diff --git a/backend/app/engine/judge.py b/backend/app/engine/judge.py
index b43d2f87..d4c3903d 100644
--- a/backend/app/engine/judge.py
+++ b/backend/app/engine/judge.py
@@ -1070,6 +1070,10 @@ def _items_of(value: Any) -> list[Any] | None:
     return value if isinstance(value, list) else None
 
 
+#: _ask 交回的没成的原因：输出到了上限被截断
+_TRUNCATED = "truncated"
+
+
 async def _ask(model: Any, messages: list[BaseMessage]) -> tuple[list[Any] | None, list[BaseMessage], str | None,
                                                                   int]:
     """问一次裁判：(判定清单, 模型的原始回复, 没成的原因, 调了几次模型)。
@@ -1111,10 +1115,6 @@ async def _ask(model: Any, messages: list[BaseMessage]) -> tuple[list[Any] | Non
     return items, raws, None if items is not None else "format", 2
 
 
-#: _ask 交回的没成的原因：输出到了上限被截断
-_TRUNCATED = "truncated"
-
-
 def _spent(error: BaseException) -> list[BaseMessage]:
     """截断时 openai 的异常带着这次的用量：折成一条空回复交给记账（额度是真花出去了，每日上限不能漏掉它）。"""
     inp, out = error_usage(error)
diff --git a/backend/app/engine/verify_agent.py b/backend/app/engine/verify_agent.py
index ad5b59a7..022706c9 100644
--- a/backend/app/engine/verify_agent.py
+++ b/backend/app/engine/verify_agent.py
@@ -963,8 +963,8 @@ async def _verify(state: _State, *, spec: ModelSpec, system: str, task: str, sou
     declared = list((resolved.get("properties") or {}) if _is_object(resolved) else {})
     if declared and not any(k in parsed for k in declared):
         # 一个要求的字段都没交：不是「查不到」，是没按结构交。不这么分，字段名对不上的回复会被当成什么都没查到
-        return await give_up("核查模型没有按结构交结果（交回的内容里没有要求的字段："
-                             + "、".join(declared[:8]) + ("等" if len(declared) > 8 else "") + "）")
+        return await give_up("核查模型没有按结构交结果（交回的内容里没有要求的字段：" + "、".join(declared[:8])
+                             + ("等" if len(declared) > 8 else "") + "）")
 
     # ---- 取值、组装理由
     cells = _Cells(queries)