The defensive half of the same pipeline: shot quality allowed and disruption, both measured against the defensive archetype a player actually resembles.
Role assignment, the leave-one-out baselines and the top-two-role blend all work exactly as described in the roles post, and both halves read the same table. What changes here is the peer group — defensive archetypes — and what gets measured against it: the FG% a defender allows, and the possessions they end.
Step 1
A defender is described by where they get shot at, not by how well they defend it. The four defended-shot frequency splits — inside 6 ft, inside 10 ft, beyond 15 ft, and threes — plus box-outs, each converted to a league percentile rank inside that season. Clint Capela taking 41% of his defended attempts inside 5 ft puts him in the 77th percentile of rim-defense frequency; that percentile is the input, not the 41%.
Four templates are defined over those five features:
Step 2
For each of the four defended zones, the expected FG% allowed is the attempt-weighted mean for that role group, blended across the defender’s top two roles — with the defender’s own attempts removed from their own group’s mean first. A big man is not held to a perimeter standard on threes, and a guard is not held to a rim-protection standard inside 6 ft.
Step 3
Where a player defends says nothing about what they take away. Steals and blocks each get their own expectation from the count baseline, in the same defensive-role peer group, and the residuals are weighted before being summed:
| Event | Weight | Why |
|---|---|---|
| Steal | 1.0 | A possession ends. On average a possession is worth about a point. |
| Block | 0.7 | A block does not always end the possession — it can go out of bounds or straight back to the offense. |
Deflections are deliberately left out. They are logged inconsistently from arena to arena, and most of what they capture already shows up inside the steal residual.
Live from role_stats. Sort any column. FG% columns are what opponents shot; the bars encode value in points, split into the two channels it comes from.
| Player | GP | Defensive Role | FG% <6ft | FG% <10ft | FG% >15ft | FG% 3PT | STL | BLK | D Shot | Role Disrupt | DRF ↓ |
|---|
FG% shown ×100. D Shot bars clip at ±4 points, Role Disrupt at ±1. First 400 rows of the current sort. Both channels are per game.
Limitations
Deterrence is invisible. A shot never taken because Wembanyama is standing there does not appear in anyone’s defended-attempt count.
Matchup assignment is credited to the nearest defender, so a player on a strong defensive team inherits some of the team’s shot suppression.
The role templates describe where a player defends, which is partly a coaching decision. A defender miscast by their team is measured against the archetype they were cast as.
Strengths
Role-adjusted and volume-weighted at once: the standard changes with where you are asked to defend, and the reward scales with how often you are asked.
Two-dimensional — D Shot and Role Disrupt are reported separately, so you can see whether a defender’s value comes from suppressing shots or from ending possessions.
Single-season and leave-one-out throughout, so no defender is measured against a baseline they themselves set.