I use this code base, it replicates, and is a low overhead environment to study reward hacking - which means it speed up research iterations.