@jcoglan but really, statistics should verify your implementation, not comparison with my potentially-also-broken version