We publish our misses. Gameweek 5's picks went out on 16 September with 290 projections attached. Here is how they did.
- We beat the crowd by our widest margin yet. Ranking players by our model scores +0.5569 against the finished gameweek. Ranking them by ownership scores +0.3294. A 0.227 gap — more than triple Gameweek 4's 0.062.
- The squad fell further behind average. Our eleven scored 36 points with the captain doubled. The average FPL manager scored 48.
- The captain call was right, clearly, for once. Haaland, doubled, returned 12. Bruno Fernandes — the only other name either of our squads considered — returned 2.
How did our picks do?
| Pos | Player | Projected | Actual |
|---|---|---|---|
| GKP | Konstantinos Tzolakis | 4.4 | 6 |
| DEF | Silva | 5.7 | 3 |
| DEF | Gabriel | 5.2 | 1 |
| DEF | Anan Khalaili | 4.3 | 6 |
| MID | Bruno Fernandes (V) | 7.1 | 2 |
| MID | Bryan Mbeumo | 6.1 | 2 |
| MID | Marcus Tavernier | 5.5 | 2 |
| MID | Mamadou Sangaré | 4.9 | 0 |
| FWD | Erling Haaland (C) | 7.3 | 6 |
| FWD | João Pedro | 5.2 | 0 |
| FWD | Dominic Calvert-Lewin | 4.8 | 2 |
60.6 projected. 30 scored by the eleven, Haaland's 6 included at face value. 36 with the captain doubled. Two of the eleven beat their number. The best score in the game that week was 126.
That is our lowest total since Gameweek 3, and it falls further short of average than Gameweek 4 did: 12 points below, against 10 the week before. Only four of our top 20 projected players hauled (6 or more points) — the fewest of the five weeks so far — and ten of them scored two points or fewer. Five of those ten blanks were sitting in our own eleven.
Was the captain call right?
Yes, clearly, for once. Last week's piece gave two different captains depending on whether you read the headline or the squad table. This week there was only one name, because our best-over-the-horizon squad and our best-for-this-week-alone squad both landed on Haaland without being asked to.
Man City hosted Sunderland and won 5-3 — eight goals, the highest-scoring match of the round. Haaland played the full 90, scored one of City's five, and picked up no bonus: 6 points, doubled to 12.
Bruno Fernandes projected 7.1, close enough to Haaland's 7.3 that we called the gap noise. He played 90 minutes at Fulham and returned 2. Captaining Haaland over Fernandes was worth 10 points on the armband alone.
One caveat, same as Gameweek 3's: Haaland was also the most-captained player in the game again, and his ownership is up to 72.3%, from 71.2% the week before. Getting the model's own top two pick right is not a contrarian win — it's agreement with the crowd, not a result against it.
What happened to our two Manchester United picks?
Man Utd drew 1-1 away at Fulham. We owned two of their attacking players, Bruno Fernandes and Bryan Mbeumo, and neither was involved in the goal. Matheus Cunha scored it, took 3 bonus points, and returned 10 — he wasn't in our squad.
Fernandes and Mbeumo both played the full 90 minutes and returned 2 points each: the appearance fee, nothing else. Two of our four midfield picks came from the same match, and the one attacking return that match produced went to the one Manchester United player we didn't select.
Where were we most wrong?
| Player | Projected | Actual | Minutes |
|---|---|---|---|
| Bruno Fernandes | 7.1 | 2 | 90 |
| Gabriel | 5.2 | 1 | 90 |
| Bryan Mbeumo | 6.1 | 2 | 90 |
| Marcus Tavernier | 5.5 | 2 | 74 |
All four played essentially the whole match and simply didn't return. Gabriel's week is the clearest case: Arsenal lost 3-0 away at Brighton, so the clean sheet he was projected against never arrived, and the goals-conceded penalty ate most of what his appearance fee gave him.
Across all 290 players we projected, our average error was 2.00 points — the second-lowest of the five weeks. Our average bias was -0.02, essentially flat, and the first week it hasn't been positive. Not a pessimistic lean; just the first time it wasn't an optimistic one.
Two more picks didn't get the chance to be wrong the normal way. João Pedro carried no injury flag, no team news and a 100% expected-minutes read into the deadline, then didn't play a single minute — there was nothing in the pre-match data for the model to see this coming from. Mamadou Sangaré, the player who'd only just crossed our 270-minute eligibility line the week before, was substituted after 21 minutes and scored zero. Neither is a projection error in the usual sense. Both are the kind of miss a model built on pre-match data cannot close, and the honest thing is to name them separately rather than average them into the four above.
Who did we miss?
| Player | Owned | Projected | Actual |
|---|---|---|---|
| Antoine Semenyo | 15.5% | 5.3 | 17 |
| Brian Brobbey | 6.2% | 2.6 | 17 |
| Jay Dasilva | 0.7% | 2.4 | 15 |
| Pascal Groß | 18.5% | 3.7 | 14 |
| Carl Rushworth | 0.9% | 2.4 | 11 |
Jay Dasilva and Carl Rushworth both play for Coventry City — two of our five biggest misses from one side, in the same match. We are not claiming these five as differentials we called; three of five are owned by more than 6% of the game, Pascal Groß by nearly a fifth. We rated all five below where the gameweek put them, and the gap ran from 8.6 points to 14.4.
The scorecard
All 290 players, no filters. The measure is how well our order matched the real order.
| Ranking by… | GW1 | GW2 | GW3 | GW4 | GW5 |
|---|---|---|---|---|---|
| Our model | +0.413 | +0.376 | +0.426 | +0.476 | +0.557 |
| Ownership — what the crowd picked | +0.360 | +0.394 | +0.414 | +0.414 | +0.329 |
| Points per game so far | +0.396 | +0.218 | +0.230 | +0.270 | +0.198 |
| Price | +0.306 | +0.286 | +0.192 | +0.276 | +0.174 |
| Average error, points per player | 2.05 | 1.99 | 2.02 | 2.19 | 2.00 |
| Hauls (6+) in our projected top 20 | 6/20 | 9/20 | 7/20 | 10/20 | 4/20 |
This is the widest gap over the crowd we've published, and it came in the same week ownership itself fell. +0.557 against ownership's +0.329 is a 0.227 gap, more than triple Gameweek 4's 0.062. Ownership actually ranked the gameweek worse than it ranked any of the previous four — the crowd's own week, not just ours relative to it, was a bad one to follow.
None of that shows up in the squad's own score. Gameweek 4's review called itself the clearest split yet between ranking accuracy and squad outcome: the best ranking week of the season against a squad that still lost to the average manager by ten. Gameweek 5 is a clearer one still — a better ranking week again, and a wider loss to average again. Two gameweeks running, the two numbers are moving in opposite directions at the same time, not just independently of each other. Ranking 290 players correctly and picking the right eleven of them for one fixed week remain different tasks; right now they are pulling against each other, and we don't yet know if that is a pattern or two unlucky squads in a row.
Tell us where this is wrong
- Did you captain Haaland too? Most of the game did — he was the most-captained player in FPL again this week. Tell us if agreeing with both the model and the crowd changed how you read this week's piece.
- Did you own Cunha, or one of the two Manchester United picks we did? One of the three scored 10. The other two scored 2 each between them. We'd rather know which one happened to you.
- Build for the gap. Gameweek 6's deadline is Saturday 10 October, 11:00 BST — three weeks out, the longest gap of the season so far. Build your squad on the Team Planner while there's time to think about it properly, rather than three days before a deadline.
Every number here comes from the file our model wrote on 16 September, before the deadline, scored against the finished gameweek.