NBA Model
How the projections are built and what the numbers mean.
Opportunity-First Modeling
NBA projections start with opportunity — how many 2-point attempts, 3-point attempts, and free throw attempts a player is likely to get — then layer conversion rates on top. This reverses the typical "predict the box score" approach and produces more stable projections when lineup context shifts.
Distribution Models
Each prop market uses distribution modeling (Normal, Negative Binomial, or Student's t with truncation) fit to the player's historical variance. Monte Carlo simulation with 25,000 draws per prop produces over/under probabilities that respect the shape of each stat distribution.
Feature Engineering
Projections incorporate defensive matchup quality (opponent-specific zone defense ratings), teammate synergy (second-order effects from off-ball players), referee crew tendencies, player archetypes (star/role/bench classification), home/away and rest-day splits, and Vegas anchoring (market-aware residual modeling).
What Edge Means
Edge is the gap between our model's fair probability and the implied probability from the sportsbook line. Larger gaps suggest the market may be mispricing a prop. Every edge still needs context — injuries, line movement, and market liquidity should all be checked before acting.
Calibration & Validation
The model is validated using walk-forward cross-validation (time-series, no random splits), leakage detection audits, and promotion gates that must pass before a model update goes live. Historical pick accuracy is tracked and published on the History page.
How to Use This Responsibly
Nothing But Bet is an informational tool. It surfaces where our model disagrees with the market — not guaranteed outcomes. It should not replace bankroll discipline, injury news review, or local compliance checks.