Gameweek 10 was a reminder that the gap between a good process and a good outcome can be enormous. Our top captain pick scored 2. Our top overall pick scored 2. And three of our top six predicted scorers all returned 2 points while finishing their matches. Here is the honest account of how our numbers held up, and where they fell apart.
How our predictions did
Quick explainer first. xP (expected points) is our prediction for how many points a player will score, frozen before the deadline. Correlation measures how well our rankings matched reality, from 0 (no link at all) to 1 (a perfect match).
See what to do with your own FPL team
Open fplwire for your teamFree. No account, no FPL password.On the surface, GW10 looks fine. The average predicted score across 670 players with a fixture was 1.31, and the average actual score was 1.37, a tiny gap. The overall correlation was 0.73, matching our season figure for GW10 exactly, which suggests our model is tracking the gameweek’s shape reasonably well in aggregate.
But that headline number is propped up by the players who didn’t play much, where predicting near-zero points for a non-starter is easy to get right. Restricting the correlation to players who actually played, the figure drops to 0.37. That is a much more honest picture of how we did at the sharp end, ranking the players who were actually out on the pitch. A 0.37 correlation among starters is weak, and it was a tough week for picking out the right performers even if the overall numbers look respectable.
Captaincy
Our top predicted scorer, and by extension our captain pick, was Haaland (MCI, FWD) at 8.6 expected points. He played the full 90 minutes and scored 2. That is about as bad as a captaincy call can go without the player missing the game entirely. Anyone following our highest expected points for the armband took a direct hit to their week.
The big scores
The week’s actual top scorers were mostly players we underrated or missed completely. Barry (EVE, FWD) top-scored with 16 points from a predicted 4.9, though we did at least have him ranked at #12 of 670, so the shape of our model saw some upside there even if the scale was off. Alysson (AVL, MID) also hit 16 points from just 1.8 predicted, and we ranked him down at #207, a total miss. Wirtz (LIV, MID) scored 13 from 3.7 predicted (#70), and Delap (NFO, FWD) scored 13 from 3.0 predicted (#124).
The most extreme miss was Dorgu (MUN, MID), who played only 24 minutes but still returned 11 points, against a prediction of just 0.2 and a rank of #461. There is no way a model built on pre-match expectation catches a return like that from limited minutes.
It wasn’t all misses, though. We did have N.Jackson (AVL, FWD) in our own top 10 at 5.2 predicted, and he delivered 11 points, our best hit of the week among the big scorers. Vuskovic (BHA, DEF) was ranked #19 with 4.7 predicted and returned 12, another one where our model was pointing in the right direction even if it undercooked the magnitude. Ødegaard (ARS, MID) at #46 and E.Le Fée (SUN, MID) at #91 also scored well without troubling our top picks, both reasonable misses in a volatile week rather than total blind spots.
The best XI in hindsight
The best possible 3-4-3 this week scored 132 points: Trafford (10), Vuskovic (12), Muharemović (10), Justin (9), Alysson (16), Wirtz (13), Ødegaard (11), Dorgu (11), Barry (16), Delap (13) and N.Jackson (11). Our predicted best XI, set up as a 4-3-3, actually returned 65 points: Raya (6), Gabriel (8), Tarkowski (2), Virgil (4), Vuskovic (12), Palmer (2), Rogers (2), Groß (10), Haaland (2), Isak (8) and Evanilson (9). That’s roughly half of the achievable ceiling, with Vuskovic the one player who made both lists.
Defensive contributions
Forty players picked up defensive contribution points this week, and they were decisive in several of the standout scores. Muharemović (LEE, DEF) led the way with 10 points including 2 for defensive contributions, followed by Justin (LEE, DEF) with 9, Mac Allister (LIV, MID) with 9, and Mukiele (SUN, DEF) with 9. Gabriel (ARS, DEF), Konsa (ARS, DEF), Mykolenko (EVE, DEF) and Elvedi (LEE, DEF) all reached 8 points with the same 2-point defensive contribution bonus attached. For defenders and defensive midfielders, these contributions are now a regular part of how a solid underlying week turns into a genuinely good score.
What we take from it
A few honest lessons from GW10. First, our model is better at identifying shape than scale, several of the week’s big scorers (Barry, Wirtz, Vuskovic) were on our radar in the rankings even where the predicted points undersold what followed. Second, the gap between overall correlation (0.73) and played-only correlation (0.37) shows that a lot of our apparent accuracy comes from correctly predicting low scores for players who barely feature, not from nailing the performances that matter. Third, defensive contributions are adding a layer of scoring that rewards players who might not top the shot or chance-creation charts, and that’s worth factoring into how we read a prediction. Finally, weeks like this one are a useful gut check: even a well calibrated model will occasionally miss its top captain pick entirely, and the only real defence against that is spreading risk rather than trusting any single highest-predicted name.
See what to do with your own FPL team
fplwire reads your live FPL team and shows you your captain options, transfer targets, chip timing and mini league standings in one place. Free, and no account needed.