We publish our misses. Gameweek 4's picks went out on 9 September with 289 projections attached. Here is how they did.
- We beat the crowd by the widest margin yet. Ranking players by our model scores +0.4764 against the finished gameweek. Ranking them by ownership scores +0.4143. A 0.062 gap — the biggest of the four weeks we've published.
- The squad still scored below average. Our eleven put up 59 points with the captain doubled. The average FPL manager scored 69.
- The article gave two different captains again. The TL;DR led with Cole Palmer. The printed squad captained Erling Haaland. Palmer returned 5. Haaland returned 9, doubled to 18. An 8-point gap depending which line of the piece you read.
How did our picks do?
| Pos | Player | Projected | Actual |
|---|---|---|---|
| GKP | Konstantinos Tzolakis | 4.2 | 2 |
| DEF | Maxim De Cuyper | 4.9 | 11 |
| DEF | Riccardo Calafiori | 4.7 | 6 |
| DEF | Silva | 4.4 | 3 |
| MID | Bruno Fernandes (V) | 6.6 | 2 |
| MID | Marcus Tavernier | 5.5 | 8 |
| MID | Bryan Mbeumo | 5.5 | 2 |
| MID | Iliman Ndiaye | 5.0 | 1 |
| MID | Elliot Anderson | 4.8 | 5 |
| FWD | Erling Haaland (C) | 6.8 | 9 |
| FWD | Igor Thiago | 5.1 | 1 |
64.5 projected. 50 scored by the eleven, Haaland's 9 included at face value. 59 with the captain doubled. Three of the eleven beat their number. The best score in the game that week was 151.
That is a big improvement on Gameweek 3's 33. It is still ten points short of what an average squad managed, because Gameweek 4 was a high-scoring week across the board — the average FPL manager scored 69, against 51 the week before.
Was the captain call right?
This is where it gets uncomfortable, and it's the second week running.
The TL;DR at the top of the Gameweek 4 piece said: "Captain Cole Palmer. We project 7.29." Palmer played 90 minutes and scored 5.
The printed £100m squad, two paragraphs later, captained Erling Haaland at 6.8 — doubled, because the article also said "if you already own Haaland, captain him," and the squad itself owned him. Haaland scored 9, doubled to 18.
Read only the headline and you banked 10 points from your armband. Read as far as the table and you banked 18. That is the same fault as Gameweek 3, with the right and wrong answers swapped. In Gameweek 3 the prose named Haaland and the squad table captained Bruno Fernandes — the prose was right, the table was wrong. In Gameweek 4 the headline named Palmer and the squad table captained Haaland — the table was right, the headline was wrong. Which half of the piece to trust has now flipped twice in two weeks. Same root cause both times: the model-wide top pick and the pick inside the optimal squad are not always the same player, and the piece has twice now stated both without resolving which one a reader should act on.
Haaland's case was also the better one on the day. Manchester City won away at Manchester United, 1-0, and he scored the goal.
What happened to the Man Utd v Man City bet?
We flagged it going in: two Manchester United players and the Manchester City captain sat in the same eleven, with 40% of the week's projected total — 25.8 of 64.5 points — riding on one game.
Manchester City won 1-0. Bruno Fernandes and Bryan Mbeumo, both Manchester United, scored 2 each. Haaland, doubled as captain, scored 18. One player captured 18 of the eleven's 59 points. The two players on the losing side combined for 4.
Same lesson as Gameweek 3's Manchester City read: picking the team that controls a 1-0 is not the same as picking which of their players it pays. This time it was our own captain who was on the right side of that coin, and the two players who weren't cost the squad more than the shirt price would suggest.
Where were we most wrong?
| Player | Projected | Actual | Minutes |
|---|---|---|---|
| Bruno Fernandes | 6.6 | 2 | 90 |
| Igor Thiago | 5.1 | 1 | 90 |
| Bryan Mbeumo | 5.5 | 2 | 90 |
| Iliman Ndiaye | 5.0 | 1 | 45 |
Three of those four played the full 90 and simply didn't return. No minutes excuse, no team-sheet surprise — the model rated the fixture and the role, and the player didn't convert either into points. Ndiaye is the partial exception: 45 minutes is a substitution, not a start withheld, and we don't see team-talks any more than we see team sheets.
Across all 289 players we projected, our average error was 2.19 points, the highest of the four weeks so far. We project 0.10 high on average — smaller than Gameweeks 2 and 3, though not Gameweek 1's 0.06, and in the same direction as every week so far: we lean optimistic, not pessimistic.
Who did we miss?
| Player | Owned | Projected | Actual |
|---|---|---|---|
| Pascal Groß | 16.2% | 4.4 | 17 |
| Jayden Bogle | 1.3% | 2.9 | 15 |
| Kevin Schade | 4.0% | 4.1 | 15 |
| Leif Davis | 5.1% | 3.4 | 14 |
| David Raya | 38.8% | 3.6 | 14 |
Jayden Bogle is on this list two weeks running. He scored 14 in Gameweek 3 and 15 in Gameweek 4, at 0.5% and 1.3% owned respectively. We projected him at 1.7 and 2.9. Whatever Leeds are doing defensively right now, our model is two weeks behind it — his rate is built almost entirely from last season's Championship record, and the 270-minute eligibility rule means his live Premier League form only starts to outweigh that prior gradually, one gameweek at a time.
Six of our under-10%-owned players scored 8 or more. We are not claiming these as differentials we called — we project cheap players who are efficient for their price, not players who are about to haul. Bogle is the clearest evidence yet that at least one of them is doing more than that, and we haven't caught up.
The scorecard
All 289 players, no filters. The measure is how well our order matched the real order.
| Ranking by… | GW1 | GW2 | GW3 | GW4 |
|---|---|---|---|---|
| Our model | +0.413 | +0.376 | +0.426 | +0.476 |
| Ownership — what the crowd picked | +0.360 | +0.394 | +0.414 | +0.414 |
| Points per game so far | +0.396 | +0.218 | +0.230 | +0.270 |
| Price | +0.306 | +0.286 | +0.192 | +0.276 |
| Average error, points per player | 2.05 | 1.99 | 2.02 | 2.19 |
| Hauls (6+) in our projected top 20 | 6/20 | 9/20 | 7/20 | 10/20 |
This is the widest gap over the crowd we've published. +0.476 against ownership's +0.414 is a 0.062 gap, more than five times Gameweek 3's 0.012. Ten of our top 20 hauled, the most of any week so far.
None of that shows up in the squad's own score. Ranking players correctly and picking eleven of them for one fixed gameweek are different tasks — the first is what the scorecard measures, the second is what a real squad has to do, bench weighting, budget and all. Gameweek 4 is the clearest split between the two we've published: our best ranking week, and a squad that still lost to the average manager by ten points.
Tell us where this is wrong
- Which line did you read — the headline or the table? If you captained Palmer off the TL;DR, you got 10 points from your armband. If you captained Haaland off the squad, you got 18. We'd rather know which one happened to you than guess.
- Did you own Bogle in Gameweek 3, Gameweek 4, or neither? He's hauled both weeks from under 2% ownership. Tell us what you're seeing in Leeds' shape that a 270-minute eligibility rule can't yet.
- Build Gameweek 5 yourself on the Team Planner. Gameweek 5's picks are up, and this time the model agrees with itself about the captain.
Every number here comes from the file our model wrote on 8 September, before the deadline, scored against the finished gameweek.