A three-layer machine learning pipeline — proprietary scoring engine, dual ML models, Half Kelly position sizing — scanning 15+ sportsbooks daily to surface statistically significant edges before lines correct.
Every morning the scanner hits 15+ sportsbooks simultaneously — pulling moneylines, spreads, and totals across NCAAB and College Baseball. Raw odds are normalized into decimal format and timestamped. No filtering, no scoring. Just prices from the market.
Each market is run through the Clyp Score engine — a rules-based model that derives a fair-value consensus line by weighting odds across all available books. The score measures three things: how far a book deviates from consensus (arbitrage strength), how much positive expected value exists at the offered price (EV%), and whether the line has been drifting in our direction (momentum). Anything below 60 is eliminated here. The ML models never see it.
Candidates that clear the Clyp Score threshold are passed to the first ML model. The Logistic Regression assigns linear weights to each feature — odds, EV%, Clyp Score, book, sport — and outputs a win probability estimate. If that probability does not exceed the market's implied probability, the candidate is discarded. The LR model is conservative by design: reliable, stable, and well-suited to smaller datasets.
The same candidate is simultaneously evaluated by a 200-tree Random Forest — an entirely separate model operating independently of the LR. Where the LR draws straight lines, the Random Forest captures complex combinations: patterns like FanDuel + NCAAB + Clyp Score above 70 that convert at rates no single variable would predict. Both models must agree independently. If the RF dissents, the signal is killed — no exceptions.
Candidates clearing all four gates receive a wager size via the Half Kelly Criterion — a mathematically proven formula that stakes proportionally to edge magnitude, protecting capital during variance while maximising long-run growth. The signal is formatted and delivered to member inboxes before tip-off, permanently timestamped and logged to the track record.
Every morning you get one email. It contains the day's signals — team, book, odds, and wager size scaled to your bankroll. That's it. No dashboard, no app, no upsells.
Cancel at any time before your next billing date. No contracts, no lock-in periods, no penalty fees.
Informational purposes only. Not financial advice. Sports betting involves risk and may be illegal in your jurisdiction. Past performance does not guarantee future results.
| Date | Sport | Team | Book | Odds | Result |
|---|---|---|---|---|---|
| Loading track record... | |||||
| How are signals delivered to members? | Every morning via email before the first game tips off. Each signal specifies team, sportsbook, odds, and a Half Kelly wager scaled to a $100,000 reference bankroll. Divide proportionally for your own stake. |
| Which sportsbooks do I need? | FanDuel and DraftKings cover the majority of signals. Bovada and BetRivers are useful additions. Every signal specifies the exact book with the best line — no manual shopping required. |
| How is this different from a tout service? | Every pick is timestamped at issuance. Results are logged in full. Nothing is edited or removed after the fact. The track record visible to members is the complete and unaltered record — every signal, every outcome. |
| What does "all 3 systems confirmed" mean? | A signal must clear three independent thresholds: Clyp Score ≥ 60, win probability above implied per the Logistic Regression, and win probability above implied per the Random Forest. Most opportunities fail at least one gate. |
| What sports and markets are covered? | Currently NCAAB and College Baseball — markets where mid-major pricing demonstrably lags sharp consensus. Additional markets are added only when the models validate consistent, measurable edge over sufficient sample size. |
| Can I cancel at any time? | Yes. Cancel before your next billing date and no further charges apply. No lock-in periods, no cancellation fees, no minimum commitment. |