Deadline: Friday 4 September, 18:30 BST. Ipswich host Liverpool ninety minutes later.
Short version, if that's all you have time for:
- Captain Erling Haaland. 7.9 projected, at home to the worst-rated team in the league.
- The model wants six Man City players in its top eight. You are allowed three.
The Gameweek 3 fixtures
Difficulty is FPL's own rating, 1 (easiest) to 5 (hardest), from each side's point of view.
| Kick-off (BST) | Home | Away | ||
|---|---|---|---|---|
| Fri 4, 20:00 | Ipswich | 4 | Liverpool | 2 |
| Sat 5, 12:30 | Newcastle | 3 | Bournemouth | 3 |
| Sat 5, 15:00 | Brighton | 2 | Leeds | 3 |
| Sat 5, 15:00 | Fulham | 3 | Crystal Palace | 3 |
| Sat 5, 15:00 | Man City | 2 | Coventry | 5 |
| Sat 5, 15:00 | Brentford | 2 | Sunderland | 3 |
| Sat 5, 15:00 | Nott'm Forest | 3 | Spurs | 3 |
| Sat 5, 17:30 | Hull City | 3 | Aston Villa | 2 |
| Sun 6, 14:00 | Everton | 4 | Man Utd | 3 |
| Sun 6, 16:30 | Arsenal | 4 | Chelsea | 5 |
Man City v Coventry, 15:00 Saturday, is the fixture this gameweek turns on. Coventry rate 20th of 20 in our team model, have lost both games and have not scored. City have won both, scoring six.
Two other things to plan around. Liverpool play first, Friday night, so a Liverpool captain locks in before anything else kicks off. And Arsenal v Chelsea is Sunday's 16:30 — the hardest fixture on the board to call, and the last chance for anything in your team to score.
The full grid, all season, is on Fixtures.
Show us your Gameweek 3 best draft — here's ours
The model's answer to "if you were starting from scratch on Friday, what would you own?" A solver working the real rules — fifteen players, 2/5/5/3, £100.0m, max three per club — picking the best starting eleven across Gameweeks 3 to 10.
£100.0m exactly. 68.0 projected points for the GW3 eleven, captain included. 4-4-2.
| Pos | Player | Club | Price | Owned | GW3 | GW3–10 |
|---|---|---|---|---|---|---|
| GKP | Caoimhín Kelleher | BRE | £5.0m | 7.3% | 4.3 | 32.0 |
| DEF | Marc Guéhi | MCI | £6.0m | 18.9% | 6.2 | 39.9 |
| DEF | Gabriel | ARS | £8.0m | 26.2% | 5.7 | 45.6 |
| DEF | Malick Thiaw | NEW | £5.0m | 2.0% | 4.6 | 38.2 |
| DEF | James Tarkowski | EVE | £6.0m | 10.5% | 4.3 | 37.9 |
| MID | Bruno Fernandes (C) | MUN | £12.0m | 48.6% | 7.1 | 61.7 |
| MID | Iliman Ndiaye (V) | MCI | £6.0m | 13.4% | 6.9 | 45.9 |
| MID | Phil Foden | MCI | £7.0m | 5.2% | 6.8 | 44.3 |
| MID | Bryan Mbeumo | MUN | £8.0m | 29.5% | 5.3 | 46.0 |
| FWD | Igor Thiago | BRE | £8.0m | 14.4% | 5.4 | 43.2 |
| FWD | Dominic Calvert-Lewin | LEE | £6.0m | 26.4% | 4.2 | 36.0 |
Bench, in order: Cole Palmer (CHE, £9.6m), Luke O'Nien (SUN, £4.0m), Joshua Zirkzee (MUN, £5.4m), and Dan Bentley (COV, £4.0m) in goal.
Three things worth explaining, because a solver output with no reasoning attached is just a list.
It doesn't own Haaland, and we're still telling you to captain him. £15.5m is 15.5% of the budget on one player, and across eight weeks the solver would rather have Bruno Fernandes at £12.0m and spend the difference elsewhere. This squad answers "starting from zero", which almost nobody is. If you already own Haaland, keep him and captain him.
Palmer benched at £9.6m. He has the best eight-week total of any midfielder outside the eleven (48.6) and still sits this week out, because Chelsea go to Arsenal. Buy-and-hold, not a Gameweek 3 starter.
