@aeva This one works fine, but it can be processor-intensive because something needs to look at both sides and figure out what the LCM is, and that's a parallelism-breaker.
Another approach is to have both sides do this, but to a common standard multiple, e.g. if one side is 4 and the other is 8, doesn't matter - they (and every other edge) tessellate up to 64 so they meet. Burns more tris, but it's easy to parallelise.