diff --git a/apps/api/src/services/incident_service.py b/apps/api/src/services/incident_service.py
index 8d3aabc4..d1290b23 100644
--- a/apps/api/src/services/incident_service.py
+++ b/apps/api/src/services/incident_service.py
@@ -1168,12 +1168,14 @@ class IncidentService:
         from src.utils.timezone import now_taipei
 
         # 1. 從 Working Memory 讀取；Redis TTL 過期時退回 PostgreSQL。
+        hydrated_from_episodic = False
         incident = await self.get_from_working_memory(incident_id)
         if incident is None:
             incident = await self.get_from_episodic_memory(incident_id)
             if incident is None:
                 logger.warning("incident_not_found_for_resolve", incident_id=incident_id)
                 return None
+            hydrated_from_episodic = True
             logger.info(
                 "incident_resolve_hydrated_from_episodic_memory",
                 incident_id=incident_id,
@@ -1197,11 +1199,15 @@ class IncidentService:
         incident.resolved_at = now_taipei()
         incident.updated_at = now_taipei()
 
-        # 3. 寫入 Working Memory
-        redis_success = await self.save_to_working_memory(incident)
-        if not redis_success:
-            logger.error("resolve_redis_write_failed", incident_id=incident_id)
-            return None
+        # 3. 寫入 Working Memory。Redis TTL 已過的歷史 DB-only 事件只更新 DB，
+        # 不重新灌回 Redis working memory，避免舊事件回流成 active workload。
+        if hydrated_from_episodic:
+            logger.info("resolve_db_only_incident", incident_id=incident_id)
+        else:
+            redis_success = await self.save_to_working_memory(incident)
+            if not redis_success:
+                logger.error("resolve_redis_write_failed", incident_id=incident_id)
+                return None
 
         # 4. 同步到 Episodic Memory
         try:
diff --git a/apps/api/tests/test_incident_service_resolve_idempotency.py b/apps/api/tests/test_incident_service_resolve_idempotency.py
index 0c86cea6..bffd08b1 100644
--- a/apps/api/tests/test_incident_service_resolve_idempotency.py
+++ b/apps/api/tests/test_incident_service_resolve_idempotency.py
@@ -12,6 +12,7 @@ test_incident_service_resolve_idempotency
 重新放大「resolve_incident 重複觸發 postmortem 洗版」的舊風險。
 """
 
+import inspect
 from datetime import UTC, datetime
 from types import SimpleNamespace
 from unittest.mock import AsyncMock
@@ -27,6 +28,42 @@ from src.services.incident_service import (
 )
 
 
+def _patch_resolve_side_effects(monkeypatch):
+    """讓 active incident resolve 測試只驗證 resolve contract，不真的觸發下游工作。"""
+    import asyncio
+
+    import src.services.anomaly_counter as anomaly_counter
+    import src.services.km_conversion_service as km_conversion_service
+    import src.services.knowledge_extractor_service as knowledge_extractor_service
+
+    def close_create_task(coro):
+        if inspect.iscoroutine(coro):
+            coro.close()
+        return SimpleNamespace()
+
+    monkeypatch.setattr(asyncio, "create_task", close_create_task)
+    monkeypatch.setattr(
+        knowledge_extractor_service,
+        "get_knowledge_extractor",
+        lambda: SimpleNamespace(extract_from_incident=AsyncMock(return_value=None)),
+    )
+    monkeypatch.setattr(
+        km_conversion_service,
+        "get_km_conversion_service",
+        lambda: SimpleNamespace(convert=AsyncMock(return_value=None)),
+    )
+    monkeypatch.setattr(
+        anomaly_counter,
+        "AnomalyCounter",
+        SimpleNamespace(derive_key_from_incident=staticmethod(lambda _incident: None)),
+    )
+    monkeypatch.setattr(
+        anomaly_counter,
+        "get_anomaly_counter",
+        lambda: SimpleNamespace(),
+    )
+
+
 @pytest.mark.asyncio
 async def test_resolve_incident_skips_when_already_resolved(monkeypatch):
     """RESOLVED 的 incident 重複 resolve 應 idempotent。"""
@@ -53,6 +90,42 @@ async def test_resolve_incident_skips_when_already_resolved(monkeypatch):
     save_mock.assert_not_called()
 
 
+@pytest.mark.asyncio
+async def test_resolve_incident_can_close_db_only_active_incident(monkeypatch):
+    """Redis TTL 過期但 DB 仍 active 的舊 incident 必須能收斂，不重新灌回 Redis。"""
+    fake_incident = SimpleNamespace(
+        incident_id="INC-DB-FALLBACK-ACTIVE-001",
+        status=IncidentStatus.INVESTIGATING,
+        resolved_at=None,
+        updated_at=None,
+        signals=[],
+        affected_services=[],
+        decision_chain=None,
+        outcome=None,
+    )
+
+    svc = IncidentService()
+    monkeypatch.setattr(
+        svc, "get_from_working_memory", AsyncMock(return_value=None)
+    )
+    episodic_mock = AsyncMock(return_value=fake_incident)
+    monkeypatch.setattr(svc, "get_from_episodic_memory", episodic_mock)
+    save_mock = AsyncMock(return_value=True)
+    monkeypatch.setattr(svc, "save_to_working_memory", save_mock)
+    _patch_resolve_side_effects(monkeypatch)
+
+    result = await svc.resolve_incident(
+        "INC-DB-FALLBACK-ACTIVE-001",
+        emit_postmortem=False,
+    )
+
+    assert result is fake_incident
+    assert fake_incident.status is IncidentStatus.RESOLVED
+    assert fake_incident.resolved_at is not None
+    episodic_mock.assert_awaited_once_with("INC-DB-FALLBACK-ACTIVE-001")
+    save_mock.assert_not_called()
+
+
 @pytest.mark.asyncio
 async def test_resolve_incident_returns_none_when_not_found(monkeypatch):
     """incident 不存在時 return None。確保 guard 不影響 not-found 路徑。"""