That’s interesting.
Okay, so the first thing is these cutpoints are not on the outcome scale, they are on the logistic scale, so setting them linearly apart may be bad.
Another thing is you may want to define the model on different quantities, for instance total points and ratio of home/away.
And even ignoring the clumping may be fine. Depends on your use case, if it’s to predict actual outcomes it’s not great, but if you are interested in latent quantities it may be fine.