How Our NBA Props Model Works
A player-level projection system that finds mispriced lines across points, rebounds, assists, threes, and combo props.
Think of a weather forecast. Meteorologists do not just look at today's temperature and guess tomorrow's. They ingest satellite data, atmospheric pressure, wind patterns, and historical trends, then run simulations to produce a probability distribution: “70% chance of rain.” Our NBA props model works the same way, except the forecast is for player performance and the “rain” is whether a stat line goes over or under.
This page explains how the model actually thinks, from raw data to the picks that surface in your terminal. We will keep it honest: you will understand exactly what powers the system without us handing out the recipe.
The Big Picture
What Are Player Props?
Instead of picking who wins a game, a prop bet asks: “Will this player score more or less than 24.5 points tonight?” Sportsbooks set that line using their own models. When our model disagrees with their number by a meaningful margin, that gap is what we call edge.
Key insight: We are not predicting exact stat lines. We are identifying situations where sportsbooks have mispriced the probability. You do not need to be right every time, just right more often than the odds imply.
Props We Cover
How the Model Thinks
Imagine you are a chef preparing a dish. You start with core ingredients (historical stats), adjust the seasoning based on who is eating tonight (the matchup), account for kitchen conditions (pace, rest, minutes), and then taste-test before it goes out (quality filters). Our model follows the same logic, but at scale.
Step 1: Gather the Ingredients
Every projection starts with data. Not just season averages, but a layered picture of who a player has been recently versus over the long haul.
Season Foundation
Full season per-game averages and standard deviations for every stat category.
Recent Form
Last 10 and last 5 game windows, weighted to capture current trajectory without overreacting.
Live Market Data
Real-time odds from 30+ sportsbooks, line movement, and injury reports updated continuously.
Why it matters: Season averages treat a November game the same as last Tuesday. A player coming off three 35-point games is not the same as one who averaged 25 over six months. We blend both signals so the projection reflects who the player is right now, anchored by who they have been all year.
Step 2: Build the Projection
The raw ingredients get blended into a base projection, then passed through an adjustment pipeline. Think of it like a GPS recalculating your route: the base route is your season average, but traffic (matchup), road conditions (rest), and speed limits (pace) all modify the final ETA.
Each adjustment factor is applied in sequence. The model also detects hot and cold streaks: if a player's last five games diverge meaningfully from their weighted average, the projection nudges in that direction, capped to prevent overreaction.
Step 3: Read the Matchup
A player does not perform in a vacuum. Imagine two identical runners: one races on a flat track, the other uphill in the rain. Same athlete, wildly different results. Our model accounts for the “track conditions” of every NBA game.
Defensive Matchup
We use per-100-possession defensive ratings, not raw points allowed. This isolates true defensive quality from pace, the only fair way to compare defenses.
A team that allows 115 points in a fast-paced game might actually be a better defense than one allowing 105 in a slow grind.
Pace and Tempo
Fast teams create more possessions, which means more opportunities for every stat category. A player facing the league's fastest team gets a boost; facing the slowest, a reduction.
This is not a guess. We calculate expected possessions and scale accordingly.
Rest and Fatigue
Back-to-back games reduce output. Extra rest days boost it. The model applies calibrated adjustments based on days since last game, because a rested player and a fatigued one are not the same bet.
Injury Redistribution
When a high-usage teammate sits out, those shots, assists, and touches do not vanish. They redistribute. We model exactly how that freed-up opportunity flows to remaining players, adjusted by position and role.
The Variance Problem
Why Most Prop Models Get This Wrong
Picture two archers. Both shoot at a target 3 inches to the right of center. Archer A groups every shot within a 1-inch circle. Archer B sprays arrows across a 3-foot spread. For Archer A, being 3 inches off-center is a meaningful, correctable bias. For Archer B, it is noise lost in the chaos. The same principle applies to player props.
Most models treat every player like the same archer. They apply a generic volatility assumption, say 15-20%, to everyone. This means a 3-point edge on a consistent scorer like Jayson Tatum gets treated the same as a 3-point edge on a wildly inconsistent role player. One is a real signal. The other is statistical noise.
Our model calculates individual volatility profiles for every player, for every stat, using actual game-to-game variance. The same raw edge can mean completely different things depending on who the player is.
Visualizing the Difference
We also use probability distributions that account for “fat tails”: the reality that blowouts, overtime games, and injury exits happen more often than standard bell curves predict. Real NBA data has heavier tails than textbook statistics assume, and our model is built for the real world.
