Curious, why can't you provide a measurement of context size for a human. Surely there must be enough science to make a good approximater.