresh_miss(tmp_path: Path) -> None:
+    db = tmp_path / "bootstrap_source_miss_unbound.db"
+
+    persistence = SQLitePersistenceBackend(str(db))
+    await persistence.initialize()
+
+    collection = DataCollection("project-bootstrap-unbound", persistence=persistence, auto_persist=False)
+    node = ProjectSyncNode(
+        node_id="stale-unbound-node",
+        data_id="proj-stale-unbound",
+        data={
+            "id": "proj-stale-unbound",
+            "type": "project",
+            "name": "Stale Unbound Project",
+            "code": "P-ST-UNBOUND",
+            "status": "active",
+            "progress_type": "C",
+            "project_type": "onshore_wind",
+        },
+        action=SyncAction.NONE,
+        status=SyncStatus.PENDING,
+        binding_status=BindingStatus.WARNING,
+    )
+    await collection.add(node)
+    await collection.persist()
+
+    reloaded = DataCollection("project-bootstrap-unbound", persistence=persistence, auto_persist=False)
+    await reloaded.load_from_persistence()
+    loaded_node = reloaded.get("stale-unbound-node")
+    assert loaded_node is not None
+    assert loaded_node.context.get("_loaded_from_persistence") is True
+
+    runtime = StateMachineRuntime(StateMachineConfig.load(CFG_PATH))
+    binding_manager = BindingManager(persistence, auto_persist=False)
+    remote_collection = DataCollection("empty-remote-unbound", persistence=persistence, auto_persist=False)
+    cleaned = await run_phase2_cleanup(
+        node_types=["project"],
+        local_collection=reloaded,
+        remote_collection=remote_collection,
+        binding_manager=binding_manager,
+        runtime=runtime,
+    )
+
+    assert cleaned == 1
+    assert reloaded.get("stale-unbound-node") is None
+
+    await persistence.close()
+
+
+@pytest.mark.asyncio
+async def test_bound_node_is_isolated_after_bootstrap_refresh_miss(tmp_path: Path) -> None:
+    db = tmp_path / "bootstrap_source_miss_bound.db"
+
+    persistence = SQLitePersistenceBackend(str(db))
+    await persistence.initialize()
+
+    collection = DataCollection("project-bootstrap-bound", persistence=persistence, auto_persist=False)
+    node = ProjectSyncNode(
+        node_id="stale-bound-local-node",
+        data_id="proj-stale-bound",
+        data={
+            "id": "proj-stale-bound",
+            "type": "project",
+            "name": "Stale Bound Project",
+            "code": "P-ST-BOUND",
+            "status": "active",
+            "progress_type": "C",
+            "project_type": "onshore_wind",
+        },
+        action=SyncAction.NONE,
+        status=SyncStatus.PENDING,
+        binding_status=BindingStatus.WARNING,
+    )
+    await collection.add(node)
+    await collection.persist()
+
+    reloaded = DataCollection("project-bootstrap-bound", persistence=persistence, auto_persist=False)
+    await reloaded.load_from_persistence()
+    loaded_node = reloaded.get("stale-bound-local-node")
+    assert loaded_node is not None
+    assert loaded_node.context.get("_loaded_from_persistence") is True
+
+    runtime = StateMachineRuntime(StateMachineConfig.load(CFG_PATH))
+    binding_manager = BindingManager(persistence, auto_persist=False)
+    await binding_manager.bind("project", "stale-bound-local-node", "remote-missing-node")
+    remote_collection = DataCollection("empty-remote-bound", persistence=persistence, auto_persist=False)
+    cleaned = await run_phase2_cleanup(
+        node_types=["project"],
+        local_collection=reloaded,
+        remote_collection=remote_collection,
+        binding_manager=binding_manager,
+        runtime=runtime,
+    )
+
+    isolated = reloaded.get("stale-bound-local-node")
+    assert cleaned == 0
+    assert isolated is not None
+    assert reloaded.is_bootstrap_blocked(isolated)
+    assert isolated.binding_status == BindingStatus.ABNORMAL
+    assert isolated.action == SyncAction.NONE
+    assert isolated.status == SyncStatus.FAILED
+    assert "BOOTSTRAP_SOURCE_MISSING" in (isolated.error or "")
+    assert isolated.context.get("bootstrap_block_reason") == "source_missing_after_persistence_refresh"
+
+    await persistence.close()
+
+
 @pytest.mark.asyncio
 async def test_invalid_persisted_enum_is_bootstrap_blocked_and_skipped_in_reset(tmp_path: Path) -> None:
     REPORT_DIR.mkdir(parents=True, exist_ok=True)