Finding the Edge
Model vs. Market
Once the model generates a projection, we compare it to what sportsbooks are offering. Every set of odds implies a probability. If a book prices a line at -110, they are implying roughly 52.4%. If our model says the true probability is 65%, the gap is our edge.
Important: Not every edge is worth taking. A small edge on a volatile player is noise. A large edge on a consistent player with good odds is a high-confidence pick. The model weighs all of these factors.
Four-Gate Quality Control
Think of it like airport security. Every potential pick must clear four independent checkpoints before it reaches your terminal. If it fails any single gate, it is rejected, no exceptions.
Gate 1: Expected Value
The mathematical profit expectation must exceed a minimum threshold. Marginal edges get rejected.
Gate 2: Projection Confidence
The edge must exceed the player's individual volatility. A 3-point edge means nothing if the player swings 10 points game to game.
Gate 3: Odds Floor
Heavy favorites beyond a certain threshold are excluded. The risk-reward ratio becomes unfavorable regardless of edge.
Gate 4: Breakeven Buffer
The edge must be thick enough to survive real-world variance. Razor-thin edges get cut even if they are technically positive EV.
Confidence Tiers
Picks that survive all four gates are ranked into confidence tiers based on how far our projection deviates from the line, scaled by the player's volatility. Higher conviction means a larger recommended position.
Highest model conviction. Large deviation relative to player volatility. Strongest edge in the slate.
Clear edge with solid projection confidence. Reliable with meaningful expected value.
Clears all four gates with adequate margin. Positive EV but lower conviction than the tiers above.
How We Validate
Closing Line Value (CLV)
This is the gold standard. CLV measures whether the line moved in your direction after you locked in your pick. If you bet a player Over 24.5 points and the line closes at Over 26.5, the market confirmed your read. You got a better price than the final consensus.
Sportsbooks use CLV to identify their sharpest customers. Consistent positive CLV is the single strongest predictor of long-term profitability, because it means you are consistently ahead of the market, not just getting lucky.
How our picks are timed:
Picks lock at 10:00 AM ET daily. This gives you time to place bets before lines move, maximizing CLV capture.
Cross-Book Odds Aggregation
The same prop can be priced differently across sportsbooks. A player might be -120 on DraftKings but -105 on FanDuel. We scan 30+ licensed books in real time and surface the best available price, because getting the best odds on a winning pick is the difference between good and great returns.
Primary Books
DraftKings, FanDuel, BetMGM, Caesars
Sharp Reference
Pinnacle (market-setting benchmark)
Total Coverage
30+ licensed sportsbooks aggregated
What Every Pick Includes
Every pick that surfaces in your terminal has been through the full pipeline. Nothing is guesswork. Here is what powers each recommendation:
Player-specific volatility modeling
Individual variance profiles, not generic assumptions
Matchup-adjusted projections
Per-100 defensive ratings, not raw points allowed
Pace and tempo scaling
Expected possessions factored into every line
Rest and fatigue adjustments
Back-to-back penalties and rest boosts calibrated from data
Real-time injury redistribution
Usage and opportunity flow modeled when teammates sit
Fat-tail probability corrections
Distributions built for the real world, not textbook curves
Trend detection
Hot and cold streaks identified and weighted appropriately
Best available odds across 30+ books
Every pick shows the best price, not just one sportsbook
Frequently Asked Questions
Why use per-100-possession defensive ratings instead of raw stats?
Raw points allowed conflates pace with efficiency. A team allowing 115 points in a fast game might be a better defense than one allowing 105 in a slow game. Per-100 normalization isolates true defensive quality, the only fair way to compare.
Why does the model weight recent games more heavily?
Season averages treat October the same as last week. Players evolve throughout a season due to role changes, injuries, and matchup adjustments. Our recency-weighted system captures these shifts without overreacting to a single outlier game.
What does "Model % vs Book %" mean on a pick?
Model % is our calculated probability that a prop hits. Book % is the sportsbook's implied probability derived from their odds. The gap between the two is the exploitable edge. Larger gaps mean higher conviction.
What is Expected Value (EV)?
EV is the average profit per bet if you made the same wager thousands of times. Positive EV means the math favors you over time. Every pick we surface must clear a minimum EV threshold to ensure margins are worth the variance.
How do you handle correlated props?
We limit exposure to one prop per player per day unless separate edges on multiple lines exceed high-confidence thresholds independently. This prevents over-concentration on a single player outcome.
How is the model validated?
Closing Line Value (CLV). If lines consistently move toward our projection after we lock, the market is confirming our edge direction. CLV is the single strongest predictor of long-term profitability in sports betting.
Ready to see the model in action? Check today's NBA prop picks in the terminal.
Open NBA Props Terminal