Many years ago I released some software with a major algorithm replacement that delivered a roughly 10x speed improvement. Within six months I discovered a bug

Many years ago I released some software with a major algorithm replacement that delivered a roughly 10x speed improvement. Within six months I discovered a bug in the new algorithm whose removal delivered roughly another 10x improvement. (That’s the most dramatic single speed up of a previously well designed algorithm I have delivered in my career.) Numerical algorithm bugs can be tricky to detect when sandwiched between dramatic improvements like that!