Calculating rolling cohort retention - with SQL

This kind of trick is great stuff. I don't even know how to categorize it. Well, "data science", of course, but this general kind of algorithmic sleight-of-hand is always attractive.