tests/unit/test_api_trace_and_load_source_logs.py

@@ -146,3 +146,39 @@ async def test_api_datasource_failed_update_does_not_use_unrelated_op_trace() ->
     assert updated.sync_log is not None
     assert "API链路[submit-failed] PUT /material/plan" not in updated.sync_log
     assert "未找到匹配 biz_id 的请求记录" in updated.sync_log
+
+
+@pytest.mark.asyncio
+async def test_api_datasource_failed_create_uses_push_id_trace_when_data_id_empty() -> None:
+    collection = DataCollection("remote")
+    ds = ApiDataSource(base_url="http://example.com", uid="u", secret="s")
+    ds.set_collection(collection)
+
+    node = _build_node(node_id="n-create-fail", data_id="", name="X")
+    with node.allow_core_state_write("test"):
+        node.binding_status = BindingStatus.NORMAL
+        node.action = SyncAction.CREATE
+        node.status = SyncStatus.IN_PROGRESS
+    await collection.add(node)
+
+    ds.client._record_trace(
+        method="POST",
+        endpoint="/construction/create",
+        url="http://example.com/construction/create",
+        request_params=None,
+        request_payload={"push_id": "push-create-fail", "data": {"task_type": "jxfx"}},
+        response={"code": 400, "message": "施工类型错误"},
+        elapsed_ms=12,
+    )
+
+    await ds._handle_task_result(
+        cast(Any, _Handler()),
+        TaskResult.failed(node_id=node.node_id, error="Create failed: 施工类型错误", push_id="push-create-fail"),
+        {},
+    )
+
+    updated = collection.get(node.node_id)
+    assert updated is not None
+    assert updated.sync_log is not None
+    assert "API链路[submit-failed] POST /construction/create" in updated.sync_log
+    assert "response={\"code\":400,\"message\":\"施工类型错误\"}" in updated.sync_log

tests/unit/test_construction_create_failure_push_id.py

@@ -0,0 +1,54 @@
+from __future__ import annotations
+
+import pytest
+
+from sync_state_machine.domain.construction.api_handler import ConstructionTaskApiHandler
+from sync_state_machine.domain.construction.sync_node import ConstructionTaskSyncNode
+
+
+@pytest.mark.asyncio
+async def test_construction_create_failure_preserves_push_id(monkeypatch: pytest.MonkeyPatch) -> None:
+    handler = ConstructionTaskApiHandler()
+    handler.api_client = object()
+
+    async def _fake_create(client, data, push_id: str, *, steps=None):
+        del client, data, steps
+        raise Exception(f"API error for {push_id}")
+
+    monkeypatch.setattr(
+        "sync_state_machine.domain.construction.api_handler.api_create_construction_task",
+        _fake_create,
+    )
+
+    node = ConstructionTaskSyncNode(
+        node_id="n1",
+        data_id="",
+        data={
+            "id": "",
+            "project_id": "p1",
+            "contract_id": "c1",
+            "show_name": "架线工程",
+            "construction_section": "SC",
+            "status": -1,
+            "task_type": "jxfx",
+            "plan_start_date": "2024-06-18",
+            "plan_complete_date": "2024-12-05",
+            "actual_complete_date": "2025-12-29",
+            "complete_quantity": 2.0,
+            "complete_rate": 100.0,
+            "contract_quantity": 2.0,
+            "plan_node": [],
+            "statistic_date": "2025-12-29",
+            "is_crucial": False,
+            "delay_days": None,
+        },
+    )
+    node.context["project_id"] = "p1"
+
+    results = await handler.create_all([node])
+
+    assert len(results) == 1
+    result = results[0]
+    assert result.status.value == "failed"
+    assert result.push_id is not None
+    assert result.error == f"API error for {result.push_id}"

