In their quest to find a solution they simply recreate the problem with a slightly better approximation of an actual work sample evaluation. It’s like watching a Newtonian approximation algorithm.