Accuracy record · 2026-27 · settled

Gameweek 4

In Gameweek 4 the model missed by 2.64 points a player across the 205 likely starters it scored, and the real team scored 83.

This page is published for every settled gameweek, good or bad, and is never edited after the fact. What follows is what the model projected before the deadline, what actually happened, how it did against the baselines on the same players, and what the real team scored.

Where Gameweek 4 went wrong

A worse week than usual: the model missed by 2.64 points a player, against 2.48 across the season so far.
Last season's points per game was closer than the model on points per player: 2.71 to 2.71 on the 162 players both scored.
Top-10k effective ownership was better than the model at spotting 10+ hauls this week: on the 205 players both scored, it ranked a hauler above a non-hauler 66% of the time, the model 58%.
FPL's own projection (ep_next) was better than the model at spotting 10+ hauls this week: on the 205 players both scored, it ranked a hauler above a non-hauler 74% of the time, the model 58%.
Spotting 10+ hauls was no better than a coin toss this week: 58%, with an interval (41%–76%) that includes 50%.
The team finished 7.6 behind a sample of top-10k managers (average 90.6, 257 sampled). This early, the sample is chosen on the very weeks it is scored on, so the gap is overstated.

What went right

The model was closer than last season's points per game on the big misses: 3.45 to 3.51 on the 162 players both scored.
The model ordered the players better than last season's points per game: 0.30 to 0.24 on the 162 players both scored.
The model was better than last season's points per game at spotting 6+ hauls: 67% to 62% on the 162 players both scored.
The model was better than last season's points per game at spotting 10+ hauls: 62% to 56% on the 162 players both scored.
The model ordered the players better than top-10k effective ownership: 0.31 to 0.18 on the 205 players both scored.
The model was better than top-10k effective ownership at spotting 6+ hauls: 70% to 61% on the 205 players both scored.
The model ordered the players better than fpl's own projection (ep_next): 0.31 to 0.26 on the 205 players both scored.
The model was better than fpl's own projection (ep_next) at spotting 6+ hauls: 70% to 60% on the 205 players both scored.
The team scored 83, 16 above the average manager's 67.

Not available this week

Only 14 players scored 10+ this week — too few to claim a result on the 10+ figure.

League-wide accuracy

Scored over predicted E[minutes] > 60 (n=205 of 654 player-fixtures). Every figure below is this gameweek only; the season-to-date line beside each one is the pooled figure across all settled gameweeks.

MAE

2.64

season to date 2.48 · mean absolute error, points per player-fixture

RMSE

3.44

season to date 3.23 · punishes big misses harder

Rank corr.

0.31 (0.18–0.43)

Spearman: 1.00 orders the players exactly as the week did, 0 is no relationship. The bracket is a 95% row-bootstrap interval — a floor on the uncertainty.

Rows

205

player-fixtures scored in this gameweek's population

Σ projected

686

points the model projected over that population

Σ realised

715

points those players actually scored — the model was −29 on the total

P(≥6) AUC

0.70 (0.62–0.78)

put a player who went on to haul above one who didn't 70% of the time; 0.50 is a coin flip. 56 hauls of 6+

P(≥10) AUC

0.58 (0.41–0.76)

put a player who went on to haul above one who didn't 58% of the time; 14 hauls — too few to claim a result

Prediction vintage: 12 Sept, 11:50 UTC run — Predictions are the last model run before each gameweek's deadline (first kickoff minus 90 minutes). 'seed' marks gameweeks that predate run retention, scored against the committed pre-season bundle.

95% percentile intervals from a row-level bootstrap (1000 resamples, seeded). Player-fixtures within a gameweek share fixtures and clean sheets, so these intervals understate the true width — read them as a floor on the uncertainty, not the uncertainty. A haul AUC on fewer than 50 hauls is too few to claim.

The same error on every stated population

SPEC §1.6 lets the unconditional figure appear only beside the conditional one. All four are printed so a reader who prefers a different population finds ours already stated — and so a comparison with someone else's figure is made on the same rows or not at all.

