Sports Intelligence · Transparency

The Honest Track Record

We predict game winners well. We have not shown we can beat the closing line. Both of those are true, and we publish both — because a track record is only worth anything if it includes the parts that don't sell picks. That's why there is nothing to buy on this page.

We'd rather you judge the model from the record than from a number we put on a banner. So we don't headline an accuracy figure — we publish every pick against the actual result, out-of-sample, and let the track record earn its own number over time.

For the record, we also don't stand behind the often-quoted 80.23% — that was a single model measured on its own selection set (in-sample, not defensible). What matters is below: are the probabilities honestly calibrated, and does any of it beat a real closing line?

Are the probabilities honest?

Calibration

Predicting the winner is only half of it — the stated confidence has to mean something too. We measure that with the Brier score (lower is better; 0.25 is a coin flip) and calibration error (how far stated confidence drifts from real hit rate).

NCAA

Under-confident

0.189

Brier

0.139

Cal. error

The model is under-confident — it wins more often than its stated confidence implies. Reliability buckets track: lower-confidence bands land at or above their stated rate, and higher-confidence bands come in at or above par.

MLB

Well-calibrated

0.244

Brier

0.012

Cal. error

Probabilities are honest (ECE 0.012) but the edge is thin — only marginally above a pick-the-favorite baseline. Calibrated confidence, marginal signal.

Can it beat the market?

Real closing lines

Every settled pick is staked against the REAL closing moneyline. ROI is the return on flat 1-unit bets; the 95% confidence interval is a nonparametric bootstrap over the settled bets. A league is STAKE-eligible only if the bootstrap ROI CI lower bound is above 0 on at least 30 settled bets. Every league fails that bar, so every league is NO_STAKE.

League	Settled	Hit rate	ROI	ROI 95% CI	Verdict
MLB	251	41.0%	-13.2%	[-26.3%, -0.3%]	NO_STAKE
WNBA	52	30.8%	-22.5%	[-56.0%, +17.6%]	NO_STAKE
NCAA	22	54.5%	-8.2%	—	NO_STAKE
SOCCER	20	35.0%	+16.8%	[-62.8%, +114.4%]	NO_STAKE

Hit rate is picks that beat their real closing line — a different, harder test than picking the winner. Every league fails the stake gate (bootstrap ROI CI lower bound ≤ 0 or fewer than 30 settled bets). No exceptions, no cherry-picking.

Why we don't sell picks

Picking winners and making money betting are two different things. A favorite can win most of the time and still be a losing bet if the price already accounts for it — the closing line is a very efficient market, and our forward record against it is negative or too thin to trust in every league above.

So we won't sell a pick, a subscription, or a “Pro tier” on sports — not until a model shows a positive return on real closing lines with enough settled bets to rule out luck. Until then the daily picks stay free to follow, and the honest asset here is the analytics: strong, calibrated winner prediction, published in the open.

Method · Winner accuracy and calibration come from the sportsify-pipeline honest-ensemble evaluation (out-of-sample tournament games and season backtests); the real-line scorecard stakes every settled pick against its actual closing moneyline and bootstraps the ROI confidence interval. Figures are a committed snapshot last updated 2026-09-06.

← Back to Sports Tips