While were at it, add option to exclude negative correlations. In particular, should exclude 1) RB-WR/TE from same team, and 2) RB and opposing RB in same game.

Game flow. Usually RBs/running plays are used more by the winning team- you're ahead, you want to control the clock/time of possession, keep the ball away from your opponent. If you're a losing team good chance you want quick points, more passes, take more shots downfield- running can be less reliable.

Obviously there are exceptions but it's a generally something you want to avoid.

I have tried the extreme (max unique players) for MLB and NFL. It is hard to find a balance since it really depends on the games in the slate. If you go too high you end up with very unique lineups but they have a very good chance to have players get 0's and if you go too low its the too many eggs in one basket effect.

I usually waver between 3 unique players and 7 unique players (7 when its a full slate of 10+ games)

To supplement this I also customize my player pool by either using the 'thumbs up' feature or by eliminating players I do not like. If you are a player that like to bring the player pool down to 25 - 30 players, going as high as 7 or 8 unique players actually works well to make completely different lineups. This will of-course only generate 3 to 7 lineups which can be used as base lineups that are completely unique that can be tweaked to your liking.

sorry this is probably dumb. I'm more of an NBA guy so why can't 2 opposing RB each have good games?

It makes sense in theory to not play opposing RBs but opposing RBs do perform well against each other (week 3 randle vs freeman), so it might not be something I want to use because I play a small number of lineups, but if you're building a lot of lineups automatically, you won't be able to analyze every lineup so you set up constraints that will work most of the time.