How to judge a sports prediction site

1. Is every pick published before the game, and kept?

A record is only worth something if every pick was public before the game started and none was removed or edited afterwards. On iPick, the pick a game is graded on is the one it carried when it started; it does not change after. Predictions computed after a game from what was known before it are labelled “retro” and counted apart.

2. Is it compared with the obvious answer?

Picking the favourite already wins about 57% of baseball games and about 75% of college football games (see How often does the favourite win?). A record should be compared with the favourite on the same games. On 432 games in September and October 2026, iPick's pick won 67.1% and the sportsbook favourite 67.4%.

3. Do the probabilities mean what they say?

If a site gives probabilities, check whether its 60% picks win about 60% of the time and its 80% picks about 80%. FiveThirtyEight used to publish exactly this check: its 70% events happened 71% of the time (FiveThirtyEight). iPick's Track record shows, for each level, how often the picks won next to how often they said they would.

4. Log-loss and Brier: better than win rate

Win rate treats a 51% pick and a 90% pick the same. Two scores do not. Log-loss punishes each prediction by the probability it gave to what happened, and punishes confident mistakes hardest; a coin flip scores 0.693, and lower is better. The Brier score is the average squared gap between the probability and the outcome; a coin flip scores 0.25. Professionals judge models with these, against the market's own prices on the same games.

5. Are there enough games?

Short records are mostly luck. A true 65% predictor can show anything from 47% to 80% over 30 games. To show, with confidence, that a site picks just one point more winners than the favourite would take thousands of games, because the two agree on most of them. Be wary of any claim built on a few weeks.

Red flags

• A record that starts on a convenient date, or shows only winners.
• “Locks”, “guaranteed” or “can't lose” picks: no honest probability is 100%.
• Win rates with no comparison and no sample size.
• Pressure to bet, pay or act fast.

iPick takes no bets and sells no picks; its predictions are information, not advice (Terms of Use).