NBA Start Score

How the NBA Start Score works

The formula. Each player's projected fantasy points for the upcoming game is estimated as: (Projected Minutes / 36) × Per-36 FP × Matchup × Availability. That projected value is then converted to a 0–100 score by percentile rank within the position — so a score of 85 means the player ranks in roughly the top 15% of playable players at their position for this game slate.

Fantasy points formula (Sleeper default). PTS × 1.0 + REB × 1.2 + AST × 1.5 + STL × 3.0 + BLK × 3.0 + 3PM × 1.0 + TOV × (−1.0). This is a labeled assumption; platform-specific formats are future toggles.

Minutes is the projected playing time — a decay-weighted rolling average of the player's last 10 games (most recent game weighted 18%, fading to 3% for the tenth). Minutes is the most stable per-game predictor in basketball: role determines the floor; everything else is a modifier.

Per-36 FP is the player's per-36-minute fantasy point rate over their recent games, regressed toward the positional baseline so small samples don't produce noise-driven extremes.

Matchup is a bounded multiplier (0.85–1.15) derived from the opposing defense's fantasy points allowed per game to this position bucket (Guard / Forward / Center) over the last 60 days, compared to the league average. A higher number means the defense has been more permissive.

Back-to-back rest risk. When a player's team plays on consecutive days, the projected minutes receive a haircut before the score is computed. High-minute veterans (≥32 avg min/game) receive a larger reduction (×0.55) reflecting documented load-management patterns; all other players receive a smaller reduction (×0.92). Players with a B2B badge carry elevated uncertainty — check news before locking them in.

Positions. NBA players are grouped as Guard (G), Forward (F), and Center (C) — the buckets reported by balldontlie.io. PG/SG/SF/PF distinction requires a secondary data source and will be added in a future version.

Data. Game logs and schedule from balldontlie.io (official API, documented terms). Scores are pre-computed daily by the ETL and cached in Supabase; this page never calls the API directly. No machine-learning model is used — every weight and assumption is published here.