Ah, super secret problem domains that have been thoroughly represented in the LLM training data. Nice.