Accuracy
Every gameweek's projections are frozen at the deadline and scored against what actually happened — next to FPL's own expected points, frozen at the same moment. No retro-fitting: once a deadline passes, the frozen file can't be touched.
Rank quality per gameweek
Spearman correlation between projected and realized points among players who played, within each position. 1 would be a perfect ordering; typical good weeks live around 0.3–0.4.
| GW | Fixtures | Rank corr | FPL | RMSE | FPL | Captain call | xPts vs actual |
|---|---|---|---|---|---|---|---|
| GW6 | 10 | 0.38 | 0.37 | 1.99 | 2.36 | Fernandes 1 · best 18 | -9% |
| GW7 | 10 | 0.39 | 0.22 | 2.00 | 2.47 | Haaland 13 · best 15 | -3% |
Under the hood, one gameweek at a time
Loading the gameweek…
How this is measured
Frozen at the deadline. After every nightly refresh, the engine writes each upcoming gameweek's projections to a frozen file — and refuses to touch that file once the gameweek's deadline has passed. What gets scored is exactly what this site was serving when the deadline hit.
Scored when final. A gameweek is scored only after FPL marks it finished and data-checked (bonus confirmed). Double-gameweek players are compared on their full week; fixtures postponed out of a gameweek are excluded and disclosed in the fixtures column, never quietly dropped.
The baselines. FPL publishes its own expected points (ep_next) — it's frozen at the same deadline and scored the same way. Rank correlation is measured within each position among players who actually played; RMSE counts everyone projected, including the ones who never left the bench.
season 2026-27 · scored 2026-10-22 01:15 · latest freeze 2026-10-21 02:35