FPL Gameweek 4 Review: What Our Picks Got Right (and Wrong)
Our Gameweek 4 captain pick was projected at 7.42 and scored 9 — a small win, though Groß was the best armband in hindsight with 17. Across 659 players our average error was 0.915 points, correlation with actual scores was 0.77, and our top 10 contained a top scorer 50% of the time.
The headline numbers
We scored 659 players in Gameweek 4. The average error came in at 0.915 points per player, with a root mean squared error of 1.853 and a correlation between projected and actual points of 0.77. Our top 10 contained a top scorer 50% of the time.
That's a respectable week, but it's worth being clear about what those figures actually describe. An average error of 0.915 points does not mean we're within a point of every player. It means the typical miss is small — and the RMSE of 1.853 tells you the misses are not evenly spread. RMSE punishes big errors harder than small ones, so when it sits meaningfully above the average error, it's telling you a handful of players went badly wrong while most went roughly right. That is exactly what happened in GW4, and we'll come to the names.
The 0.77 correlation is the number we care about most. It says the ordering was largely sound: players we rated higher did, on the whole, score more. Football being football, that relationship will never be perfect, and anyone claiming otherwise is selling something.
The captain call
We projected 7.42 for our captain pick. It returned 9. Doubled, that's a solid if unspectacular week — the armband did its job without winning anyone a mini-league.
The pain is in the hindsight column. Groß scored 17. We had him projected at 5.71, which is to say we saw a starter with a plausible route to points and nothing more. He delivered 90 minutes, a goal, two assists and three bonus points — the full haul. Nobody was captaining a 5.71 projection, and we wouldn't have told you to. But it's a reminder that the single highest-scoring option in a gameweek is very often not the highest-projected one. Projections are about the average of all the ways a match could go. Groß played the version where everything landed.
Where we got it right
The wins in GW4 were less about pinpoint accuracy and more about being pointed in the right direction.
Groß is a strange one to file under "right" given we projected 5.71 for a 17-point return, but the underlying read — a nailed starter with creative involvement — held up. Bogle was the cleanest call: projected 6.18, scored 15, with 82 minutes, a goal and three bonus. That's a defender we rated above the crowd doing precisely the thing we rated him for. Schade was similar in shape: projected 5.31, delivered 15 off 90 minutes, two goals and three bonus points.
What links all three is minutes. Every one of them played 82 minutes or more. When a projection is built on a player being on the pitch and that assumption holds, the rest is a coin-flip we're happy to take repeatedly. When the minutes assumption breaks, nothing else matters — which brings us to the other half of the ledger.
Where we got it wrong
Gakpo is the clearest miss. We projected 5.32 and he returned 1 from 30 minutes. No goals, no assists, no bonus. This is a minutes failure rather than a performance failure: a projection of 5.32 implicitly assumed something close to a full shift, and 30 minutes makes that projection irrelevant before the game has really started. We flag rotation risk where we see it, and in this case we didn't see enough of it.
Cherki was the same story in miniature — projected 4.96, scored 1, 45 minutes. A half is a half. There is no version of a 45-minute cameo that delivers a mid-range projection unless the player scores, and we shouldn't have been that confident on the minutes.
B.Fernandes is the more uncomfortable one, because there's no excuse available. He played the full 90. We projected 6.15 and he returned 2 — no goal, no assist, no bonus. That is a pure attacking-return blank from a player who had every opportunity to deliver. Nothing in the process failed; the ball simply didn't go in. Those weeks happen to every premium midfielder and there is no adjustment that eliminates them. If we told you we'd fixed it for next week, you should stop reading us.
The GW4 scorecard
| Player | Projected | Actual | Minutes |
|---|---|---|---|
| Groß | 5.71 | 17 | 90 |
| Bogle | 6.18 | 15 | 82 |
| Schade | 5.31 | 15 | 90 |
| B.Fernandes | 6.15 | 2 | 90 |
| Gakpo | 5.32 | 1 | 30 |
| Cherki | 4.96 | 1 | 45 |
What this means for next week's picks
Here's the honest framing. A 50% hit rate on having a top scorer inside our top 10 is genuinely useful — it means half the time, the highest ceiling in the gameweek was sitting in a list you could have read in two minutes. It also means half the time it wasn't, and the week's biggest score came from somewhere we didn't highlight. Both halves of that sentence are true and you should hold them together.
The practical lesson from GW4 is about minutes, not about form. Two of our three worst misses — Gakpo at 30 minutes and Cherki at 45 — were players who never got the chance to justify the number next to their name. The three big wins all played 82 minutes or more. If you're choosing between two players with similar projections, the tiebreaker that actually moves the needle is which one you're more confident starts and finishes.
And on the B.Fernandes type of miss: 90 minutes, zero returns, a 6.15 projection turning into 2. That is not an error to correct, it's variance to absorb. The way to absorb it is to keep making decisions that are right on average rather than chasing the player who happened to score 17 last Saturday. An average error of 0.915 points across 659 players says the process is broadly sound. It does not, and never will, say the process is safe.
Questions
- How accurate were coachme's Gameweek 4 projections?
- Across 659 players scored, our average error was 0.915 points per player with a root mean squared error of 1.853. The correlation between projected and actual points was 0.77, meaning the ordering of players was broadly right even where individual scores missed. Our top 10 contained a top scorer 50% of the time.
- Who was the best FPL captain in Gameweek 4?
- In hindsight, Groß was the best captain option with 17 points — 90 minutes, a goal, two assists and three bonus. We had him projected at 5.71, so he was not a realistic armband call at the time. Our actual captain pick was projected at 7.42 and returned 9.
- Which Gameweek 4 picks did coachme get wrong?
- Three clear misses. Gakpo was projected at 5.32 and scored 1 from just 30 minutes, and Cherki was projected at 4.96 and scored 1 from 45 minutes — both minutes failures. B.Fernandes played the full 90, was projected at 6.15 and returned 2 with no goal, assist or bonus, which was simply a blank week rather than a process error.
- Should I trust FPL projections after a week like this?
- Use them for ordering rather than as a promise. A correlation of 0.77 between projected and actual points says higher-rated players did generally score more in Gameweek 4, but an RMSE of 1.853 against an average error of 0.915 shows a small number of players missed badly. Where two picks look close, favour the one with more secure minutes — that was the difference between our wins and our misses this week.