FPL Gameweek 7 Review: What Our Picks Got Right (and Wrong)

In Gameweek 7 our captain pick was projected at 6.62 and scored 6 — a fair call, but Neto's 15 was the haul in hindsight. Across 668 players our average error was 1.247 points, with a 0.555 correlation between projected and actual. Haaland (projected 5.75, scored 13) was the standout hit; Cherki's zero the clearest miss.

The headline numbers

Gameweek 7 is in the books, and here is the scorecard without the spin. We projected 668 players. The average error came in at 1.247 points per player, the root mean squared error was 2.045, and the correlation between what we projected and what players actually scored was 0.555.

That middle figure — RMSE of 2.045 — is the one worth dwelling on, because it is deliberately harsh on big misses. Average error treats a one-point slip the same way whether it happened to a defender or a premium forward; RMSE squares the gap first, so the Cherki-shaped holes in a week hurt it far more than a hundred near-misses on rotation-risk midfielders. The gap between 1.247 and 2.045 is, in plain terms, the signature of a week where most players landed roughly where expected and a handful did something wild.

A correlation of 0.555 is a reasonable, unspectacular number. It means the ordering was broadly right — the players we liked tended to outscore the players we did not — without being anything like precise. Nobody projecting football a week out should be claiming better, and if they are, ask to see the receipts.

The weakest number on the sheet: our top 10 contained a top scorer 20% of the time. That is the stat to hold us to. It says that on any given week, the chance of our headline shortlist including one of the gameweek's genuine hauls is modest. Useful for building a squad that does not bleed points. Not a lottery ticket.

The captain call

We projected 6.62 for our captain pick. It returned 6. On the narrow question of whether the projection was accurate, that is about as close as this gets — the player did almost exactly what the numbers said he would.

On the question of whether it was the right armband, no. Neto scored 15, and doubled that is a different gameweek entirely. This is the honest tension in captaincy advice: projections optimise for the most likely outcome, and the most likely outcome for a premium attacker is a quiet six-pointer. The captain haul almost always comes from a player whose median week is unremarkable but whose ceiling is enormous. We were right about the player and wrong about the week.

If you want our view on how to use that: the projection tells you who is unlikely to blank. It does not tell you who is about to score 15. Those are different questions and we would rather say so than pretend otherwise.

What went right

Three picks did more than hold up.

Haaland was projected 5.75 and scored 13 — 90 minutes, two goals, three bonus points. Nothing complicated here; the minutes were never in doubt, the volume of chances was always going to be there, and when a player of that profile converts twice the bonus follows automatically.

Evanilson was projected 5.24 and scored 12 off 78 minutes, with a goal, an assist and three bonus. That is the kind of return that makes a mid-priced forward worth the squad slot — the goal involvement came in a package, not in isolation, and he was still on the pitch deep into the second half despite not finishing the game.

Tavernier was projected 4.69 and scored 10 — a goal and two bonus from 78 minutes. A sub-5 projection delivering double figures is not us being clever; it is a reminder that the spread around any single projection is wide in both directions. We flagged him as a sensible option, not a ceiling play, and the ceiling turned up anyway.

What went wrong

This is the part that matters more.

Cherki was projected 5.33 and scored 0, because he played 0 minutes. That is not a modelling failure about football — it is a failure about availability. A projection that assumes minutes and gets none is worthless, and no amount of underlying numbers rescues it. If you brought him in on the back of our write-up, that is on us, and the lesson is the oldest one in FPL: check the team news before the deadline, every time, even when the projection looks safe.

N.Jackson was projected 5.85 and scored 2 from a full 90 minutes — no goals, no assists, no bonus. This one is a cleaner kind of wrong. The minutes were there, the role was there, the ball did not go in. Over a season those weeks average out; in a single gameweek they cost you. Our highest-projected of the three misses produced the second-lowest return, which is exactly the sort of event that pushes RMSE up to 2.045.

Guéhi was projected 5.22 and scored 1 across 90 minutes. Defender projections lean heavily on the chance of a clean sheet, and when that does not arrive there is very little floor left underneath. No attacking returns, no bonus, one appearance point. Nothing unusual happened — the projection simply backed an outcome that did not occur.

---

The gameweek 7 scorecard

So should you trust next week's picks?

Trust them for what they are. An average error of 1.247 points per player across 668 players means the typical projection lands close enough to be useful for ranking transfer options and avoiding obvious traps. A correlation of 0.555 means the shape of the week was right more often than not.

But a 20% hit rate on top scorers in our top 10 is the number that should set your expectations for captaincy and for chasing hauls. Use the projections to build a squad that does not blank. Use your own read on fixtures, team news and nerve for the armband — and if a player's minutes are in any doubt at all, let the Cherki zero be the reminder that no projection survives a benching.