The squad, not the theory
Forget the general theory for a minute — here's the actual fifteen produced by our own Squad Builder for Gameweek 1, at the full £100.0m budget. This is the exact formation and squad, pulled straight from the optimizer, and it's fixed as the reference point for the rest of this piece: whatever the live tool suggests later in the season, this is the squad we're talking about here.
Formation
3-5-2| Player | Position | Club | Price | GW1 | GW2 | GW3 | GW4 | GW5 | GW6 | Total |
|---|---|---|---|---|---|---|---|---|---|---|
| Robert Sánchez | GKP | CHE | £5.0m | 2.9 | 3.4 | 3.4 | 3.7 | 2.8 | 3.4 | 19.4 |
| Bart Verbruggen (bench) | GKP | BHA | £4.5m | 2.8 | 2.7 | 3.0 | 2.8 | 3.3 | 2.6 | 17.3 |
| Nico O'Reilly | DEF | MCI | £6.5m | 6.0 | 5.6 | 6.4 | 5.0 | 6.2 | 4.8 | 34.0 |
| Marc Guéhi | DEF | MCI | £6.0m | 5.7 | 5.3 | 6.2 | 4.7 | 6.0 | 4.5 | 32.4 |
| Nordi Mukiele | DEF | SUN | £5.5m | 4.2 | 5.6 | 4.3 | 3.6 | 3.1 | 5.2 | 25.9 |
| Malick Thiaw (bench) | DEF | NEW | £5.0m | 3.8 | 3.8 | 5.3 | 4.3 | 6.5 | 4.1 | 27.8 |
| Issa Diop (bench) | DEF | IPS | £4.0m | 2.4 | 1.6 | 1.6 | 2.1 | 2.6 | 2.9 | 13.1 |
| Bruno Fernandes (C) | MID | MUN | £12.0m | 8.2 | 7.9 | 6.6 | 7.0 | 7.4 | 8.7 | 45.7 |
| Antoine Semenyo | MID | MCI | £8.5m | 7.6 | 6.7 | 7.5 | 6.7 | 7.2 | 6.8 | 42.4 |
| Bryan Mbeumo | MID | MUN | £8.0m | 7.2 | 6.8 | 5.3 | 6.0 | 6.5 | 7.3 | 39.1 |
| Florian Wirtz | MID | LIV | £7.5m | 6.4 | 7.1 | 6.5 | 7.3 | 6.5 | 6.6 | 40.3 |
| Cody Gakpo | MID | LIV | £7.0m | 5.5 | 6.8 | 5.9 | 7.0 | 6.0 | 6.2 | 37.3 |
| Ollie Watkins (VC) | FWD | AVL | £8.0m | 6.1 | 6.7 | 7.6 | 7.5 | 7.0 | 8.1 | 43.0 |
| Dominic Calvert-Lewin | FWD | LEE | £6.0m | 5.6 | 6.2 | 4.5 | 6.0 | 6.0 | 4.3 | 32.8 |
| Jean-Philippe Mateta (bench) | FWD | CRY | £6.5m | 5.1 | 5.1 | 5.4 | 6.0 | 5.5 | 6.2 | 33.2 |
£100.0m spent to the penny, no club supplying more than three players, and the model's own projected total for this exact fifteen sits at 448.07 points across Gameweeks 1–6 with captaincy applied. Four names carry it: Bruno Fernandes, Ollie Watkins, Antoine Semenyo and Florian Wirtz account for £36.0m of the budget and over a third of the projected output between them. The other eleven exist to protect the bank, cover rotation, and keep the bench playing when it actually has to.
That's the concrete example. Everything below explains the reasoning that produced it — specifically, why the squad was optimised around a six-gameweek window rather than one gameweek or eight.
Why five to six gameweeks is the right planning window
Every projection in that table has a shelf life. Player output isn't a fixed number waiting to be discovered — it's a stochastic process, a noisy signal wrapped around an underlying rate that shifts week to week. Form has a short memory: meaningful autocorrelation in a player's output rarely survives more than two or three gameweeks once you strip out the fixture underneath it. Fixture difficulty, by contrast, is known in advance and stays informative for considerably longer — you can trust a fixture run in gameweek six with far more confidence than you can trust a player's current hot streak.
That's the whole argument for a rolling horizon in one sentence: fixtures decay slowly, form decays fast, so the window you plan around needs to be long enough to capture fixture signal and short enough to avoid pretending you know things you don't.
Formally, the job is to maximise the sum of expected points across a horizon H, subject to the constraints that actually govern your squad — the £100.0m budget, no more than three players per club, exact position quotas, the free transfers you've banked, and the four-point cost of any extra transfer you're tempted into. Change H and you change the answer. A one-week horizon pushes you toward double-gameweek specialists who blank for a month either side. An eight-week horizon pushes you toward players picked for fixtures that haven't yet survived a transfer window, an injury, or a manager getting sacked.
