diff --git a/backend/app/api/knowledge.py b/backend/app/api/knowledge.py
index 98eecae..c035ce5 100644
--- a/backend/app/api/knowledge.py
+++ b/backend/app/api/knowledge.py
@@ -328,7 +328,7 @@ async def list_knowledge(
         stmt = stmt.where(
             or_(
                 Item.updated_at < cur.created_at,
-                and_(Item.updated_at == cur.created_at, Item.id < cur.item_id),
+                and_(Item.updated_at == cur.created_at, Item.id < cur.id),
             )
         )
 
diff --git a/backend/tests/test_knowledge.py b/backend/tests/test_knowledge.py
index 76cdebc..568eb87 100644
--- a/backend/tests/test_knowledge.py
+++ b/backend/tests/test_knowledge.py
@@ -381,3 +381,45 @@ async def test_draft_timeline_reads_like_a_story(client: AsyncClient) -> None:
     body = doc["body_md"]
     assert "待办 → 进行中" in body
     assert "进行中 → 已完成" in body
+
+
+@pytest.mark.asyncio
+async def test_list_pages_through_every_document(client: AsyncClient) -> None:
+    """翻页要能翻到底，且不重不漏。
+
+    这条用例是有来历的：知识列表的游标解出来是 `Cursor(created_at, id)`，
+    查询里却写成了 `cur.item_id`——属性不存在，第二页必 500。
+    库小于一页（默认 30 条）时永远走不到这条路径，所以一直没暴露，
+    直到把旧备忘导进来 84 条，用户点「加载更多」就看到"服务器内部错误"。
+    """
+    owner = await login(client)
+    ws = await create_workspace(client, owner, "翻页")
+    titles = [f"备忘 {i:02d}" for i in range(25)]
+    for title in titles:
+        unwrap(
+            await client.post(
+                f"{V1}/workspaces/{ws['id']}/knowledge",
+                json={"title": title},
+                headers=owner["headers"],
+            )
+        )
+
+    seen: list[str] = []
+    cursor: str | None = None
+    for _ in range(10):  # 25 条 / 每页 10 条，三轮就该到底，多留几次余地
+        params: dict[str, Any] = {"limit": 10}
+        if cursor:
+            params["cursor"] = cursor
+        page = unwrap(
+            await client.get(
+                f"{V1}/workspaces/{ws['id']}/knowledge", params=params, headers=owner["headers"]
+            )
+        )
+        seen.extend(i["title"] for i in page["items"])
+        cursor = page["next_cursor"]
+        if not page["has_more"] or not cursor:
+            break
+
+    assert cursor is None
+    assert sorted(seen) == titles
+    assert len(seen) == len(set(seen)), "翻页翻出了重复条目"