Forecast accuracy
How pre-deadline point estimates compare with completed FPL gameweeks.
Accuracy results are being collected
The first results will appear after Gameweek 6 finishes and final points are imported. Until then, the method below explains what the model measures.
What we'll measure
Latest pre-deadline forecast for the top 100 projected players compared with final gameweek points; no retrospective predictions. Mean absolute error is the average distance between predicted and actual points; lower is better. A positive mean error means the model over-predicted. Forecasts blend playing time, expected goals and assists, clean-sheet chances, saves, bonus and defensive-contribution history with a regressed form baseline and opponent results. They are estimates, not guaranteed returns.
Only forecasts saved before a gameweek deadline are scored. Earlier gameweeks are deliberately excluded because reconstructing predictions from later data would give a misleading accuracy claim. The new model has not yet accumulated a completed scored gameweek, so its weights are provisional.
How player rankings differ from xPts
The 0–100 ranking score weights recent form and season points at 25% each, expected goal involvement at 15%, and playing time, upcoming fixture difficulty and value at 10% each, with availability at 5%. Most components are normalised within each position, so a player’s rank can move when another player’s form changes. This is a comparison score, not a points forecast.
xPts is the separate component-forecast-v4 estimate for the next open gameweek. It uses opponent-aware projections and is saved before the deadline. If no model forecast exists, xPts is shown as “—”, never replaced with FPL Form.
Awaiting final results
GW6: 667 forecasts · GW7: 667 forecasts · GW8: 667 forecasts · GW9: 667 forecasts · GW10: 667 forecasts
Data refresh
Latest successful import: . 667 players and 380 fixtures in the current snapshot. Last import outcome: succeeded.