Skip to content
// methodology

How Our AI Predicts Sports

No black box. No vague claims. Here's exactly how our system works, what data it uses, and why we publish every single loss.

The 30-Second Version

Five different machine learning algorithms independently analyze every game. Each one looks at different patterns in the data — team stats, recent form, matchup history, market odds, and sport-specific factors. When they agree, we publish the pick with a confidence tier (LEAN, CONFIDENT, STRONG, or LOCK). When they disagree, we still publish — because every game gets a pick. Then we track every single prediction publicly on our dashboard. Wins and losses. No hiding.

The 5-Model Ensemble

Instead of relying on a single algorithm, we run five different models on every game. Each model has different strengths, and together they catch patterns that any single model would miss.

XGBoostPrimary predictor

Gradient-boosted decision trees. Excels at finding non-linear patterns in structured data — like how rest days interact with travel distance to affect NBA performance.

LightGBMSpeed & breadth

Similar to XGBoost but optimized for speed and large datasets. Handles the 5,000+ tennis matches and 1,000+ college basketball games in our training data efficiently.

Random ForestStability anchor

Builds hundreds of independent decision trees and averages them. Less likely to overfit to recent results, which keeps the ensemble grounded during hot or cold streaks.

Neural NetworkEdge case detection

A multi-layer perceptron that detects subtle interactions between features that tree-based models might miss — like how a specific coaching matchup changes game dynamics.

Logistic RegressionCalibration baseline

The simplest model in the ensemble. If the complex models disagree but logistic regression is confident, that signal matters. It also serves as a calibration check — if it consistently outperforms the others, something is wrong.

Confidence Tiers

Not all predictions are created equal. Our confidence tiers tell you how much edge the model sees, so you can size your bets accordingly.

LOCK
80%+
Full model agreement. Strongest edge.
STRONG
70-80%
High confidence. Clear advantage.
CONFIDENT
60-70%
Solid edge. Worth a standard bet.
LEAN
50-60%
Slight edge. Proceed with caution.

Our LOCK picks hit at 82.2% across 208 predictions. See the full tier breakdown on the dashboard.

What Data We Use

For Every Sport

  • Win-loss records, home/away splits, recent form
  • Point differentials, scoring trends, defensive ratings
  • Head-to-head matchup history
  • Market odds from The Odds API (what Vegas thinks)
  • Strength of schedule and conference rankings

Sport-Specific Features

NBA
Rest days, back-to-backs, pace factor, net rating, travel fatigue, injury impact
NCAAB
Conference strength, KenPom-style efficiency, home court advantage (60% in college), experience, tournament context
NHL
Power play %, penalty kill %, goal differentials, back-to-back detection, travel distance
Tennis
Surface type (clay/hard/grass), serve stats, head-to-head record, tournament round, fatigue

The Feedback Loop

This is the part most prediction sites skip. We don't just make picks and move on. Every result feeds back into the system.

1
Predict
Model analyzes the game
2
Publish
Pick goes live publicly
3
Result
Game outcome recorded
4
Learn
Model weights adjusted
5
Improve
Next prediction is better

Models retrain automatically twice a week with fresh game data. Sport-specific blend weights adapt based on which algorithm is performing best for each sport. The system literally gets smarter with every game.

Why We Publish Every Loss

Most prediction sites only show their wins. We show everything — every correct call and every miss. Here's why:

  • ✓You can verify our track record yourself. We're not asking you to trust a claim — we're showing you the data.
  • ✓It makes the AI better. Every loss is training data. If we hid our misses, we'd be hiding the signal that improves the model.
  • ✓It builds trust. Anyone can claim 80% accuracy. We prove ours is 62.6% across 949 games — and that's real, verified, honest.
  • ✓It's what we'd want. If we were paying for picks, we'd want to see the full record. So that's what we give you.

Current Performance

SportRecordAccuracy
NBA53-1874.6%
NCAAB277-13567.2%
Tennis240-17458.0%
NHL24-2846.2%
Overall594-35562.6%

Updated daily. Full breakdown on the dashboard.

See It in Action

949+ predictions. Every result public. 99¢ gets you every pick, every sport, forever.

Get All Picks — 99¢

5 sports · 99¢ lifetime

99¢