tests/unit/test_e01_bootstrap_matrix.py

@@ -52,7 +52,13 @@ def test_e01_bootstrap_accepts_all_combinations(action: SyncAction, status: Sync
     original_error = node.error
 
     is_zombie = _is_create_zombie(node)
-    decision = e01_bootstrap(runtime, node=node, is_create_zombie=is_zombie)
+    decision = e01_bootstrap(
+        runtime,
+        node=node,
+        is_create_zombie=is_zombie,
+        bootstrap_source_present=True,
+        has_binding_record=False,
+    )
 
     if is_zombie:
         assert decision.to_state == "S15"
@@ -66,3 +72,48 @@ def test_e01_bootstrap_accepts_all_combinations(action: SyncAction, status: Sync
         assert node.action == SyncAction.NONE
         assert node.status == SyncStatus.PENDING
         assert node.error is None
+
+
+def test_e01_bootstrap_cleans_unbound_persistence_only_node_on_source_miss() -> None:
+    runtime = StateMachineRuntime(StateMachineConfig.load(CFG_PATH))
+    node = _new_node(
+        node_id="e01-stale-unbound",
+        action=SyncAction.NONE,
+        status=SyncStatus.PENDING,
+        data_id="id-1",
+    )
+
+    decision = e01_bootstrap(
+        runtime,
+        node=node,
+        is_create_zombie=False,
+        bootstrap_source_present=False,
+        has_binding_record=False,
+    )
+
+    assert decision.to_state == "S15"
+
+
+def test_e01_bootstrap_isolates_bound_node_on_source_miss() -> None:
+    runtime = StateMachineRuntime(StateMachineConfig.load(CFG_PATH))
+    node = _new_node(
+        node_id="e01-stale-bound",
+        action=SyncAction.NONE,
+        status=SyncStatus.PENDING,
+        data_id="id-1",
+    )
+
+    decision = e01_bootstrap(
+        runtime,
+        node=node,
+        is_create_zombie=False,
+        bootstrap_source_present=False,
+        has_binding_record=True,
+    )
+
+    assert decision.to_state == "S17"
+    assert node.binding_status == BindingStatus.ABNORMAL
+    assert node.action == SyncAction.NONE
+    assert node.status == SyncStatus.FAILED
+    assert node.context["bootstrap_blocked"] is True
+    assert "BOOTSTRAP_SOURCE_MISSING" in (node.error or "")

tests/unit/test_full_sync_pipeline_downstream_reload.py

@@ -7,7 +7,10 @@ import pytest
 
 from sync_state_machine.common.binding import BindingManager
 from sync_state_machine.common.collection import DataCollection
+from sync_state_machine.common.sqlite_backend import SQLitePersistenceBackend
+from sync_state_machine.common.types import BindingStatus, SyncAction, SyncStatus
 from sync_state_machine.datasource.jsonl import JsonlDataSource
+from sync_state_machine.domain.project.sync_node import ProjectSyncNode
 import sync_state_machine.pipeline.full_sync_pipeline as full_sync_pipeline_module
 from sync_state_machine.pipeline.full_sync_pipeline import FullSyncPipeline, NodeTypeLoadError, WriteOutcome
 
@@ -30,6 +33,37 @@ class _FakeDataSource:
         return None
 
 
+class _BootstrapRefreshDataSource(_FakeDataSource):
+    def __init__(self, *, node_data: dict[str, str]) -> None:
+        self.node_data = dict(node_data)
+
+    async def load_all(self, order=None, data_type=None, raise_on_error=False) -> None:
+        del data_type, raise_on_error
+        if self.collection is None:
+            raise RuntimeError("collection not set")
+        if not order or tuple(order) != ("project",):
+            return
+
+        existing = self.collection.get_by_data_id("project", self.node_data["id"])
+        if existing is None:
+            node = ProjectSyncNode(
+                node_id="bootstrap-refreshed-project",
+                data_id=self.node_data["id"],
+                data=self.node_data,
+            )
+            node.append_log(f"加载来源: {self.__class__.__name__}.load 新增节点")
+            await self.collection.add(node)
+            return
+
+        existing.set_data(self.node_data)
+        existing.set_origin_data(self.node_data)
+        if existing.context.pop("_loaded_from_persistence", False):
+            existing.append_log(f"加载来源: persistence -> {self.__class__.__name__}.load 覆盖刷新")
+        else:
+            existing.append_log(f"加载来源: {self.__class__.__name__}.load 覆盖刷新")
+        await self.collection.update(existing)
+
+
 class _StubStrategy:
     def __init__(self, node_type: str):
         self.node_type = node_type
@@ -349,3 +383,72 @@ async def test_run_persists_before_reraising_load_failure(monkeypatch: pytest.Mo
 
