Who is beating expectations?
How the model has revised each player's expected impact since preseason. Positive means the player’s impact rating has moved up; negative means it has moved down. This uses the Board’s existing player model, updated from team game margins and actual minutes.
Early-season estimates are uncertain. Teammates share the same game results, so their changes are related. A small positive revision alone is weak evidence of over-performance. The intervals compare current impact with a fixed preseason expectation; they are not intervals for a player's change in underlying talent.
Player impact versus preseason
Ratings: points per 100 possessions above league average. Dots: revision; blue lines: 90% marginal model intervals, including the live season simulation’s uncertainty scale. Zero is the preseason expectation.
| Player | Team | Games | Minutes | MPG / projected | Preseason | Now | Revision | 90% interval | Revision & uncertainty | P(above preseason) | Evidence |
|---|
How to read this model
The preseason column is the adjusted prior used by the live v2 filter (release, rim defence, BPM and young-player adjustments). The current column is its saved player posterior. The revision is current minus that fixed prior mean. It is independent of fair salary and contract value.
Each game supplies a capped team margin. The filter attributes the surprise jointly to players, weighted by actual minutes, and to a team residual. It adjusts for opposition, home court and rest. These are impact estimates inferred from team results, not separate player box-score production grades or causal estimates of individual contribution. The model does not identify separate new offence and defence ratings from a single game margin.
Above/below expectation requires at least 100 minutes and a 90% interval entirely above/below zero. These are marginal model statements; many players are inspected and the player-level probabilities have not been independently calibrated. With fewer minutes, the evidence label says so even if a rating moves through shared covariance. Players outside the filter state get no invented rating. Later roster additions retain their current impact rating but have no preseason comparison; their roster-supplement prior is available in the CSV.
Minutes and appearances come only from completed regular-season games included in the same archived run. DNPs count as zero minutes and no appearance. A departed player remains available by turning off the roster filter; their historical performance is retained.
This page reads the existing forecast archive. It does not fit a new model, change forecast parameters, or record new predictions.