generate a short context cost after every N queries to how much context you are eating up