This NBA model update improves how PropsBot generates Confidence Scores by removing sportsbook pricing from core player prop predictions and introducing a new AI signal to surface high-value betting opportunities.
Retrained 6 core NBA player prop models to operate independently from sportsbook lines
Disagreement bets (where PropsBot differs from the market) flipped from unprofitable to profitable in backtesting
Added a new AI signal: “Line is Nuts 🥜” to highlight high-value opportunities
Confidence Scores now reflect true model signal, not price mirroring
New AI signal: Line is Nuts 🥜 — flags props where the model and sportsbook disagree on direction
Signal appears on ~10–15% of NBA props, highlighting only the strongest opportunities (no noise flooding)
Updated NBA markets:
Points
Rebounds
Assists
Threes
Field Goals
Blocks + Steals
(All combo markets like PRA, PR, PA, double/triple doubles remain unchanged)
Retrained all 6 NBA player prop models to remove bookmaker pricing as an input feature
Models now rely entirely on:
Player statistics
Matchup data
Usage patterns
Rest & schedule context
Defense-vs-position metrics
This ensures that when PropsBot disagrees with the sportsbook line, it’s finding real signal — not price echoing
Backtesting shows these disagreement bets are now profitable, reversing prior losses in those same situations
Previously, models leaned heavily on sportsbook consensus, which looked accurate on paper but failed exactly where value exists. Now, Confidence Scores reflect independent analysis — giving you sharper edges when the market is wrong.
Combo NBA markets (PRA, PR, PA, doubles/triples) remain unchanged
MLB and NHL models unchanged (sportsbook pricing still complements thinner statistical features)
NFL models will be updated separately ahead of the 2026 season, with odds inputs removed across all 17 markets
