The model calls 67.2% of moneylines across 2,597 games played after it was trained — picking home every time would have gone 55.1%. Every pick is logged before tip-off and graded where anyone can check, no account needed. The stats library comes with a free account.
Measured once, in a sealed pre-registered evaluation, on 2,597 games from 2024-25 and 2025-26 — none in the training data. 95% CI 65.32–68.93%. Statistically ahead of both our previous model (65.8%, p = 0.041) and the better-record baseline (65.3%, p = 0.039). Accuracy is not profitability: that test measured no return; the Track Record measures it live, with its range, from opening night 2026-27.
A free account. Free forever.
Ask the model, not the internet.
How much to bet, and how you are doing.
A free account gets all of this, and the Track Record is open to anyone. Pro and Elite build on it.
The XGBoost model called 67.2% of moneylines across 2,597 games from 2024-25 and 2025-26, none of which it was trained on — measured once, in a sealed pre-registered evaluation. Picking the home team every time would have gone 55.1%; picking the better record, 65.3%. 95% CI 65.32–68.93%. Accuracy is not profitability: that test measured no return; the Track Record measures it live from opening night 2026-27. Every live pick is logged before tip-off and graded at /track-record, hits and misses both.
A formula for bet size that assumes your win probability is exactly right. Because it never is, most people bet a fraction of it. We show the full-Kelly number and label it as such; it is not a recommendation.
Yes, absolutely. All plans can be canceled anytime with no penalties. Your access continues until the end of your current billing period.
No. We show you the math — probabilities, expected value and, on Elite, Kelly sizing — and then we show you our record. Accuracy is not profit. Betting always involves risk, and you can lose money using this product.
The stats database ingests the previous night's box scores every morning, and odds are recorded through the day, each time a line moves. Predictions are written to the ledger before tip-off and never edited after the fact.
It shows its work. Model probability, expected value, Kelly sizing and correlation warnings on parlays — plus a free reference library deep enough to check the model against, and a track record anyone can open to check us against, no account needed.
Start on the free library and pay nothing. Upgrade when you want the model's probabilities and expected value in one place, and to Elite for bet sizing and your own record.