Expected points explained: inside the FPLRogue model

The short answer

Expected points, or xPts, is an estimate of the average FPL return across possible outcomes. A forecast of 7.91 is not a promise of seven or eight points: a player can blank, miss the match or produce a large haul. Use the estimate to compare decisions under the same assumptions.

This guide explains how to read FPLRogue projections. The worked chart below is deliberately frozen at 6 August 2026. Current recommendations need the latest projection release and team news, not the old player values in that example.

How it works

Playing time determines how much opportunity a player has to earn points. Expected minutes and start probability describe different things: a likely starter can still be substituted early, while a substitute may collect a short appearance. Injury, rotation and workload evidence can change both.

The points estimate combines playing time with position-specific routes to goals, assists, clean sheets, saves, defensive contributions, appearance points and bonus, alongside potential deductions. Team strength, opponent and home or away context affect those opportunities. An attractive team fixture does not guarantee a useful individual role.

At the 6 October 2026 check, the production status endpoint identified xPts v1 and Dixon-Coles 3.1 as the live engines, with v2 distribution bands disabled. A candidate model or a favourable research backtest is not proof that the live tools use that candidate.

The current v1 floor and ceiling are indicative display ranges, not calibrated probability intervals or guaranteed minimum and maximum scores. Do not read them as a statement that a particular percentage of outcomes falls between the endpoints. The August chart preserves the labels and values displayed at that time; it does not establish their statistical coverage.

A practical example

In the original August example, Haaland had 7.91 GW1 xPts, with displayed floor and ceiling values of 6.02 and 9.80. His three-match estimate was 24.63, compared with Bruno at 22.31. The difference over that horizon was 2.32 projected points before captaincy.

Bruno cost £3.5m less in that comparison. Choosing him could be sensible if the saving improved the rest of the squad by more than the projected gap, but the chart does not contain every alternative squad needed to prove that claim. Captaincy also changes the comparison: doubling one player in a specific week is different from comparing uncaptained three-week totals.

Reading an accuracy scorecard

A model can have a modest average error across the whole player pool while missing by much more among the players you are choosing between. The GW5 scorecard makes that distinction visible: it evaluates the same frozen forecast across all captured players, those projected for at least 45 minutes, and the highest 25 forecasts. The groups overlap, so their sizes must not be added together.

Mean absolute error, or MAE, is the average size of a forecast miss, ignoring whether the prediction was too high or too low. The 659-player pool had an MAE of 1.51 points; among the 289 players projected for at least 45 minutes it was 2.63. For the top 25 forecasts it was 3.77. The larger decision-relevant errors are a reason to inspect the cohort, rather than use the smallest headline number as a claim of precision.

For the likely-player group, 37.7% of forecasts finished within two points of the result. That is an observed hit rate in one completed Gameweek. It is not a promise about the next Gameweek, and it does not validate the floor and ceiling as confidence limits.

Expected-minutes error averaged 24.5 minutes in that group. Minutes uncertainty therefore deserves a place beside the points forecast when comparing candidates. This table alone cannot show how much of the points error came from playing time, finishing, clean sheets or other events; that needs a player-level breakdown.

Inspect the complete GW5 scorecard, including signed bias and individual outcomes

Common mistakes

Keep the horizon consistent. A next-Gameweek estimate cannot be ranked directly against a three-Gameweek total, and a single-fixture forecast is different from a Double Gameweek total. Check the fixture list and the units before comparing values.

A fresh injury update does not automatically make an old forecast injury-aware. Check when the projection was captured and whether the affected role was included. The same applies to transfers, changing club assignments and postponed fixtures.

One correct captain or one failed pick cannot establish model accuracy. Compare frozen forecasts with outcomes across a stated cohort and period, including misses. The model-accuracy pages separate research evaluations from release decisions and describe the limits of each test.

Decision checklist

For a current decision, check the Gameweek horizon, capture time, model release, player price, expected minutes and availability. Compare affordable alternatives within the same squad constraints, then account for captaincy, transfer costs and bench cover.

Prefer a decision that remains reasonable when a doubtful starter loses minutes over a tiny numerical edge that depends on every uncertain assumption going right. Keep the forecast used for the decision so the eventual result can be reviewed without rewriting the original prediction.

See current projections and capture times

Read model evaluation results and release decisions