     assert calls == ["bootstrap", "sync", "persist"]
     assert persistence.close_calls == 1
+
+
+@pytest.mark.asyncio
+async def test_phase_bootstrap_refreshes_persisted_nodes_before_e01_cleanup(tmp_path: Path) -> None:
+    db_path = tmp_path / "bootstrap-refresh.db"
+    persistence = SQLitePersistenceBackend(str(db_path))
+    await persistence.initialize()
+
+    seed_remote = DataCollection("remote", persistence=persistence, auto_persist=False)
+    stale_node = ProjectSyncNode(
+        node_id="persisted-project-node",
+        data_id="proj-bootstrap-1",
+        data={
+            "id": "proj-bootstrap-1",
+            "type": "project",
+            "name": "Persisted Project",
+            "code": "P-BOOT",
+            "status": "active",
+            "progress_type": "C",
+            "project_type": "onshore_wind",
+        },
+        action=SyncAction.NONE,
+        status=SyncStatus.PENDING,
+        binding_status=BindingStatus.WARNING,
+    )
+    await seed_remote.add(stale_node)
+    await seed_remote.persist()
+
+    local_collection = DataCollection("local", persistence=persistence, auto_persist=False)
+    remote_collection = DataCollection("remote", persistence=persistence, auto_persist=False)
+    binding_manager = BindingManager(persistence, auto_persist=False)
+
+    pipeline = FullSyncPipeline(
+        local_datasource=_FakeDataSource(),
+        remote_datasource=_BootstrapRefreshDataSource(
+            node_data={
+                "id": "proj-bootstrap-1",
+                "type": "project",
+                "name": "Refreshed Project",
+                "code": "P-BOOT",
+                "status": "active",
+                "progress_type": "C",
+                "project_type": "onshore_wind",
+            }
+        ),
+        strategies=[_StubStrategy("project")],
+        persistence=persistence,
+        local_collection=local_collection,
+        remote_collection=remote_collection,
+        binding_manager=binding_manager,
+        node_types=["project"],
+        load_order=["project"],
+        sync_order=["project"],
+        enable_persistence=True,
+    )
+
+    await pipeline.phase_bootstrap()
+
+    refreshed = remote_collection.get("persisted-project-node")
+    assert refreshed is not None
+    assert refreshed.binding_status == BindingStatus.UNCHECKED
+    assert refreshed.action == SyncAction.NONE
+    assert refreshed.status == SyncStatus.PENDING
+    assert remote_collection.is_bootstrap_blocked(refreshed) is False
+    assert refreshed.sync_log is not None
+    assert "加载来源: persistence 恢复节点" in refreshed.sync_log
+    assert "加载来源: persistence -> _BootstrapRefreshDataSource.load 覆盖刷新" in refreshed.sync_log
+
+    await pipeline.close()

tests/unit/test_scope_persistence_partitioning.py

@@ -56,3 +56,21 @@ async def test_scope_partitioning_and_copy_between_scopes(tmp_path: Path) -> Non
     assert len(reloaded_global.filter(node_type="test_project")) == 1
 
     await persistence.close()
+
+
+@pytest.mark.asyncio
+async def test_sqlite_backend_creates_restore_join_index(tmp_path: Path) -> None:
+    db_path = tmp_path / "scope_partitioning.db"
+    persistence = SQLitePersistenceBackend(str(db_path))
+    await persistence.initialize()
+
+    columns, rows = PersistenceBackend.query_records(
+        backend="sqlite",
+        db_path=str(db_path),
+        sql="PRAGMA index_list('nodes')",
+    )
+
+    assert columns == ["seq", "name", "unique", "origin", "partial"]
+    assert any(row["name"] == "idx_nodes_scope_type_node_id" for row in rows)
+
+    await persistence.close()