diff --git a/backend/agenteval/storage/repository.py b/backend/agenteval/storage/repository.py
index 442714c..8ffddc9 100644
--- a/backend/agenteval/storage/repository.py
+++ b/backend/agenteval/storage/repository.py
@@ -277,6 +277,25 @@ class RunRepository:
         self.session.refresh(db)
         return _run_from_db(db)
 
+    def mark_orphans_failed(self) -> int:
+        """服务启动时清理：把遗留的 running/pending 运行标记为 failed。
+
+        评测任务是进程内 asyncio 任务，服务重启后不会恢复；不清理则这些
+        运行永远停留在 running（僵尸运行）。
+        """
+        statement = select(EvalRunDB).where(EvalRunDB.status.in_(["running", "pending"]))  # type: ignore[attr-defined]
+        orphans = self.session.exec(statement).all()
+        for db in orphans:
+            db.status = "failed"
+            db.completed_at = db.completed_at or utc_now()
+            summary = db.get_summary() or {}
+            summary["error"] = {"code": "interrupted", "message": "服务重启导致评测中断"}
+            db.set_summary(summary)
+            self.session.add(db)
+        if orphans:
+            self.session.commit()
+        return len(orphans)
+
     def update(self, run: EvalRun) -> Optional[EvalRun]:
         existing = self.session.get(EvalRunDB, run.id)
         if not existing:
diff --git a/backend/agenteval/web/app.py b/backend/agenteval/web/app.py
index 5435410..3f17752 100644
--- a/backend/agenteval/web/app.py
+++ b/backend/agenteval/web/app.py
@@ -1,5 +1,6 @@
 """FastAPI web backend for AgentEvalTool."""
 
+import logging
 from contextlib import asynccontextmanager
 from pathlib import Path
 
@@ -8,7 +9,8 @@ from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import FileResponse, JSONResponse
 
 from agenteval.config import get_settings
-from agenteval.storage.db import init_db
+from agenteval.storage.db import get_session, init_db
+from agenteval.storage.repository import RunRepository
 from agenteval.version import get_build_info, get_version
 from agenteval.web.deps import require_api_key
 from agenteval.web.routers import auth, files, model_configs, proxy, reports, runs, scenarios, stats, targets
@@ -18,6 +20,17 @@ from agenteval.web.websocket import ws_manager
 @asynccontextmanager
 async def lifespan(_: FastAPI):
     init_db()
+    # 评测任务是进程内 asyncio 任务，重启后不会恢复——清理僵尸运行（尽力而为，不阻断启动）
+    try:
+        session = get_session()
+        try:
+            count = RunRepository(session).mark_orphans_failed()
+            if count:
+                logging.getLogger("agenteval").warning("启动清理：%d 个中断的运行已标记为 failed", count)
+        finally:
+            session.close()
+    except Exception as exc:
+        logging.getLogger("agenteval").warning("启动清理失败（忽略）: %s", exc)
     yield
 
 
diff --git a/frontend/web/src/pages/Home.tsx b/frontend/web/src/pages/Home.tsx
index 060972b..048a842 100644
--- a/frontend/web/src/pages/Home.tsx
+++ b/frontend/web/src/pages/Home.tsx
@@ -216,14 +216,14 @@ function RecentRunRow({ run, onOpen }: { run: Run; onOpen: () => void }) {
         <span style={{ fontWeight: 500 }}>{run.scenario_name || run.scenario_id.slice(0, 8)}</span>
         <span style={{ color: colors.textMuted }}> · {run.target_name || run.target_id.slice(0, 8)}</span>
       </div>
+      <Tag color={triggerColors[trigger] ?? 'default'} style={{ margin: 0, fontSize: 10, lineHeight: '16px', padding: '0 4px' }}>
+        {triggerLabels[trigger] ?? trigger}
+      </Tag>
       {run.scenario_version != null && (
         <Tag style={{ margin: 0, fontSize: 10, lineHeight: '16px', padding: '0 4px' }}>
           v{run.scenario_version}
         </Tag>
       )}
-      <Tag color={triggerColors[trigger] ?? 'default'} style={{ margin: 0, fontSize: 10, lineHeight: '16px', padding: '0 4px' }}>
-        {triggerLabels[trigger] ?? trigger}
-      </Tag>
       <span style={{ fontSize: 12, color: dotColor, width: 48, flexShrink: 0 }}>{statusLabels[run.status] ?? run.status}</span>
       <span style={{
         fontSize: 12, fontWeight: 600, width: 44, textAlign: 'right', flexShrink: 0,
diff --git a/tests/unit/test_orphan_runs.py b/tests/unit/test_orphan_runs.py
new file mode 100644
index 0000000..a665fae
--- /dev/null
+++ b/tests/unit/test_orphan_runs.py
@@ -0,0 +1,38 @@
+"""Startup cleanup of orphan runs (interrupted by server restart)."""
+
+from agenteval.models import EvalRun, RunStatus
+from agenteval.storage.repository import RunRepository
+
+
+def _make_run(session, status: RunStatus) -> str:
+    run = RunRepository(session).create(EvalRun(
+        target_id="t-1", scenario_id="s-1", status=status,
+    ))
+    return run.id
+
+
+def test_mark_orphans_failed(db_session):
+    repo = RunRepository(db_session)
+    running_id = _make_run(db_session, RunStatus.RUNNING)
+    pending_id = _make_run(db_session, RunStatus.PENDING)
+    completed_id = _make_run(db_session, RunStatus.COMPLETED)
+    failed_id = _make_run(db_session, RunStatus.FAILED)
+
+    count = repo.mark_orphans_failed()
+
+    assert count == 2
+    for rid in (running_id, pending_id):
+        run = repo.get(rid)
+        assert run.status == RunStatus.FAILED
+        assert run.summary["error"]["code"] == "interrupted"
+        assert run.completed_at is not None
+    # 已完结的运行不受影响
+    assert repo.get(completed_id).status == RunStatus.COMPLETED
+    assert repo.get(failed_id).status == RunStatus.FAILED
+    assert repo.get(completed_id).summary is None
+
+
+def test_mark_orphans_failed_noop_when_clean(db_session):
+    repo = RunRepository(db_session)
+    _make_run(db_session, RunStatus.COMPLETED)
+    assert repo.mark_orphans_failed() == 0