How Our Predictions Work
A transparent, data-driven approach to sports prediction. No black boxes — here's exactly how we generate every pick.
The 3-Model Ensemble
Every prediction combines three independent models, each approaching the game from a different angle. This ensemble approach reduces individual model bias and produces more robust probabilities.
Statistical Model
Analyzes team performance metrics, win/loss records, and head-to-head matchup data. Uses the Log5 method to convert team strength into matchup probability — the same mathematical framework used in baseball analytics since Bill James.
Market Model
Ingests real-time sportsbook odds from major books (DraftKings, FanDuel, BetMGM) and converts them to true probabilities. Removes the vig (house edge), corrects for favorite-longshot bias, and produces a market-implied probability.
Situational Model
Captures context the other models miss: home court advantage, rest days between games, recent momentum (hot/cold streaks), and injury impact. Returns an adjustment factor that shifts the ensemble probability based on real-world conditions.
Ensemble Combination & Calibration
The three model outputs are combined using optimized weights. The Market Model typically carries the heaviest weight (markets are efficient), with Statistical and Situational models providing edge where markets are slow to adjust.
The raw ensemble probability is passed through a Platt scaling calibration layer. This logistic regression transform ensures that when we say "70% chance," the outcome actually occurs roughly 70% of the time. Calibration parameters are fitted on historical prediction data.
The calibrated probability is compared against market-implied odds to compute the "edge" — the gap between our assessment and the market's. Positive edge means we see value; larger edges indicate stronger conviction.
Closing Line Value (CLV)
Closing Line Value is widely regarded as the single best metric for evaluating prediction skill. Here's why:
What is CLV?
CLV compares the odds at the time of our prediction to the closing odds (final odds before tip-off/kickoff). If we consistently get better numbers than the closing line, it demonstrates we're identifying value before the market fully prices it in.
Moneyline CLV
Tracks whether our predicted probability was closer to the true outcome probability than the opening market odds. Positive moneyline CLV means our model identified the direction of line movement before it happened.
Spread CLV
Compares the point spread at prediction time vs the closing spread. Getting a better number (e.g., picking a team at -3 that closes at -5) indicates our spread assessment leads the market.
Full Transparency
All predictions are stored with timestamps, odds snapshots, and model outputs. Nothing is retroactively edited or deleted.
We use Brier scores — the gold standard for probability forecast evaluation — broken down by sport. Lower is better; 0.25 is random; anything below 0.20 shows meaningful skill.
Our Track Record page shows real-time performance across all sports with calibration charts, CLV trends, and sport-by-sport breakdowns.
Frequently Asked Questions
What sports does Edgewise cover?
We currently generate predictions for NBA, NFL, NCAA Basketball, MLB, and PGA Tour golf. Each sport uses the same three-model ensemble framework, calibrated independently for that sport's characteristics.
How often are predictions updated?
Predictions are generated daily when games are scheduled. Our system runs an hourly cron job that auto-resolves completed games and updates performance metrics. Best bets are refreshed each morning with the latest odds data.
How accurate are the predictions?
We measure accuracy using Brier scores (lower is better) and calibration curves. Visit our Track Record page to see real-time accuracy metrics broken down by sport. We believe in full transparency — every prediction is logged and publicly tracked.
What is "edge" in a prediction?
Edge is the difference between our model's probability and the implied probability from sportsbook odds. A positive edge means our model believes the outcome is more likely than the market suggests. Larger edges indicate stronger disagreements with the market.
How does Closing Line Value (CLV) work?
CLV measures whether our predictions capture value by comparing the odds at the time of our pick to the closing odds (final odds before game time). Consistently positive CLV is widely considered the strongest indicator of predictive skill — it means we're identifying value before the market corrects.
Is this gambling advice?
No. Edgewise is for entertainment and research purposes only. Our predictions are the output of statistical models and should not be considered financial or gambling advice. Always gamble responsibly and within your means.
See our predictions in action