Written by a human, with a model doing the arithmetic.
Most first-draft advice is somebody's opinion about who you should own. What it usually doesn't admit is how much maintenance that draft needs from you afterwards.
We all sit somewhere on the risk-to-effort scale. If you're a felt cute, might check my team in a few weeks manager — genuinely, love that for you — you're probably better off drafting the highest-owned players your budget allows and filling the gaps with names you've never heard of and the few shillings you have left.
If you're reading this, that's probably not you. My guess is you're one of two types. Either you enjoy the friendly competition in your group chat, you'd like the bragging rights, but mostly you're trying not to come last. Or you're a proper nerd — every game, every highlight, we've seen it; the latest news and hot tips, we saw it coming.
We don't discriminate. Both are welcome. What this site tries to do is surface the data you actually need to make the call yourself. So let's get into it.
A caveat that matters, up front. Our projection model normally can't run before a season starts. It trains on gameweeks already played and needs eight of them to warm up. With zero fixtures played it stops cleanly, which is correct behaviour. So we ran its machinery at the season boundary instead: team strengths fitted on all 380 finished fixtures of 2025/26, every player's rates carried to whichever club he's at now, then a solver picking the best legal fifteen under the real rules.
None of this has passed our evaluation gate. It isn't in the model ledger. The caveats are at the bottom and they aren't decoration — one of them decides the goalkeeper.
Everything below was re-run on 18 August 2026, with that morning's prices and ownership.
First, the thing almost everyone gets backwards
Haaland is in 70.9% of squads. Before a ball is kicked, seven managers in ten have already made the same decision.
Here's the bit that trips people up: ownership does not change your expected points. It can't.
Your rank is your score minus the field's score. The field's score doesn't depend on which squad you pick. So maximising expected points already maximises expected rank, and there is no ownership adjustment that improves on it.
What ownership changes is the shape of the outcome, not its average. Own the template and your rank moves slowly in both directions. Avoid it and your rank swings hard.
So "fade Haaland because 71% own him" is a bet, not an edge. That's not an argument against it. It's an argument for knowing the price — which we can now quote exactly.
Squad A — the model's free choice
£100.0m exactly. 538.6 projected starting-XI points across GW1–8, captain doubled.
| Pos | Player | Club | Price | Owned | GW1 | GW1–8 | Starts |
|---|---|---|---|---|---|---|---|
| GKP | Roefs | SUN | £5.0m | 4.8% | 4.3 | 32.4 | 5/8 |
| GKP | Verbruggen | BHA | £4.5m | 18.8% | 3.9 | 31.1 | 3/8 |
| DEF | Gabriel | ARS | £8.0m | 28.7% | 6.7 | 46.3 | 8/8 |
| DEF | Guéhi | MCI | £6.0m | 20.4% | 4.8 | 38.3 | 7/8 |
| DEF | Thiaw | NEW | £5.0m | 1.9% | 4.5 | 37.9 | 7/8 |
| DEF | Senesi | TOT | £6.0m | 9.6% | 4.2 | 37.1 | 5/8 |
| DEF | Tarkowski | EVE | £6.0m | 9.5% | 5.3 | 36.9 | 6/8 |
| MID | B.Fernandes (C) | MUN | £12.0m | 48.8% | 8.3 | 62.1 | 8/8 |
| MID | Bruno G. | ARS | £7.0m | 9.1% | 6.7 | 44.6 | 8/8 |
| MID | Mbeumo | MUN | £8.0m | 30.2% | 5.9 | 44.3 | 8/8 |
| MID | Enzo | CHE | £7.0m | 5.3% | 5.1 | 43.3 | 7/8 |
| MID | Semenyo | MCI | £8.5m | 26.6% | 5.3 | 41.0 | 8/8 |
| FWD | Thiago | BRE | £8.0m | 17.5% | 5.9 | 43.6 | 8/8 |
| FWD | Kusi-Asare | FUL | £4.5m | 7.2% | 0.4 | 2.9 | 0/8 |
| FWD | Neave | NEW | £4.5m | 0.7% | 0.3 | 2.2 | 0/8 |
Bold = the core seven. We swept the planning horizon at 1, 3, 5 and 8 gameweeks. Roughly a third of the squad changes every time you move the window. Only these seven are picked at every horizon. They cost £54.0m of the £100.0m, and they are the only part of this squad we'd defend hard.
The other eight slots are inside the noise. Fill them to taste.
