Every player in his first three NBA seasons who is playing, but not playing much. Where he sits on the two things that predict this, and the fourteen careers closest to his.
Every sentence on a claim page is a count. It is either true or it is a bug, and it shows you the full list of everyone who qualifies. Nothing here works that way. These are comparisons, and a comparison can be honest and still be wrong about a particular player.
Three ways to pick the fringe seasons most likely to turn into a rotation role, measured on the players whose careers have already played out. The last column is the one that matters: shuffle the outcomes at random and run the same pick again, times, and count how many of those runs did as well as the real thing.
The second row is not a second piece of evidence. Age, minutes and games played sit inside the profile distance, so ranking on the profile is partly re-measuring the row above it. The only way to find out whether the box score adds anything of its own is to hold age and games fixed and look at what is left over.
Split the cohort into four bands by age and games played. Inside each band, take the players whose comps did best and the players whose comps did worst, and compare what actually happened to them. If the box-score profile knows something that age and games do not, the best-profile group should beat the worst-profile group in every band.
Grouped by where the player sits on age and games, because that is the part that predicts. The two are combined into one score, so a slightly older player who already gets a lot of games can land in the same band as a very young one who does not. Inside a group the order is alphabetical: ordering players by their comps would be exactly the ranking the section above says does not hold up.