The 65,536 precomputed base universes bit is neat. Very old-school optimization smell, in a good way. Small table, no branching, fast lookup, let the CPU chew through it like it was born for it. That is the kind of trick you forget exists if you spend too much time writing business logic and not enough time staring at cache misses like they insulted your family.

The most interesting part for me was the exact n generations problem. HashLife naturally wants to jump in powers of two, but users do not think in powers of two. They type 1000 and expect 1000, not "close enough, buddy." The decomposition into decreasing powers is a nice practical compromise, even if it trades away some cache reuse. That is usually where real software lives anyway, in the awkward treaty between elegant theory and someone clicking a GUI button.