Every FPL planner tool will happily let you map out the next eight gameweeks. Almost none of them tell you how much of that plan is real.
We can, because we run the optimisation ourselves. Here is what happens when you change one setting — how far ahead you're planning — and change nothing else.
The experiment
Same £100m budget. Same squad rules. Same projections. The only thing that moves is the planning horizon: pick the best fifteen for the next 1, 3, 5 or 8 gameweeks.
If planning ahead were straightforward, the four squads would mostly agree. They don't.
| Horizon | Projected XI points | Per gameweek | Players changed vs previous horizon |
|---|---|---|---|
| 1 gameweek | 71.9 | 71.9 | — |
| 3 gameweeks | 207.8 | 69.3 | 7 of 15 |
| 5 gameweeks | 335.7 | 67.1 | 6 of 15 |
| 8 gameweeks | 538.6 | 67.3 | 4 of 15 |
Read the right-hand column again. Going from "best team for this week" to "best team for the next three weeks" replaces nearly half the squad. Not the bench — the squad.
The bit that is actually stable
Seven players are picked at every horizon:
| Pos | Player | Club | Price | Owned |
|---|---|---|---|---|
| MID | Bruno Fernandes | MUN | £12.0m | 48.8% |
| DEF | Gabriel | ARS | £8.0m | 28.7% |
| MID | Bruno Guimarães | ARS | £7.0m | 9.1% |
| FWD | Igor Thiago | BRE | £8.0m | 17.5% |
| MID | Bryan Mbeumo | MUN | £8.0m | 30.2% |
| DEF | Marc Guéhi | MCI | £6.0m | 20.4% |
| GKP | Robin Roefs | SUN | £5.0m | 4.8% |
£54.0m of the £100.0m. These seven are the plan. Whether you're thinking about Saturday or about October, the solver wants them.
The other eight slots are the solver reacting to whichever fixtures happen to fall inside the window you gave it. Change the window and it changes its mind. That isn't a bug — it's an accurate report of how much the evidence actually supports.
What that means for your planner
Plan the core. Fill the rest. If roughly half your squad flips when you nudge the horizon by two weeks, then half your squad was never a decision — it was an artefact of an assumption you made without noticing.
Multi-week transfer plans are worth less than they feel. We ran one. It proposed moves worth 0.2 to 0.7 points each. Our model's measured error is 2.4 points per player per gameweek. The moves were an order of magnitude smaller than the noise they were being chosen out of, and in one week the plan bought back a player it had sold two weeks earlier.
We didn't publish it, and that's the honest recommendation: roll your transfer unless someone is injured, suspended, or has genuinely lost his place.
The exception is the fixture swing. Planning pays where the effect is large enough to clear the noise, and there's one place it reliably is: cheap defenders on a fixture run. A defender scores 4.94 in the easiest fixture bracket and 1.89 in the hardest — see does fixture difficulty actually matter. That gap is big enough to plan around. A 0.4-point midfield upgrade is not.
Why the horizon changes the answer at all
Two reasons, and they pull in opposite directions.
Fixtures rotate. A player with three home games in the next three weeks looks fantastic at a three-week horizon and ordinary at eight. The solver is right to prefer him at H=3 and right to drop him at H=8.
Expensive players amortise. The longer the window, the more a premium's edge compounds against the fixed cost of the bench slot you sacrificed to afford him. Forcing Haaland into the squad costs 1.13 points a gameweek at a one-week horizon and 0.78 at eight. Same player, same price, and the case for him improves the further out you look purely because of arithmetic.
That second point is the practical one. If you're building around a premium, plan long. If you're playing week to week, don't buy one.
How to use a planner without fooling yourself
- Lock the core. Whoever your model, your gut or ours says is stable across horizons — those names don't move.
- Plan two weeks, not eight. Beyond that you're plotting against fixtures whose difficulty will have changed by the time you get there.
- Re-plan every week anyway. The plan is a way of noticing what's coming, not a commitment.
- Treat any move worth under a point as a non-move. It's inside the error bar.
- Never take a hit on a projection alone. −4 needs a swing you can see without a model.
The limits
This is one squad optimisation on one set of pre-season projections, run on 18 August 2026. The projections themselves haven't passed our evaluation gate — the model can't train properly until eight gameweeks are played, so these carry a season-boundary substitution described in full in the GW1 draft piece.
The instability finding is more robust than the specific names. Fewer than half the picks surviving a change of horizon is a statement about how little separates the candidates, and that would hold with a better model — it's a property of football, not of our code.
Build it out in the Team Planner, sort the pool on Players, or check the run of fixtures on the Fixtures grid. This week's picks are in Gameweek 1 model picks.