Population	n	MAE	RMSE	Season MAE
Every player-fixture with a prediction	654	1.18	2.13	1.23
Predicted E[minutes] > 0	549	1.41	2.33	1.41
Played at least one minute (selected on the outcome)hindsight	307	2.13	3.00	2.04
Predicted E[minutes] > 60headline	205	2.64	3.44	2.48

Reliability of the stated probabilities

In each bin of predicted probability, what fraction actually happened, with a Wilson 95% interval. A calibrated forecaster's observed column tracks its predicted column. The minutes model's P(60+) is read on every gameweek, the haul probabilities on the gameweeks whose run retained them.

P(plays 60+ minutes)

played 60+ minutes · every projected player-fixture with a retained p_start60 (n=654 of 654)

Predicted	Observed	95%	n
0%	0%	0–3	131
0%	0%	0–6	65
1%	3%	1–10	66
4%	2%	0–8	65
14%	9%	4–19	65
42%	39%	28–51	66
72%	66%	54–76	65
88%	88%	78–94	65
94%	94%	85–98	66

P(scores 6+)

scored 6+ points · every projected player-fixture with a retained p_ge_6 (n=654 of 654)

Predicted	Observed	95%	n
0%	0%	0–3	134
0%	0%	0–6	63
1%	0%	0–6	66
2%	5%	2–13	64
5%	2%	0–8	65
9%	9%	4–18	66
15%	14%	7–24	65
21%	23%	15–35	65
33%	47%	35–59	66

P(scores 10+)

scored 10+ points · every projected player-fixture with a retained p_ge_10 (n=654 of 654)

Predicted	Observed	95%	n
0%	0%	0–3	147
0%	0%	0–7	49
0%	0%	0–5	69
0%	0%	0–6	62
1%	2%	0–8	66
2%	3%	1–11	65
4%	2%	0–8	66
5%	6%	2–15	64
10%	9%	4–18	66

Against the comparators

Our error, next to something else's, on the same rows. An arm that cannot cover every row the model scored is compared only over the rows it does cover — with the model re-scored on exactly those rows, never against its headline figure for the week.

Comparator arm

Last season's points per game

Each player's total FPL points last season divided by his appearances — the yardstick a human reaches for before a ball is kicked.

Different rows, stated plainly. This arm scored 162 of the model's 205 rows for this gameweek; 43 of the model’s rows are excluded because the arm has nothing to say about them. Every “model” number in this table is the model re-scored on exactly the 162 rows the arm covers — so its headline error for the gameweek (2.64, over 205 rows) is not the figure to read against this arm.

Population: reference arm's predicted E[minutes] > 60, imposed on every arm (n=162 of 409 player-fixtures); players with a prior-season appearance only

Last season's points per game against the model on the 162 rows both cover, Gameweek 4
Metric	Model · these 162 rows	Last season's points per game
MAE arm ahead	2.71	2.71
RMSE model ahead	3.45	3.51
Spearman rank correlation model ahead	0.30	0.24
Haul AUC, 6+ points model ahead	0.67	0.62
Haul AUC, 10+ points model ahead	0.62	0.56

“—” means the payload carries no figure for that metric on this arm; it is not a zero. “No direction stated” means the payload does not say which way is better for that metric, so this page does not call a winner on it.

Comparator arm

Top-10k effective ownership

How heavily the sampled top-10k managers held each player at the last capture before the deadline, captaincy counted — the crowd's ranking of the field. A rank, not a points forecast.

Rows covered. This arm scored 205 of the model's 205 rows for this gameweek, with none of the model’s rows excluded. Every “model” number in this table is the model re-scored on exactly the 205 rows the arm covers, which is the only like-for-like comparison.

Population: reference arm's predicted E[minutes] > 60, imposed on every arm (n=205 of 654 player-fixtures); every model row; double-gameweek rows excluded (ownership is per gameweek, the record is per fixture)

Top-10k effective ownership against the model on the 205 rows both cover, Gameweek 4
Metric	Model · these 205 rows	Top-10k effective ownership
MAE not carried by this arm	—	—
RMSE not carried by this arm	—	—
Spearman rank correlation model ahead	0.31	0.18
Haul AUC, 6+ points model ahead	0.70	0.61
Haul AUC, 10+ points arm ahead	0.58	0.66

Comparator arm

FPL's own projection (ep_next)

The expected-points figure FPL publishes for each player before the gameweek, captured between the previous deadline and this one. Scored as a rank: it is per gameweek, so it says nothing about a double's second fixture.