Squad B — the best team we can build around Haaland
If you're keeping him, this is what the same solver does with the constraint that he must be in the squad. Same £100.0m, same rules.
| Pos | Player | Club | Price | Owned | GW1 | GW1–8 | Starts |
|---|---|---|---|---|---|---|---|
| GKP | Verbruggen | BHA | £4.5m | 18.8% | 3.9 | 31.1 | 8/8 |
| GKP | Woodman | LIV | £4.0m | 1.1% | 1.8 | 14.7 | 0/8 |
| DEF | Gabriel | ARS | £8.0m | 28.7% | 6.7 | 46.3 | 8/8 |
| DEF | Guéhi | MCI | £6.0m | 20.4% | 4.8 | 38.3 | 7/8 |
| DEF | Thiaw | NEW | £5.0m | 1.9% | 4.5 | 37.9 | 8/8 |
| DEF | Keane | EVE | £5.0m | 1.7% | 5.0 | 34.9 | 5/8 |
| DEF | Bassey | FUL | £4.5m | 0.7% | 3.7 | 31.5 | 5/8 |
| MID | B.Fernandes (C) | MUN | £12.0m | 48.8% | 8.3 | 62.1 | 8/8 |
| MID | Bruno G. | ARS | £7.0m | 9.1% | 6.7 | 44.6 | 8/8 |
| MID | Enzo | CHE | £7.0m | 5.3% | 5.1 | 43.3 | 7/8 |
| MID | Anderson | MCI | £6.5m | 9.6% | 5.0 | 39.4 | 8/8 |
| MID | Tavernier | BOU | £6.0m | 1.8% | 3.9 | 38.8 | 8/8 |
| FWD | Haaland | MCI | £15.5m | 70.9% | 6.6 | 49.6 | 8/8 |
| FWD | Neave | NEW | £4.5m | 0.7% | 0.3 | 2.2 | 0/8 |
| FWD | Scarlett | TOT | £4.5m | 1.6% | 0.0 | 0.0 | 0/8 |
532.4 points — 6.2 behind Squad A over eight gameweeks, or 0.78 a week.
Look at what changes. Mbeumo, Thiago, Semenyo and Tarkowski all go — most of the £6m-plus midfield and attack outside Fernandes. In come Anderson (City, £6.5m, a real differential at 9.6%) and Tavernier, plus a cheaper defence and a £4.0m second keeper who never plays.
That's where the 6.2 points go. Not to Haaland's output. He is projected second-highest of any player in the game over the window, at 49.6.
Fernandes captains all eight weeks in Squad A bar GW4. In Squad B he takes five, Haaland takes GW3 and GW7, and Enzo GW4. Both squads captain Fernandes in Gameweek 1.
The price of the bet, exactly
Haaland's projected contribution to Squad B, captaincy included, is 65.1 points. Squad B trails by 6.21. So:
Haaland has to beat his projection by 9.5% for the Haaland squad to win. That's 8.13 points a gameweek projected, needing 8.91.
His cost also falls the further ahead you look — 1.13 points a gameweek at a one-week horizon, 0.78 at eight. His edge compounds; the dead-slot penalty you pay to afford him doesn't. If you're keeping him, plan long.
And why that number still isn't the argument
The field is already priced for him to hit his projection. Only the miss or the beat moves anyone's rank. A non-owner's edge is therefore:
−ownership × (actual − projected)
because the field carries that deviation 70.9% of the time and you carry none of it.
| If Haaland averages | vs projection | Non-owner edge per GW |
|---|---|---|
| A blank (2) | −6.1 | +4.3 |
| Below par (5) | −3.1 | +2.2 |
| His projection (8.1) | 0.0 | 0.0 |
| A good week (9) | +0.9 | −0.6 |
| A haul (15) | +6.9 | −4.9 |
| A monster (20) | +11.9 | −8.4 |
Roughly symmetric in points. Wildly asymmetric in rank. When he hauls you lose to 71% of the field at once. When he blanks you beat all of them at once. Same arithmetic, completely different distribution of outcomes — and rank is what the game scores.
One honest gap: we hold ownership but not captaincy share. His true effective ownership is well above 70.9%, so every figure in that table is a lower bound on the swing.
So which squad is for you
Squad A if you're playing for a rank you don't currently have — a mini-league you're behind in, or an overall target that needs you to beat most of the field rather than track it. It costs 0.78 points a gameweek and buys you the entire upside of a Haaland blank. It's also the less template squad by our own measure: 19.5% mean ownership on the players carrying its points, against 21.0% for Squad B.
Squad B if you're defending a position, or you'd genuinely rather be wrong alongside everyone than wrong alone. That isn't a soft reason. The discipline of not chasing is worth points to most managers, and 6.2 points over eight weeks is well inside this model's own error bar.
What is not a good reason to own him: that 70.9% of managers do. Their ownership is exactly what makes fading him profitable when he fails.
