mempalace_check_duplicate misses clearly existing content at its default threshold of 0.9.

I tested it against text that already exists in the palace, including exact sentences copied from indexed content, and still got:

{
"is_duplicate": false,
"matches": []
}

Example 1:
马丁·海德格尔（Martin Heidegger）出生于德国巴登——符腾堡（Baden-Württemberg）梅斯基尔希的一个贫寒的天主教家庭中。

This only returned true after lowering the threshold to 0.4.

Example 2:
本标准使用重新起草法参考 ISO 690:2010(E)《信息和文献 参考文献和信息资源引用指南》编制，与 ISO 690:2010 的一致性程度为非等效。

This only returned true after lowering the threshold to 0.15.

So on real indexed content, the default 0.9 appears too high to detect even exact existing text.