Ask the same solver for Gameweek 3 alone and it shares only 7 of 15 with this squad — against 10 of 15 last week. That one-week version spends £15.5m on Haaland, captains him, and projects 71.4. Eight players moving on a change of horizon is a warning that this week is unusually fixture-driven, and one fixture is doing most of it.
Build your own on the Team Planner — real rules, saved in your browser, no account, and it gives you a link to share.
Who should I captain in Gameweek 3?
Erling Haaland, 7.9 projected — the highest number the model has produced for anyone this season. City are at home to the worst-rated team in the division and he takes their penalties. At 71% owned this is the safe pick rather than the clever one, and the model doesn't think this is a week to be clever.
| Player | Pos | Club | Price | Owned | GW3 |
|---|---|---|---|---|---|
| Erling Haaland | FWD | MCI | £15.5m | 71.0% | 7.9 |
| Bruno Fernandes | MID | MUN | £12.0m | 48.6% | 7.1 |
| Iliman Ndiaye | MID | MCI | £6.0m | 13.4% | 6.9 |
| Phil Foden | MID | MCI | £7.0m | 5.2% | 6.8 |
| Antoine Semenyo | MID | MCI | £8.5m | 20.5% | 6.3 |
| Enzo Fernández | MID | MCI | £6.9m | 3.4% | 6.3 |
The gap from Haaland to Bruno is 0.8 points, and our own average error is 2.1. So these are two defensible captains and the model prefers one of them slightly. If you own Bruno and not Haaland, it is not asking you to take a hit.
What it is saying clearly: six of the eight highest-projected players in the game play for Manchester City, and the rules let you own three. That constraint, not the captain call, is the real puzzle this week.
Best value on the board
Projected GW3 points per £1m.
| Player | Pos | Club | Price | Owned | GW3 | Per £m |
|---|---|---|---|---|---|---|
| Iliman Ndiaye | MID | MCI | £6.0m | 13.4% | 6.9 | 1.15 |
| Maxim De Cuyper | DEF | BHA | £4.7m | 15.2% | 4.9 | 1.05 |
| Marc Guéhi | DEF | MCI | £6.0m | 18.9% | 6.2 | 1.04 |
| Antonee Robinson | DEF | FUL | £4.5m | 1.6% | 4.4 | 0.97 |
| Phil Foden | MID | MCI | £7.0m | 5.2% | 6.8 | 0.97 |
| Malick Thiaw | DEF | NEW | £5.0m | 2.0% | 4.6 | 0.93 |
Ndiaye at £6.0m is the pick of the week. A Man City midfielder at a Brighton price, 13% owned, in the fixture the model likes most on the board. If you want City exposure and can't afford a third premium, that is where it points.
Differentials
Under 5% owned, ranked over eight gameweeks rather than one — a differential bought for a single week is a coin toss.
| Player | Pos | Club | Price | Owned | GW3 | GW3–10 |
|---|---|---|---|---|---|---|
| Marcus Tavernier | MID | BOU | £6.0m | 2.3% | 4.8 | 40.9 |
| Enzo Fernández | MID | MCI | £6.9m | 3.4% | 6.3 | 39.9 |
| Malick Thiaw | DEF | NEW | £5.0m | 2.0% | 4.6 | 38.2 |
| Sven Botman | DEF | NEW | £5.0m | 0.6% | 4.2 | 35.4 |
| Kevin Schade | MID | BRE | £6.0m | 4.0% | 4.5 | 35.1 |
| Nathan Collins | DEF | BRE | £5.5m | 2.2% | 4.9 | 34.1 |
Enzo Fernández at 3.4% owned is the interesting one — a City midfielder in the best fixture of the week, in fewer than one team in twenty-five.
Tavernier tops this table for the third week running, and his record on it is split: we said 3.9 in GW1 and he scored 10, then said 5.3 in GW2 and he scored 1. Both at 90 minutes. That is what a 2.1-point average error looks like from the inside.
More, updated daily, on Differentials.
Who's flagged
Check Watkins first. He has left the Premier League permanently and is still in 5.1% of teams — a guaranteed zero every week until you fix it, and the transfer market will not do it for you.
| Player | Club | Owned | Status |
|---|---|---|---|
| Ollie Watkins | AVL | 5.1% | Left the league — joined Al Hilal |
| Joe Rodon | LEE | 2.9% | Hamstring — 50% |
| Jean-Philippe Mateta | CRY | 2.9% | Hamstring — back 11 Oct |
| Moisés Caicedo | CHE | 2.1% | Unspecified — 75% |
| Jack Hinshelwood | BHA | 1.4% | Unspecified — 50% |
| Dan Burn | NEW | 1.2% | Ankle — back 14 Sep |
Who the model can't see
We need 270 minutes of Premier League record — three full matches — before we publish a number on a player. Invent one and you are guessing with extra steps.
