Model points are an experimental estimate, not a promise. They estimate how many FPL points each player will score in the next gameweek and over the next five. Captain Picks, Transfer Suggestions, Points Projections and Rival Radar all read from the same numbers.
What goes in
- Points per appearance this season, and FPL form (average points per match over the last 30 days).
- Expected goal involvement (xG + xA) per 90 minutes, counted once a player has 180 minutes and capped at 1.5.
- Share of the available minutes the player has actually played.
- The official FPL fixture difficulty, home or away, and price.
Each position has its own set of weights, fitted on the 2023/24 and 2024/25 seasons. The estimate is then multiplied by the chance of playing that FPL publishes; injured, suspended and unavailable players get zero. A double gameweek adds both fixtures; a blank scores zero.
How we tested it
We compare the model with FPL's own estimate this season, which follows form, on gameweeks the model never saw. We only score regular starters: players with at least half the available minutes, three earlier matches, and one fixture that gameweek. Correlation runs from 0 (no link) to 1 (perfect); higher is better. Error (RMSE) is the typical miss in points; lower is better. "Top pick" is the average real score of the player each method ranked first in each gameweek.
This season, GW5–9, with the weights the site uses today
| Position | Correlation: model | Correlation: form | Error (RMSE): model | Error (RMSE): form | Top pick: model | Top pick: form | Player-gameweeks |
|---|---|---|---|---|---|---|---|
| GKP | 0.26 | 0.14 | 2.92 | 3.27 | 2.8 | 3.0 | 102 |
| DEF | 0.30 | 0.20 | 3.14 | 3.50 | 5.0 | 3.8 | 426 |
| MID | 0.23 | 0.16 | 2.76 | 3.17 | 3.6 | 5.0 | 459 |
| FWD | 0.31 | 0.18 | 3.60 | 4.01 | 7.0 | 4.8 | 100 |
Three past seasons, refitted each time
We fitted the same recipe on two of the 2022/23, 2023/24 and 2024/25 seasons and tested it on the third, three times over. The table shows the average. This checks the recipe rather than the exact weights in use.
| Position | Correlation: model | Correlation: form | Error (RMSE): model | Error (RMSE): form | Top pick: model | Top pick: form | Player-gameweeks |
|---|---|---|---|---|---|---|---|
| GKP | 0.24 | 0.18 | 2.82 | 3.19 | 3.6 | 3.1 | 624 |
| DEF | 0.28 | 0.21 | 2.77 | 3.12 | 4.4 | 3.8 | 2,487 |
| MID | 0.31 | 0.24 | 3.05 | 3.41 | 6.5 | 5.4 | 2,866 |
| FWD | 0.30 | 0.21 | 3.57 | 4.04 | 5.6 | 4.5 | 632 |
What this does and does not show: the model's estimates line up with real points better than form at every position, in both tests. The gap is modest, and FPL scores are noisy, so no estimate is close to certain. This season's top midfield pick by the model averaged 3.6 points against 5.0 for form; that is five gameweeks of data, and over the three past seasons the model's top midfield pick did better (6.5 against 5.4). Goalkeepers carry a "limited validation" label because this season's sample is small.
Known limits
- Defensive contribution (DEFCON) points have no term of their own: the fitting seasons predate them.
- The five-gameweek total reuses today's form for every week.
- After the next gameweek, availability follows the return date in FPL's injury news, or a doubt clearing after one week. This rule is not yet tested.
- The test uses current prices, which partly reflect recent form, so it flatters the model slightly.
- The model knows nothing FPL does not publish: team news after the last update, rotation hints in press conferences, or a new manager.
How each tool uses it
- Captain Picks ranks midfielders and forwards who have a fixture and are not doubts, by next-gameweek model points.
- Transfer Suggestions lists a swap only when it gains at least 2 model points over five gameweeks. When we replayed past seasons, smaller gains were not worth a transfer, and real gains came in at about half the estimate.
- Price Watch does not use the model: it shows FPL's own price-change projections.
Corrections and changes
When we find an error we fix it and log the change here with its date. Spotted one first? Email founder@fplpicks.com.
- 6 Nov 2026 — v1 tested position by position on 2026/27 GW5–9; defenders no longer marked unvalidated.