FPL Gameweek 9 Review: What Our Picks Got Right (and Wrong)

Gameweek 9 was a middling week for us. Across 670 players scored, our average error was 1.071 points per player with a correlation of 0.598 between projected and actual. The damage was at the top: we captained Haaland on a projection of 8.08 and he returned 2, while Van Hecke scored 15.

The short version

We scored 670 players in Gameweek 9. The average error was 1.071 points per player, the root mean squared error was 1.845, and the correlation between what we projected and what actually happened was 0.598. Our top 10 contained a top scorer 30% of the time.

That is a week where the broad shape of the projections held up and the headline call went badly wrong. Those two things can both be true, and it is worth separating them, because the second one is the bit that actually cost managers points.

The captain call

We projected 8.08 for Haaland and put the armband on him. He played 90 minutes, did not score, did not assist, took no bonus, and finished on 2.

There is no clever framing for this. A blank from a captain is the single most expensive event in a Fantasy Premier League week, and it does not matter that the underlying case was sound — minutes were never in doubt, he was on the pitch for the full 90, and the projection was built on a player who gets chances regardless of opponent. He simply did not convert, and neither did anyone around him.

The best option in hindsight was Van Hecke, who scored 15. We had him projected at 4.8. So let us be completely straight about this: we did not see that coming either. A centre-half scoring 15 is not something any honest projection flags in advance, and anyone telling you they had Van Hecke captained in Gameweek 9 is telling you a story. Listing him under "where we were right" is generous to ourselves — we had him as a reasonable pick, not a hauler.

What we got right about him was the floor. He played 90 minutes, scored a goal and picked up 3 bonus. The minutes call was correct and the goal threat was in the projection; the magnitude was not.

What held up

Two other calls landed in a way that is more instructive than the Van Hecke lottery win.

Rogers was projected 4.92 and scored 9 — 90 minutes, a goal, 1 bonus point. That is the kind of hit that comes from getting the minutes right on an attacking midfielder and then letting variance do the rest in your favour. We were not predicting the goal specifically; we were predicting that he would be on the pitch long enough for one to be plausible.

Evanilson was the cleanest of the lot. Projected 6.18, scored 9, from 75 minutes with a goal and 3 bonus. A forward projected in that range who delivers 9 is almost exactly what a working projection looks like in practice — you are not aiming to nail the number, you are aiming to put him in the right band and let the distribution play out.

What went wrong

Beyond the captain, two misses are worth owning.

Mukiele was projected 4.33 and scored 0, because he played 0 minutes. That is the worst kind of error we can make — not a judgement call about attacking output, but a straightforward failure on availability. A projection of 4.33 on a player who does not get on the pitch is noise fed into your squad as if it were signal. There is no variance excuse here. If a player does not start, the number should not have been there.

Egan was projected 4.9 and scored 1 from 45 minutes. Half a game, no goal, no assist, no bonus. A 45-minute appearance for a defender is the worst of all worlds in FPL — too long to count as a rest, too short to get a clean sheet. This one sits somewhere between a rotation misread and ordinary bad luck, but the practical effect is the same: a mid-range projection on a player who was always going to be vulnerable to being hooked.

The common thread between Mukiele and Egan is minutes, not finishing. Haaland played his 90 and blanked, which is simply football. The other two failed before the football started.

---

Gameweek 9 at a glance

What the error figures actually mean for you

An average error of 1.071 points per player across 670 players sounds small, and in a sense it is — most players score close to what you would expect, because most players score very little. The more honest number is the root mean squared error of 1.845, which punishes the big misses rather than letting them average out against hundreds of correctly-projected 2-pointers. The gap between those two figures is the story of every FPL week: a sea of players doing roughly what you expected, and a handful doing something wild.

The correlation of 0.598 is the figure to hold onto if you are deciding whether to trust next week's picks. It says the ranking is useful and clearly better than guessing, but it is not a ranking you should follow blindly to three decimal places. A player we project at 6.18 is genuinely more likely to return than one we project at 4.33 — Evanilson and Mukiele happened to prove that neatly — but the ordering within half a point of projected points is close to a coin flip.

And that 30% figure, how often our top 10 contained a top scorer, is the one to keep in mind at captaincy time. Seven weeks in ten, the biggest haul of the gameweek was not in our top 10 at all. Gameweek 9 was one of those weeks. That is not an argument for ignoring the projections — it is an argument for captaining the player with the highest floor and the most secure minutes, rather than chasing the ceiling, because the ceiling is largely unpredictable for everybody.

Next week we will be more cautious on anyone whose start is in doubt. The Mukiele 0 is the error we should not repeat.