How iPick makes a pick — and why it often matches the favourite

A model for each league

Each of the six leagues has its own statistical model, trained on that league's past games. Every model rates the teams from their results and margins, and weighs rest, recent form and the season so far. Each adds what matters most in its sport: the starting pitcher in baseball, the quarterback in the NFL, who is out in the NBA, shots and the schedule in hockey, roster talent and returning players in college football. The details are on What each model knows.

A model is judged only on seasons it never saw while it was trained, and it replaces the previous one only if it is not worse on those seasons.

Weighed against outside estimates

Before a game, other estimates of the same game exist: sportsbooks' prices and prediction markets. Every week, iPick measures — on recent games the model did not train on — whether its own model adds anything to those outside estimates. When it does not, the pick follows the outside estimate, recalibrated if needed.

In October 2026 that test came out the same way in all six leagues: the outside estimates were more accurate than the model on its own. So for a game with an outside estimate, iPick's pick is, in practice, the market's favourite with the market's probability; the model decides on its own only when there is none.

Why that is not a failure

Sportsbooks' closing prices are the most accurate public forecasts there are for these leagues; large, well-known models such as ESPN's FPI and FiveThirtyEight's Elo did not beat them over several seasons (ECU/DSI). A site that pretends to beat them every day is selling something. iPick shows the probability honestly instead — how likely the pick is, at which level, and how picks at that level have really done.

What was tried and rejected

Every idea is tested the same way — fixed in advance, judged over about ten past seasons it was not tuned on — and kept only if it clearly lowers the model's error. Some passed and are used: the NFL quarterback, the MLB starting pitcher, NBA availability, NHL schedule, college rosters. Others did not, and are not used: NHL goalies by goals saved above expected, NHL expected goals, NFL efficiency per play, NFL weather for the winner, and NBA back-to-backs, travel and altitude. Negative results are results too.

When the pick is made

The pick is recalculated every few minutes until the game starts, so late news is taken in. The pick a game carries when it starts is the one it is graded on afterwards, and it never changes. Every graded pick, the misses with the hits, counts in the Track record.