Draft models cannot see defensive movement yet

College spacing is so different from the professional game that the defensive numbers scouts rely on barely transfer.

Every draft model I have seen handles offence reasonably and defence terribly. That is not because the analysts are lazy. It is because the defensive signal in college basketball is contaminated by the floor it was produced on, and nobody has cleaned it properly.

Start with spacing. A defender in a league where two of the five opponents cannot shoot spends his possessions in a crowded box. His rotations are short, his recoveries are forgiving, and his blocks and steals come from sitting in help positions that would be suicide against a professional lineup. Move the same body into a league where all five men on the floor can shoot and those short rotations become long sprints with hard stops at the end. Some players make that jump. Some do not. The college box score cannot tell you which.

The combine does not rescue you either. The lane agility drill is run cold, on a known path, with no ball and no decision. It measures a cone to cone time. What you actually want to know is whether a player can pick the right man to leave, start late, and still arrive under control. Those are perception problems dressed as athletic ones, and no timed drill separates them.

So models fall back on box score defence, which mostly measures reach and gambling, and on subjective scout grades, which are honest but inconsistent between viewers. Then teams are surprised when a lottery pick spends two years unplayable on that end.

What would improve it is not exotic. Tracking data now exists at the college level for a growing share of games. Build a feature that describes the geometry the prospect actually defended, how far the average rotation was, how many shooters were on the floor, how often his team switched. Then adjust his numbers for that context the same way you adjust a shooting percentage for shot difficulty. It is an unglamorous correction and it would move players up and down the board by ten spots.

The counter argument is that defence is coachable and offence is not, so you should draft the shot and fix the rest. I used to believe that. I believe less of it now, because the thing that fails to transfer is rarely willingness. It is the movement vocabulary, and that gets laid down early.

Draft the player who has already defended a spread floor. He is not guessing about the distances.