Most Valuable Player
Production, efficiency, availability, and team performance. One formula for every season.
A fixed production core adds points, rebounds, assists, steals and blocks, then subtracts turnovers. True shooting is clipped between 0 and 1 and scales positive production by 0.8 + 0.2 × TS. Playing-time confidence applies to the core and learned support. Fifteen monotone trees provide nonnegative adjustments from production, shooting and team results within fixed caps.
Michael Jordan Trophy| Input | Scoring behavior |
|---|---|
| Points per game | positive Each additional point enters the fixed core with coefficient +1, before the shared true-shooting and confidence multipliers. |
| Rebounds per game | positive Each additional rebound enters the fixed core with coefficient +1. |
| Assists per game | positive Each additional assist enters the fixed core with coefficient +1. |
| Steals per game | positive Each additional steal enters the fixed core with coefficient +1. |
| Blocks per game | positive Each additional block enters the fixed core with coefficient +1. |
| Turnovers per game | negative Each additional turnover enters the fixed core with coefficient −1. The production total is clamped to zero. |
| True shooting | positive TS is a decimal: 0.60 means 60%. The multiplier is 0.8+0.2×clip(TS,0,1), so efficiency supports production without a large standalone bonus. |
| Playing-time confidence | positive Min(MP82/1500,1) × min(MP/500,1) limits short samples. This is model reliability, not an NBA eligibility rule. |
| Production support | monotone favorable inputs Positive trees compare sustained scoring, rebounding, playmaking, defensive box production and ball security. |
| Team support | positive Team success and team strength receive workload credit. The complete team-dependent support can contribute at most 9.00 score points. |
| Shooting support | positive Attempt-aware shooting and efficiency support can contribute at most 1.89 score points, beyond the fixed true-shooting multiplier. |
Each rule checks a small set of statistical thresholds and adds a contribution. The effect of an input depends on the other inputs; it has no single fixed coefficient. Every threshold and contribution is included in the downloadable model data.





