The Onside Prediction Engine

Instead of one formula predicting FPL points, Onside runs nine independent signals through a single model that has learned how much to trust each one — based on how well it has actually predicted points in real gameweeks. The model is retrained every week on the latest results, so those weights keep adjusting as the season plays out.

The output is a single number — xP (expected points) — for each player, for each gameweek. xP is Onside's best estimate of how many FPL points that player will score, accounting for their fixtures, fitness, form, and attacking role.

How those signals are built, weighted, and combined is proprietary — it is the part that makes Onside Onside. What we publish instead is the thing that actually matters to you: whether it is right. See the graded record →

Captain picks are tracked every gameweek and scored after the final whistle. Accuracy is measured as: did the recommended captain outscore the FPL average captain?

This is a deliberately strict benchmark — it compares against the collective wisdom of millions of FPL managers, not a random baseline. The engine scores well on it because it weighs who is likely to score, whether they are likely to play, and how they are performing right now, all at once — rather than leaning on any single one.

Late team news, surprise rotations, and red cards introduce variance that no model can predict. A striker who misses the warm-up bus on Saturday morning has 0 xP in reality, regardless of what the engine projected on Thursday.

Onside refreshes xP scores as close to kickoff as possible using live market and availability data — but the final team sheets are announced 75 minutes before kickoff, and sometimes managers make unexpected choices even then.

The engine is calibrated to be right more often than it is wrong over a full season. It is not designed to guarantee the correct captain pick every single week — no tool can do that. The goal is to give you a statistically better starting point than gut feel alone.