The bit that makes this low-risk rather than a one-shot bet: you're never actually locked into the full horizon. The squad is re-optimised every single gameweek as new information lands — injuries, press conferences, price changes, actual minutes played. Each week's projection updates on the last, which is a fancy way of saying the model does what any decent manager does anyway: takes the new evidence, adjusts the estimate, and moves on. Because of that constant re-optimisation, planning six gameweeks ahead doesn't commit you to six gameweeks of anything. It just gives this week's transfer the right context.
And this isn't a hunch. Walk-forward validated squad-selection models — the kind tested only on data they've never seen, gameweek by gameweek, rather than fitted neatly to a season that's already happened — consistently land on a horizon of five to six gameweeks as the sweet spot. So do the better community optimisers that have quietly been tuned against real season results for years. When that many independent approaches converge on the same number, it stops being a coincidence.
Too short overreacts, too long overtrusts the model
Go too short and you overreact. A one-to-three gameweek horizon hands too much weight to a single fixture, which pushes you toward chasing the next blank or double rather than building a squad that scores well on average. It also forces your hand: you'll take avoidable hits chasing a short-term swing, or panic into a wildcard you hadn't planned for, because the model can't see far enough ahead to tell you the storm passes.
Go too long and you overtrust. An eight-plus gameweek horizon does drag in fixtures that are still genuinely useful, but it also drags in noise dressed up as signal — injury risk, transfer windows, a new signing nobody's minutes model has calibrated for yet, a manager who gets sacked in October. Locking a chunk of your budget against a forecast that far out quietly kills the thing that makes FPL forgiving in the first place: the option value of your free transfer and your first wildcard. Optionality is worth something. An eight-week horizon prices it at zero.
- Horizon too short (1–3 GWs): overweights one fixture, forces avoidable hits, risks an unplanned wildcard
- Horizon too long (8+ GWs): trusts low-confidence projections, ties up budget, discounts the value of flexibility
The case for 6–8, and why we don't go that far
You'll see good analysts argue for a six-to-eight gameweek window instead, and it isn't a lazy take — it's directionally reasonable, especially if you're planning to hold your first wildcard back rather than firing it in gameweek four or five. If you already know you're sitting tight until gameweek eight, there's a genuine case for planning your squad with that whole runway in mind, because you truly aren't going to touch it before then anyway.
But for a standard template squad with no wildcard pre-committed — which is what most of us are actually running in August — the centre of mass sits closer to six. Weekly transfers are still doing the fine-tuning, so you don't need the model betting on gameweek eight fixtures to get the benefit of planning for them; you'll simply re-run the horizon next week and gameweek eight will still be five or six weeks out, informative and current. Stretching to eight now mostly just adds forecast error without adding anything you couldn't get by re-optimising weekly instead.
How the horizon should move through the season
The horizon isn't a fixed number you set once in August and forget about. It should move with the season, and it should move for specific, identifiable reasons — not vibes.
- Default state: roll a 5–6 gameweek window every week — this is where the model lives for roughly 80% of the season
- Wildcard or Free Hit imminent: shrink the horizon down to just the gameweeks remaining before the chip, then let the fresh squad plan its own six from there
- Approaching a known double/blank gameweek cluster: stretch the horizon to cover the whole cluster, even if that means looking eight-plus weeks ahead for that one specific decision
- Injury crisis or heavy uncertainty: shorten slightly and lean harder on expected-minutes data — certainty of starting beats projected ceiling when half the squad is a fitness doubt
- Price volatility and bandwagon periods: keep multi-gameweek expected points as the base layer, then add short-term ownership and price-change considerations on top rather than letting them drive the squad alone
- Late season: shorten again if rotation risk is high — dead rubbers, cup finals, kids getting run-outs — and lengthen if the run-in is stable with every side still playing for something
Takeaways
The squad at the top of this piece isn't the point. The process that built it is.
- Plan on a rolling 5–6 gameweek horizon by default — long enough for fixtures to matter, short enough to trust the numbers
- Re-run that horizon every gameweek; you're never locked into more than one week of actual commitment
- Shrink the horizon before a wildcard or free hit, stretch it to cover a known double/blank cluster
- Prioritise expected minutes over ceiling whenever injury news is flying around
- Treat ownership and price-change data as a layer on top of multi-gameweek projections, never as the primary driver
- Judge any squad — including this one — on its six-gameweek total, not on how it looks after gameweek one alone
Fergie Time Tim
Guest Analyst
Written by the FPL King editorial network - tactical FPL analysis, captaincy edges, transfer strategy and fixture-led planning.
FPL manager chat
Discussion
Loading discussion…