Frequently asked questions

BetMachine is a statistical sports betting signal system. A mathematical model analyzes team efficiency data, adjusts for opponent strength, and identifies games where the market's implied probability appears mispriced. When the model finds a meaningful gap — an "edge" — it publishes a pick with a recommended bet size. No human is selecting picks, and no AI language model is generating them.

The model measures how efficiently teams score and prevent scoring, adjusts those ratings for the strength of opponents faced, and generates a predicted margin for each game. That margin is converted to a win probability, which is then compared to the market's fair probability (the Vegas line with the bookmaker's margin removed). The difference is the edge. Bets are only published when the edge clears a minimum threshold derived from historical optimization. Picks that pass filtering are sized 1–5 units based on edge magnitude.

Walk-forward out-of-sample win rates by sport (push-excluded):

"Out-of-sample" means these results were measured on seasons the model never saw during training — not backtested on the same data used to build it. NFL 2023–24 (49.4%) is included in the NFL aggregate.

NFL, NCAAF, NBA, NCAAB, and NHL. MLB is not covered by the model — starting pitcher assignments are typically made within hours of game time, after the model runs, making efficiency-based baseball prediction structurally unreliable. MLB games and odds still appear on the games board.

It varies by sport and time of season. The model is selective — it publishes bets only when the edge exceeds a minimum threshold. On a heavy NCAAB day, there may be 15–30 picks. On a light NFL week, there may be 3–8. The model does not force picks to fill a quota.

Picks are published automatically by the pipeline before game time. The model runs multiple times daily. Picks for evening games are typically available by morning ET.

Win rate is the percentage of decided bets (pushes excluded) that were correct. A win rate above 52.4% is profitable at standard -110 juice pricing. The win rate is independent of juice — it measures prediction accuracy, not dollar return. ROI (return on investment) is the better measure of profitability, since it accounts for unit sizing. The model's ROI figures are published before juice; subtract approximately 4–5 percentage points to estimate after-juice ROI.

Spread: a bet on whether a team covers a point handicap. If the spread is -6.5, the favored team must win by 7 or more to cover; spread bets are typically priced at -110 (bet $110 to win $100). Total (over/under): a bet on the combined score of both teams. If the total is 214.5, you bet whether the final score sum will be over or under that number. Moneyline: a bet on which team wins outright, no point handicap. Favorites are priced negative (-180 means bet $180 to win $100); underdogs positive (+155 means bet $100 to win $155).

BetMachine uses a unit system. One unit equals whatever base bet size you choose — for example, $25, $50, or $100 per unit. The model recommends 1–5 units per pick, scaled by edge magnitude. A 5-unit pick represents the model's highest confidence. Using a consistent unit size lets you track performance and risk exposure accurately over time. Never bet more than you can afford to lose on any single game.

AI Picks — GPT-4.1, Claude Sonnet, and Gemini 2.5 Flash picks for every game across all sports. No account required to view. Free account holders additionally see BetMachine Model moneyline picks with edge% included.

Full BetMachine Model access: spread picks, total picks, moneyline picks, full edge%, and unit sizing across all five sports. Pro subscribers see every pick the model publishes with complete context.

Visit the pricing page to see current plan options. Pro is available at $35/week or $99/month. You can upgrade directly from your account page after signing in. Cancel anytime — no long-term commitment required.

Subscriptions can be cancelled at any time before the next billing cycle. Because BetMachine provides statistical signals and not guaranteed outcomes, completed billing periods are generally non-refundable. See our refund policy for full details.

Every pick published by the model is stored with a timestamp, bet type, pick direction, edge%, and unit recommendation. Results are recorded automatically when game outcomes become available. The complete published record — wins, losses, and pushes — is visible on the results page. Nothing is deleted. Nothing is curated.

Picks are published automatically before games start and timestamped at publication. Results are recorded from official game outcomes after the fact — no human can alter a pick after the game begins. The pick record is not third-party audited, but the automated pipeline eliminates the ability to retroactively edit or delete picks. The NFL 2023–24 season (49.4% win rate — our only below-50% season) remains in the published record as evidence of this commitment.

Backtest (walk-forward): performance simulated on historical seasons using a model trained on earlier data. Parameters are derived on seasons before the test window — the model never saw the outcomes it was tested on. This is how the headline win rates (55.5%–67.2%) are calculated. Live performance: actual published picks going forward in real time. Live results are newer and have smaller sample sizes, but they represent the model operating in real conditions with real closing lines. Both are published on the results page. How closely they match over time is the real test of the model's validity.