We replayed last season blind. The model learned only from seasons through 2024-25, projected every returning player's 2025-26 line from his 2024-25 season, and was then checked against what actually happened.
Everything predicted before the season, compared with the simplest honest guess: that a player repeats last season's line. Error is the average miss, per game.
The same players scored at the minutes they actually played, which isolates the part of the model that is about the player: how each stat develops from one season to the next and how much of a hot or cold year carries over. "Same per minute" repeats last season's rates at this season's minutes.