CSPs and performance of all-different variants

I've got a CSP solver powering things like a sudoku solver and a crossword builder on my site (https://dawnofthe.dad/). Recently I've been playing with NxN sudokus and created https://dawnofthe.dad/ndoku to try out a few algorithms/implementations of the all-different constraint, which is at the heart of sudokus. To expand on that, I decided to evaluate the performance of the various implementations of the all-different constraint in a somewhat more rigorous fashion, and this is what this article is about.