This is a very clear comparison of file-based context vs a memory layer. I liked the way it derived the queries into different categories, it makes it easy to understand the metrics.