Draft models cannot see a guard fight through a screen

The single most transferable defensive skill for a young guard is one that produces no box score event at all.

Steals are noisy. Blocks belong to somebody else. Defensive rating in college tells you mostly about the four players around the prospect, and every serious model already knows this, which is why the honest ones apply large error bars to defensive projection and move on.

The skill that actually transfers is getting through screens without losing the ball handler, and it is invisible by construction. A guard who slides over the top cleanly produces no event. Nothing is recorded. The possession ends with a contested jumper that gets attributed to the shooter having a bad night, and the guard who caused it collects a zero in every column.

Physically it is a specific package and not a general athletic one. Hip mobility, the willingness to take a shoulder to the chest, and above all a sense of when to go early rather than fast. The best screen fighters do not accelerate through contact, they leave before the contact exists, which means the trait is closer to anticipation than to speed and therefore does not show up in a combine either.

This is why the good college defenders who bust are so often the tall, long, event-generating ones, and why unremarkable guards from unfashionable programmes occasionally become defensive starters on good teams. The model was measuring what could be measured.

If I were building the evaluation, I would take a hundred clips per prospect of nothing but ball screen defense, tag the outcome of each, and treat the resulting rate as a real number rather than an eye-test note. That is expensive manual work and it is exactly the sort of thing clubs pay analysts to avoid doing.

There is a second invisible skill that travels just as well, which is the willingness to accept a screen on the strong side rather than dying under it. Some young defenders fight everything and lose track of the ball for two seconds each time. Others take the hit, stay attached, and concede two feet of space they can win back. Nothing in the data separates those two players, and they are not the same player at all.

The scouting report says he competes. What it means, if the scout is any good, is that he arrives before the screen does.