01Recent games matter more
Older games fade gradually. Validation chose a 240-day half-life, so a short hot streak cannot overwhelm a player's established level.
02History and age stabilize the answer
The model blends the current season with longer-term performance and the normal way players change as they age.
03Context had to earn its place
Opponent, rest, team-change, and injury-return inputs were tested. They did not improve ability forecasts, so they were not forced into the score.
04Lineup evidence comes later
This version intentionally excludes teammate and opponent lineup effects. Those will be tested only after the frozen model passes prospective checkpoints.
Research preview—not the live Impact Rate.