> Classifying data by cacheability ... Notice how we’re combining the three types of data in the first iteration?

This is the key insight. But I would phrase it more as "segregate data by mutability" as it's about more than caching. More than once I've seen pain caused by a design that failed to do this.