The one finding that survives everything
A player is strictly dominated when someone in his position is both cheaper and higher projected. No budget assumption, no horizon assumption, no judgement.
| Owned | Player | Price | GW1–5 | Beaten by |
|---|---|---|---|---|
| 59.6% | João Pedro (CHE) | £7.5m | 17.0 | Calvert-Lewin (LEE) £6.0m, 23.1 — and 5 others |
| 41.2% | Szoboszlai (LIV) | £7.0m | 24.6 | Anderson (MCI) £6.5m, 24.7 |
| 33.9% | Raya (ARS) | £6.0m | 17.8 | Kelleher (BRE) £5.0m, 20.2 — and 10 others |
| 30.2% | Mbeumo (MUN) | £8.0m | 27.2 | Bruno G. (ARS) £7.0m, 27.7 |
| 26.7% | Rogers (CHE) | £7.5m | 16.9 | Bruno G. (ARS) £7.0m, 27.7 — and 27 others |
| 26.6% | Semenyo (MCI) | £8.5m | 25.7 | Bruno G. (ARS) £7.0m, 27.7 — and 3 others |
| 26.0% | Calafiori (ARS) | £5.5m | 15.4 | Thiaw (NEW) £5.0m, 23.6 — and 23 others |
| 23.1% | Pedro Porro (TOT) | £5.5m | 17.5 | Thiaw (NEW) £5.0m, 23.6 — and 11 others |
| 21.6% | O'Reilly (MCI) | £6.5m | 20.5 | Guéhi (MCI) £6.0m, 24.3 — and 8 others |
| 20.1% | Rice (ARS) | £7.5m | 22.0 | Bruno G. (ARS) £7.0m, 27.7 — and 8 others |
João Pedro is in 59.6% of squads and is beaten by six cheaper forwards.
Read this as pairwise, not as a squad rule — it's the easiest thing on the page to misread. Mbeumo appears here beaten by Bruno G. and is still in our core seven. Semenyo appears here and is still in Squad A. Both for the same reason: the squad already owns Bruno G., and you can't own him twice. The list says "prefer the dominator if you don't already have him". Nothing more.
What this model can and can't do for you
| Strategy | Why | |
|---|---|---|
| ✅ | Value and squad structure | Literally the objective function |
| ✅ | Differential hunting | Projection and ownership are both in the data |
| ✅ | Fixture-swing timing | The team layer prices every fixture |
| ⚠️ | Template following | We can measure how template a squad is; copying the crowd isn't a model output |
| ⚠️ | Stars and scrubs | We can price it exactly, but not say whether the variance suits you |
| ❌ | Rank-state risk taking | Needs your rank and a distribution over the field |
| ❌ | Chip timing | Needs squad-level distributions, not per-player ones |
| ❌ | Price-rise trading | Driven by net transfers, which we don't model |
Where this is weakest
1. The promoted clubs are a guess, and it's load-bearing. Coventry, Hull and Ipswich have no top-flight record, so they inherit the average of the three clubs they replaced. That single assumption sets the difficulty of a large share of every opening fixture — and it's why Roefs is the first-choice keeper. Sunderland open away at Ipswich, and the model only "knows" Ipswich are weak because we told it.
2. Expected minutes are three months stale. They come from each player's last five gameweeks of last season — before pre-season, before new managers, before this summer's transfers.
3. 185 players are invisible. Anyone with no Premier League record is excluded rather than handed an invented number: every Coventry, Hull and Ipswich player, plus this summer's overseas signings. This model structurally cannot recommend one of them. A further 63 are flagged as unavailable and excluded on that basis.
4. The goalkeeper choice is arbitrary. The top keepers span 20.2 to 19.4 points over five gameweeks — under 0.2 a week. Roefs over Kelleher is a rounding error dressed as a decision.
5. The squad is not stable. Change the horizon and 4–7 of the 15 change with it. Only the core seven survive all four windows.
We also ran a transfer plan and are not publishing it, because it proposed moves worth 0.2–0.7 points against a model whose measured error is 2.4 points per player per gameweek in backtest, and one week it bought back a player it had sold two weeks earlier. Roll your transfer unless someone is injured.
The code is model/preseason_squad.py, model/preseason_optimise.py and model/preseason_strategy.py, with the caveats in the headers. When eight gameweeks are played the real walk-forward model takes over — and we'll come back and score both squads against what actually happened.
Next: this piece is about your fifteen. For the Gameweek 1 XI, the captain call and the fixtures themselves, read Gameweek 1 model picks — the first of a weekly series that runs to May.
You can also check any name above against his own page — Haaland, Bruno Fernandes, Bruno Guimarães, Igor Thiago, Gabriel — or search the full database on Players. Build the squad yourself in the Team Planner, and see who the field is avoiding on Differentials.