Will the RLVR mechanism be improved upon or is it in some sense optimal?