Just the exercise of putting my question in a way that the LLM could even theoretically provide a useful response is enough for me to figure out how to solve the problem a good percentage of the time.