- Mark harvested files as Scanned until vocab write finishes - Prompt and normalize speakers into short title-case nameplates - Keep speaker finalize chunking on the Settings batch size