Nothing scored yet · 0 of 1 published weeks played
One week has been published and it has not been played. Below are the measures each week will be scored on, fixed before any results exist.
One row per published gameweek, added after the week is played.
| Week | Model squad, actual | Entered squad, actual | Difference | MAE | RMSE | Spearman |
|---|---|---|---|---|---|---|
| GW1 · 2026/27 | Published 2026-08-21; awaiting the result. Declared in advance: objective 186.87 for the model’s squad against 178.24 for mine, a gap of -8.63. | |||||
| Season to date | — no completed weeks — | |||||
Each is computed over the fifteen named players in each published squad, against their real FPL scores, and reported per week and cumulatively. If a measure is added later, the date it was added is recorded and it is backfilled across every prior week, or it is not used.
| Measure | Symbol | What it tells you | Value |
|---|---|---|---|
| Mean absolute error | MAE | Mean of |predicted − actual| over every player in the squad, per gameweek. The plainest measure, in points, of how far off a typical player was. | — |
| Root mean squared error | RMSE | Punishes large misses more than small ones. Reported next to MAE because the gap between them says whether the model is steadily off or occasionally catastrophic. | — |
| Bias | ME | Mean signed error. If this is not near zero the model is systematically optimistic or pessimistic, which is a different and more fixable fault than being noisy. | — |
| Spearman rank correlation | ρ | Whether the model ordered players correctly, ignoring whether the magnitudes were right. Ordering is what actually picks a squad. | — |
| Calibration of P(start) | — | Of the players given a 0.9 chance of starting, did about nine in ten start? A probability that is not calibrated is not a probability. | — |
| Squad points, model vs. entered | pts | What each of the two squads really scored, under real FPL rules, with the captain and bench as each was actually set. | — |
Each overrule is scored on its own. A squad can win while every call inside it was wrong, and a set of good calls can lose.
| # | Call, as declared before the deadline | Verdict |
|---|---|---|
| 1 | Belgian → Premier League attack factor set to 0.58 | — |
| 2 | Bruno Fernandes fitness factor set to 0.85 | — |
| 3 | Diop P(start) set to 0.85 | — |
| 4 | Raya started over Lammens, against the model | — |
| 5 | Le Fée over Ampadu | — |
| 6 | Egeli over Thomas-Asante | — |
| 7 | Tzolis captained, decided on the day | — |
Declared on the gameweek 1 page before the deadline of 2026-08-21T17:30:00Z.