The more complexity you add the less likely the LLM is to give you a valid response in one shot. It’s still going to be easier to get the LLM to supply values to a fixed scheme than to get the LLM to give the answers and the scheme