diff --git a/tests/test_automation_ck_health.py b/tests/test_automation_ck_health.py
index dc58006..f5dce36 100644
--- a/tests/test_automation_ck_health.py
+++ b/tests/test_automation_ck_health.py
@@ -582,3 +582,36 @@ class TestFailureClassification(_Base):
             batch_size=200,
             concurrency=3,
         )
+
+
+class TestOrphanRunCleanup(_Base):
+    """回归：进程重启后，卡在 running 的运行记录必须被收尾。
+
+    作业在调度线程里同步跑，进程一重启那轮就没机会写回结果，记录会永远停在
+    running，页面上一直显示「执行中」。登录任务早有孤儿清理，运行记录原先漏了
+    —— 线上部署时打断了一轮，run 14 就是这么卡住的。
+    """
+
+    def test_running_runs_are_closed_on_startup(self):
+        from web.backend.services.automation.scheduler import cleanup_orphan_runs
+
+        stale = AutomationRun(
+            job_key="ck_relogin", started_at=datetime.now(UTC), status="running"
+        )
+        done = AutomationRun(
+            job_key="ck_health",
+            started_at=datetime.now(UTC),
+            finished_at=datetime.now(UTC),
+            status="success",
+        )
+        self.session.add_all([stale, done])
+        self.session.commit()
+
+        cleaned = cleanup_orphan_runs(self.session)
+        self.session.refresh(stale)
+        self.session.refresh(done)
+        assert cleaned == 1
+        assert stale.status == "failed"
+        assert stale.finished_at is not None
+        assert stale.message == "服务重启中断"
+        assert done.status == "success"  # 已完成的不受影响
diff --git a/web/backend/main.py b/web/backend/main.py
index 7dc8ce6..2933a47 100644
--- a/web/backend/main.py
+++ b/web/backend/main.py
@@ -66,6 +66,11 @@ async def lifespan(app: FastAPI):
         cleaned_relogin = cleanup_orphan_relogin_tasks(db)
         if cleaned_relogin:
             logger.info(f"启动清理 CK 重登残留任务: {cleaned_relogin} 条")
+        from .services.automation.scheduler import cleanup_orphan_runs
+
+        cleaned_runs = cleanup_orphan_runs(db)
+        if cleaned_runs:
+            logger.info(f"启动收尾中断的自动化运行记录: {cleaned_runs} 条")
         cleaned_login = cleanup_orphan_login_tasks(db)
         if cleaned_login:
             logger.info(f"启动清理登录残留任务: {cleaned_login} 条")
diff --git a/web/backend/services/automation/scheduler.py b/web/backend/services/automation/scheduler.py
index aea822b..ca9326d 100644
--- a/web/backend/services/automation/scheduler.py
+++ b/web/backend/services/automation/scheduler.py
@@ -118,6 +118,26 @@ def _due(job: AutomationJob, now: datetime) -> bool:
     return next_run_at is None or next_run_at <= now
 
 
+def cleanup_orphan_runs(db: Session, message: str = "服务重启中断") -> int:
+    """把上次进程留下的 ``running`` 运行记录收尾。
+
+    作业是在调度线程里同步跑的：进程一重启（部署、崩溃、手工重启），那轮作业就
+    再没机会写回结果，``automation_runs`` 会永远停在 ``running``，页面上一直显示
+    「执行中」。登录任务有自己的孤儿清理，运行记录原先漏了。
+
+    收尾为 ``failed`` 而不是 ``success``：那轮确实没跑完，如实记录；被中断的账号
+    由各自的孤儿清理负责标记，并会按退避重试。
+    """
+    orphans = db.query(AutomationRun).filter(AutomationRun.status == "running").all()
+    for run in orphans:
+        run.status = "failed"
+        run.finished_at = datetime.now(UTC)
+        run.message = message
+    if orphans:
+        db.commit()
+    return len(orphans)
+
+
 def run_due_jobs(db: Session, *, now: datetime | None = None) -> list[dict[str, Any]]:
     """跑一轮「到点的作业」，返回每轮结果。