That means 104 players who have already featured this season have no projection from us, and the best of them are the promoted clubs' defenders:
| Player | Club | Pos | Price | Owned | Points so far |
|---|---|---|---|---|---|
| Semi Ajayi | HUL | DEF | £4.1m | 8.2% | 20 |
| Konstantinos Tzolakis | HUL | GKP | £4.6m | 7.2% | 20 |
| Mamadou Sangaré | BRE | MID | £5.7m | 13.5% | 18 |
| John Egan | HUL | DEF | £4.0m | 4.6% | 17 |
| Nobel Mendy | HUL | DEF | £4.0m | 5.5% | 16 |
Four of those five are Hull defenders. Hull have the best-value defence in the game right now and we have nothing useful to say about it — the honest cost of refusing to make a number up. It fixes itself as the season plays: Sangaré qualifies after one more start. Treat silence from us about a promoted club as ignorance, not a verdict.
Watch the players you're deciding between
Two of the calls above are close enough to be worth checking again on Friday — Haaland against Bruno, and whether Rodon's 50% becomes a start.
Every player page has a Watch button, and every row of the players table has a star. Watching a player snapshots his price and injury news at that moment, and the next time you open his page it tells you what changed since — the price moved, the flag cleared, the news updated. No account, no email; it lives in your browser.
It exists because the honest problem with a stats site is that the page looks the same every time you open it. A watched player's page doesn't.
How did the model do in Gameweek 2?
Mixed, and one part of it is a loss.
We projected 9.4 for our Gameweek 2 captain, Bruno Fernandes, after he had blanked in GW1 and 139,000 managers sold him. He scored 23. That is the best call we've made.
Then the uncomfortable half. We scored all 278 players we published a number for, unfiltered:
| Ranking by… | GW1 | GW2 |
|---|---|---|
| Our model | +0.413 | +0.376 |
| Ownership — what the crowd picked | +0.360 | +0.394 |
| Last season's points per game | +0.396 | +0.218 |
| Mean error, points per player | 2.05 | 1.99 |
The crowd beat us. Ownership ranked Gameweek 2 better than our model did, 0.394 against 0.376. It is a 0.018 gap on one gameweek, nowhere near enough to conclude the crowd is better — but we would have called 0.018 our way a win, so we are calling this one a loss. Two gameweeks, one each.
The rest of the card: mean error 1.99, better than GW1's 2.05. Nine of our projected top 20 returned 6+ points, against six in GW1. Two of the top 20 didn't play at all. Over the 219 who actually got on the pitch, the correlation falls to +0.338.
That last number is the model's weakest joint and it has not moved: three of our six worst misses never got on the pitch. Predicting a footballer's points is close to solved next to predicting whether the manager picks him. You can watch that gap update daily on Over/underperformers.
What this is, and what it isn't. These weekly numbers run at a point in the season where the model's real backtest cannot — that gate needs eight played gameweeks and there are two, so this has not passed our evaluation gate. The record it falls back on is the 2025/26 walk-forward: a rank correlation of 0.359 against 0.268 for a points-per-game baseline, beating that baseline in 31 gameweeks out of 31. Two gameweeks is two gameweeks. We will have thirty-six more before any of this means anything.
Tell us where this is wrong
Three specific asks, and they are the reason this is a page rather than just a projection file.
- Show us your Gameweek 3 draft and why. Build it on the Team Planner, share the link, and name the one pick you would defend against the model. The disagreements are the interesting part — if you are not captaining Haaland, we want to know what you are seeing that we are not.
- Tell us when a number looks wrong. Not "the model is bad" — this player, this figure, here's why. That is the most valuable message we can get.
- Tell us what you would want to watch. The Watch button tracks price and news today. If it should tell you about a fixture swing, a set-piece change or a press conference, say so.
Send it through the feedback form for anything specific, or join the Discord if it needs a back-and-forth. It is read every day.
Next: Gameweek 4 picks and the GW3 scorecard. In the meantime — Players for the full database, Fixtures for the difficulty grid, Team Planner to build your fifteen, and My Team to check your own score against this.