FPL Dave Projection Accuracy
Projected points are frozen before every deadline and graded here after the gameweek, next to two simple baselines. 1 gameweek graded in 2026/27.
2026/27 so far: 1 gameweek graded
1.87
FPL Dave avg error (pts)
2.06
Season-average baseline
1.99
Form baseline
1
Captain picks, total pts
Average error is in points per player, and lower is better. The model has the lowest error of the three so far. Against the most-captained player, the captain pick scored more in 0 gameweeks, the same in 0 and fewer in 1. The 10 highest-projected players averaged 4.20 points a gameweek; the 10 most-owned averaged 7.10.
Gameweek by gameweek
Average error in points per player. The lowest of the three in each row is in bold.
How the record is kept
- Freeze. Within 9 hours of the deadline, projected points for every player with a fixture are written to a file. The projection comes from the same engine the app uses, including the market-odds blend.
- Lock. A frozen file is never rewritten. A file whose timestamp is not before its deadline is ignored.
- Grade. Once FPL marks the gameweek final (bonus points included), each projection is compared with the points actually scored.
- Compare. The same error is worked out for two baselines saved at the same moment: the player's season average and FPL's form figure.
Frequently Asked Questions
How is FPL Dave's accuracy measured?
Within 9 hours of each FPL deadline, the projected points for every player with a fixture are saved to a file that is never changed afterwards. Once the gameweek is final, each projection is compared with the points the player actually scored. The headline figure is the average error in points per player: lower is better.
Why does the projection have to be frozen before the deadline?
A projection recalculated after the matches would use stats that already include those matches, which makes any model look better than it was. Freezing first is the only way a track record means anything. Each row shows when its snapshot was taken and links to the raw file.
What are the baselines?
Two simple guesses saved at the same moment: the player's season average (points per game) and FPL's own form figure. If the model cannot beat "assume he scores his average", it is not adding much. Where a baseline did better in a gameweek, the table shows it.
What counts as a good average error?
FPL points are noisy. A single goal is worth 4 to 6 points, so any model will often be a few points out for an individual player in a single gameweek. What matters is whether the model beats the baselines over many gameweeks, not any one week.
How is the captain pick judged?
The captain pick is the player with the highest projected points in the frozen file. It is compared with the most-captained player across all FPL managers that gameweek, and with the highest scorer in hindsight.