Every Tuesday the projections are saved into the public data history before the next deadline. Once a gameweek is played, this page scores what was published against what actually happened. Nothing is re-run or re-tuned afterwards. The forecasts are the ones you could see at the time.
The comparison is FPL's own expected points (ep_next) from the same snapshot. Lower average error is better.
The forecast for each gameweek is the latest one published before its deadline
| GW | Published | Players | Avg error | FPL error | Error | Rank corr. | Top-20 avg | Model's captain | FPL's captain |
|---|---|---|---|---|---|---|---|---|---|
| Loading… | |||||||||
Scored players are everyone projected at least 1 point or who scored anything that week, which leaves out the long tail of squad players who were never going to feature. Avg error is the mean absolute difference between projected and actual points. Rank corr. is how well the order matched (Spearman, 1 is perfect). Top-20 avg is the average actual score of the 20 highest-projected players. Captain is the single highest projection. A few weeks is a small sample, so read the totals as early signal. For a full-season test see the walk-forward backtest.