Would you say it is the context length or both access and management of it? I have my doubts with merely maxing the context length. It looks to me that you need a intelligent context composer to get the most horsepower from a model.