This is really cool James. I especially like how you compared the primary feature of the cluster against everyone else to sanity check how different your identified clusters were.

I think you mentioned it already but a cool next step would be to have a metric (could be something already available like PER or Win Shares) to look at both the variance in performance of the players within the group, but also to approximate the average level of production/contribution of the average player in a cluster.

Also, it might be interesting to check out whether the distance between clusters means anything (maybe it correlates to traditional positions, but it would be cool if it did not). I would assume that you would want to put together a team where the players are sort of spread apart from each other so that they all have their own specializations (with some doubling down on the most important attributes).

Finally, I would guess that there are multi-cluster players. These multi-talented players might be worth more than the average player due to their versatility (if they are stars) or it might mean they were never able to fully master a skill set or playing style (for the non-stars) lol.

Great blog!