I disagree that non-toxic negative reinforcement learning is not helpful. I agree that we could have more positive reinforcement learning. I tried to accomplish this with the top town, top scum, top improved leaderboard idea, but that was shot down. It could admittedly lead to an even more unhealthily competitive atmosphere, but meeh I still think it could have worked out.
If someone has bad reads, then that actively involves your win condition in mafia, because you have to get them to trust your reads and direction to attain majority, so there is always a drive to correct them. The way how you go about that can differ, but yes you will be challenged and pressured on your viewpoints. That’s literally part of social deduction.