And the bug was quite silly -- it does a simple standard thing (remove obvious symmetries) in a more complicated way (at the end of a long computation, instead of at the start), incorrectly.

Always good to test code with and without such optimizations to verify the results are the same. That's not always practical though for large problems and optimizations that yield significant time savings.

As the saying goes, there are only two hard problems in computer science: naming things, cache invalidation, and off by one errors.