Rim protection, with uncertainty.
Rim protection, possession-ending events and team context in a new current-season Stan model.
Rim evidence and experimental components
Rim effectiveness is the model-implied difference in points conceded at the same estimated attempt rate, relative to the league reference. Its intervals remain uncertain and it is not a complete defensive rating. The main overall composite overcredits defensive-rebound associations in this release. Composites and their top-ten frequencies are diagnostic outputs, not All-Defence recommendations. The hustle extension has no held-out validation claim.
| Player / last observed team | Rim effectiveness / 100 | Model rim FG% allowed | Observed rim makes / attempts | Experimental composite / 100 | Composite contribution | Composite top-ten frequency | Observed minutes |
|---|
Intervals contain the central 80% of model draws. Composite top-ten frequency ranks the experimental composite contribution in each draw; it does not rank rim effectiveness or predict official selection. Every row retains its uncertainty; the table does not assign first or second All-Defence teams. Click a player for components and matches.
What contributes to the estimate?
| Component | Points / 100 · 80% interval |
|---|
How to read it
Rim effectiveness measures allowed shooting relative to the model's league reference. Rim opportunity accounts for attributed attempts and the remaining-score relationship. Other event contributions use learned conditional coefficients. The residual represents what these channels do not explain. Team context stays separate.
The main composite is dominated by defensive-rebound associations. Rebounds follow missed shots and depend on opportunity, role and teammates; the current model does not isolate their individual defensive value. These are statistical associations, not identified causal effects. No player receives a position or reputation bonus. Missing hustle data stays unavailable.
Current-season matches
| Date | Opponent | Minutes | Rim makes / attempts | Steals | DREB | Fouls | Deflections | 3PT contests | Defensive box-outs |
|---|
Did it improve predictions?
Two chronological development checks use later games, conditional on observed minutes and estimated possessions in every model. Future event counts and rim outcomes are simulated for the main comparison. The comparison conditioned on actual events is a separate diagnostic. This is a small, inspected development sample, rather than a prospective pregame test.
| Comparison | CRPS difference | 90% game-bootstrap interval | Paired games |
|---|
Sources and model boundaries
Official NBA player/team game logs, defensive-impact tracking and hustle data. The separate under-six-foot dashboard is not added as another independent rim-shot sample.
Rim makes enter a binomial channel. Those points are removed before fitting remaining scoring, which conditions on realized event counts. The model uses individual events alongside team and opponent context; it cannot establish which defender caused every outcome. Raw data, implementation, draws and derived results have dated integrity receipts. Minute-share exposures and box-score possession estimates are approximate. Opponent-team adjustments do not identify individual shooter quality or responsibility for every shot. Four-chain convergence checks establish sampling quality, not accurate player attribution.