What if I built an OJ that works like chess.com/lichess puzzle training? The issue is that you are encountering new problems one by one, with rating changes based on 100% or partial test case passes. Does it seem you're interested, or is it a bad idea?