I'm honestly relieved to see that I'm not the only one to test optimized/newer implementations against less optimized/older implementations. It's what I came up with after thinking about it for quite a while and I've been wondering why so little other software seems to do that. In my case, it's about serialization code. My test compare self-reported reader / writer states and serialized data.

Ladybird goes further by also running the engines in parallel in debug builds during normal use. They have a way to extract test cases from any bugs found that way.