Comments on https://lichess.org/@/heroku/blog/bandit-chess-development-journey/sI4fjQQm
one small problem, sometimes two moves are indistinguishable so its just a 50/50. Also, the point system is just weird it prefers to sac the exchange rather than win the exchange in a completely winning endgame so uh yeah I dunno how that works, and also the versions of stockfish used might change the scoring.