A two-stage approach: first predicting minutes, then forecasting stat rates via Monte Carlo simulation. Every projection comes with a full probability distribution — not just a single number.
A gradient-boosted regression model predicts how many minutes each player will see on the court. It uses rotation patterns, matchup data, pace metrics, rest days, and recent usage trends. Minutes are the foundation — they determine opportunity volume for every stat.
With predicted minutes as the base, thousands of game scenarios are simulated by sampling stat rates conditioned on minutes played. For combo markets (PRA, PR, PA, RA), correlated multivariate normal distributions preserve the natural relationships between stats — a high-assist game tends to correlate with high points, and the model captures this.
Unlike tools that output a single projected number, Propboard shows the full distribution of outcomes. You see the 10th percentile, median, 90th percentile, and everything between. This gives you the complete picture: not just what's likely, but how wide the range of outcomes is.
Each simulated distribution is compared against real sportsbook lines from 10 books. When the model's probability for an over or under diverges meaningfully from the implied odds, the edge is surfaced with a confidence grade. Isotonic calibration ensures that when the model says 65%, it means 65%.
Seven guardrail checks run on every projection: blowout risk, minutes turbulence, new role detection, volatility flags, injury context, model uncertainty, and low-confidence warnings. When a guardrail fires, the pick is flagged so you can weigh the risk before acting.
Player stats and game data come from BallDontLie, covering multiple seasons of historical performance. Odds are pulled in real-time from The Odds API, covering 10 sportsbooks including DraftKings, FanDuel, Pinnacle, BetMGM, and others. All data is refreshed throughout the day during active game windows.
3-day free trial. Cancel anytime.