Show the math or shut up
Every pick ships with SHAP values. You see which factors moved the model, by how much, and whether the weighting makes sense for the matchup. No black box, no vibes.
What we hold ourselves to
Losing weeks get the same update as winning weeks
When variance hits us, you see it. When the model misses a trend for a week, we say so. The weekly recap is honest whether the numbers are good or bad.
Responsible by default
Self-exclusion that cannot be reversed early. Bankroll caps enforced at the command level. 1-800-GAMBLER on every pick embed. We would rather lose a customer than ship a harmful product.
Nobody on this platform sells guaranteed locks
Cappers caught using tout language (hammer, lock, insider, must-play) get delisted permanently. Marketplace sharps win on CLV, not hype.
How the AI actually works
01
Data in
Live odds from The Odds API (12 books), injury feeds, weather feeds, line movement history, CLV tracker, tracked bet results from the community.
02
Model scoring
XGBoost ensemble per sport. Features: recent form, rest differential, pace, venue, injury impact, line movement direction, public betting %, contrarian money indicators.
03
SHAP explainability
Every prediction decomposed into per-feature contributions. You see that the pick is +3.5% edge from rest diff, +1.2% from injury, -0.8% from venue, etc. Read the reasoning before you decide.
04
Pick posted
Confidence 1-5, Kelly stake sized to your bankroll, best current line across 12 books, reasoning, 18+ and 1-800-GAMBLER footer. Every time.
A small team, building in public
NuroBets is built by a small independent team with a betting background. We track every shipped feature, every model update, every edge case, and every incident in public. No VCs, no corporate overhead, no pressure to inflate the model.
Want to follow along? The blog covers strategy posts plus build updates.