Skip to content

NAIL-RAPM v1.2.2: Defensive-Rebound Profile

NAIL-RAPM v1.2.2 is a strict one-feature extension of v1.2.1. It adds the sum of the five players' forward-safe, stat-specifically padded defensive-rebound percentages as a ninth additive player-profile coordinate. The two retained non-additive terms, value-conditioned aging prior, cold-start gate, gap-returner bridge, and context penalty are unchanged.

Contract

For unit (U), the new additive coordinate is

\[ \operatorname{DRB\%}(U) = \sum_{i \in U} \widetilde{\operatorname{DRB\%}}_i, \]

where the tilde denotes the same lagged, 108.26-pseudo-possession padded player rate already used by the profile mart. The fitted home-minus-away coefficient can be compiled into the player-attributable additive rating; it is not a new non-additive lineup effect.

Decision Plan

The branch will be trained recursively through 2025-26, then compared directly with v1.2.1 on the frozen 2023-24 through 2025-26 regular seasons and playoffs. The DRB% trajectory will be retained as a defensible profile feature only if it has a positive sign in at least 80% of the 29 source-season fits. Promotion also requires the existing paired full-game-RMSE non-promotion gate to pass in the pooled replay and each frozen season.

Coefficient Audit

The recursive fit produced 29 source-season coefficients: 1997-98 through 2025-26. Each panel is a standardized conditional Ridge coefficient, so it answers how the home-minus-away stint estimate changes for a one-standard- deviation increase in the named unit total while the other eight additive coordinates and the two retained non-additive terms are held fixed.

NAIL-RAPM v1.2.2 additive profile coefficient trajectories

defensive_rebound_pct is positive in 21 of 29 source seasons (72.4%), with a median standardized coefficient of +0.126, mean absolute coefficient of 0.269, and range from -0.356 to +0.776. It therefore fails the predeclared 80% positive-sign gate. The other eight inherited additive coordinates retain their resolved dominant signs, so adding DRB% did not displace a previously stable profile feature; it simply fails to earn a place alongside them.

The two retained non-additive terms remain directionally resolved in this joint fit: usage concentration is positive in all 29 source seasons, and top-two assists is positive in 26 of 29.

NAIL-RAPM v1.2.2 retained non-additive coefficient trajectories

Frozen Results

The direct incumbent is v1.2.1. Both models forecast 2023-24 through 2025-26 from the matching preseason information set; target-season outcomes are never available to the forecast.

Model Poss. RMSE Eligible game RMSE Full-game RMSE Winner accuracy Team NetRtg RMSE Pythagorean-win RMSE Playoff eligible-game RMSE
NAIL-RAPM v1.2.1 1.197952 14.0246 14.2521 68.24% 3.2706 7.0351 16.5942
NAIL-RAPM v1.2.2 1.197952 14.0238 14.2529 68.39% 3.2723 7.0434 16.5875

v1.2.2 is essentially tied at possession level, modestly improves eligible game RMSE, winner accuracy, and playoff eligible-game RMSE, but is slightly worse on the primary full-game and team-level metrics. The point estimates are too small to settle the decision, so the predeclared paired bootstrap remains the predictive guardrail.

Differences below are v1.2.2 minus v1.2.1; negative values favor the DRB% candidate. The non-promotion threshold is +0.5% of the incumbent full-game RMSE in each scope.

Scope Full-game RMSE difference Paired 95% interval Gate threshold Gate
Pooled +0.0008 [-0.0036, +0.0051] +0.0713 Pass
2023-24 +0.0038 [-0.0095, +0.0167] +0.0698 Pass
2024-25 -0.0021 [-0.0057, +0.0013] +0.0719 Pass
2025-26 +0.0007 [-0.0016, +0.0031] +0.0721 Pass

Decision

Not promoted. The direct bootstrap clears the no-material-harm gate, but the new DRB% coefficient is positive in only 72.4% of source seasons, below the agreed 80% interpretability requirement. The minimal predictive movement does not justify adding a directionally unresolved ninth additive player profile term. NAIL-RAPM v1.2.1 remains the promoted model and website bundle.

Artifacts

  • Recursive candidate: artifacts/models/forward_nail_rapm_v122_defensive_rebound_profile/2025-26/forward-nail-rapm-v122-defensive-rebound-profile-2025-26-20260822T213244Z-706ca7b2
  • Frozen replay: artifacts/models/nail_v122_defensive_rebound_profile_frozen_backtest/frozen_multiseason_backtest/2023-24_to_2025-26/frozen_multiseason_backtest-2023-24-to-2025-26-20260822T214052Z-9e7684d0
  • Paired bootstrap: artifacts/models/nail_v122_defensive_rebound_profile_bootstrap/2023-24_to_2025-26/nail-v122-defensive-rebound-profile-bootstrap-20260822T214132Z-4453ca36
  • Coefficient audit: artifacts/models/analysis/nail_v122_defensive_rebound_profile_weight_audit/nail-v122-defensive-rebound-profile-weight-audit-20260822T213316Z-2ac00473

Reproduction

See Train NAIL-RAPM v1.2.2 Defensive-Rebound Profile.