I verify the voice context before I ask a model to write

A small local Quillgrain example: compile a voice pack, verify its scope, and catch a language mismatch before inference.