This only works as "proof" up until someone innovates an "authenticity" flag on the LLM output.
https://github.com/ethel-dev/misspell
https://github.com/ethel-dev/misspell