Hey @douzatan thanks for pointing it out. Honestly, the cache is simple right now. Each segment is cached on disk, [keyed on the text + language + voice, so a changed segment only re-synthesizes its own audio / keyed on the source segment, so a manual correction wouldn't invalidate it yet]. The terminology side is looser than it should be. Technical terms are kept in English through the translation prompt and a slang replacement map, not through a real glossary, so changing those rules won't invalidate anything already cached. For on-screen UI names, you'd really want a glossary that's part of the key, so editing one term rebuilds only the segments that use it. Good catch, I'm adding that to the list. Thanks for pushing on it.