Population: reference arm's predicted E[minutes] > 60, imposed on every arm (n=205 of 654 player-fixtures); players with an FPL projection captured in the pre-deadline window; double-gameweek rows excluded

FPL's own projection (ep_next) against the model on the 205 rows both cover, Gameweek 4
Metric	Model · these 205 rows	FPL's own projection (ep_next)
MAE not carried by this arm	—	—
RMSE not carried by this arm	—	—
Spearman rank correlation model ahead	0.31	0.26
Haul AUC, 6+ points model ahead	0.70	0.60
Haul AUC, 10+ points arm ahead	0.58	0.74

How the arms are scored Lower is better for MAE and RMSE; higher is better for Spearman and haul AUC (every metric states its own direction in `metrics`). Each baseline is scored on the model's own likely-starter rows (predicted E[minutes] > 60) and reports the model's score over exactly those rows, so the two numbers are like for like even when the baseline cannot cover every row. Overall Spearman and AUC are means over the gameweeks where BOTH arms have a value — never a cross-gameweek pool, which would rank a GW1 blank against a GW7 haul.

Legal-squad Team of the Week

Picked from the last model run before the deadline, at the prices captured before it, then scored on what actually happened.

Not comparable with the real entry. A fresh £100.0m squad, bought from scratch every gameweek: it pays no transfer hits, carries nothing over from last week and banks no price rises. It is a weekly mark on the model's picks under the real squad rules — not a strategy any manager can run, and not comparable with a managed entry on equal terms.

XI projected

57.2

with captain 64.2

XI realised

with captain 86

Capture

51%

of the 146-point hindsight XI — a ceiling with no budget and no club cap, not a squad

Spend

£100.0m

of £100.0m · 3-4-3

The legal-squad Team of the Week for Gameweek 4
Player	Club	Pos	Points
Verbruggen	Brighton	GK	8
Lacroix	Chelsea	DEF	1
Gabriel	Arsenal	DEF	9
De Cuyper	Brighton	DEF	11
Rogers	Chelsea	MID	8
Szoboszlai	Liverpool	MID	3
Wirtz	Liverpool	MID	3
Tavernier	Bournemouth	MID	8
João Pedro(C)	Chelsea	FWD	12
Isak	Liverpool	FWD	2
Haaland	Man City	FWD	9
Bench (does not score)
Dubravka	Spurs	GK	0
Thomas	Coventry City	DEF	0
Davis	Ipswich Town	DEF	14
Olusesi	Spurs	MID	0

Against the real entry

TOTW

XI with the captain doubled

Entry

before hits; none were taken this week

Difference

not a like-for-like contest

Read this as a mark, not a match. A fresh £100.0m squad, bought from scratch every gameweek: it pays no transfer hits, carries nothing over from last week and banks no price rises. It is a weekly mark on the model's picks under the real squad rules — not a strategy any manager can run, and not comparable with a managed entry on equal terms.

Deadline 12 Sept, 12:30 UTC · prediction 12 Sept, 11:50 UTC (run) · prices 18 Jul, 13:53 UTC–12 Sept, 11:50 UTC (185 from an earlier capture) · pool 654 of 654 predicted

The real entry · Sideline's Team

The squad the model actually ran — with transfer hits, bench points and continuity, all of which the Team of the Week above is free of.

Points

no hits taken

vs average

+16

field average 67 (FPL publishes it net of hits)

vs top 10k

−7.6

sampled average 90.6 from 257 managers, sampled as of the capture

Overall rank

1,277,769

gameweek rank 1,771,884

Projected

63.0

Σ multiplier × xP over 15 of 15 picks, from the last pre-deadline run (12 Sept, 11:50 UTC)

Bench

left on the bench

Transfers

0 paid

Squad value

£97.0m

£3.2m in the bank · captain João Pedro

The entry’s squad in Gameweek 4
Player	Club	Mins	Bonus	Points
Raya	Arsenal	90	3	14
Gabriel	Arsenal	90	2	9
Lacroix	Chelsea	4	0	1
Virgil	Liverpool	90	0	8
Calafiori	Arsenal	90	0	6
Tavernier	Bournemouth	82	2	8
Palmer(V)	Chelsea	90	0	5
Szoboszlai	Liverpool	90	0	3
Wirtz	Liverpool	90	0	3
Mbeumo	Man Utd	90	0	2
João Pedro(C)	Chelsea	90	3	12
Phillipsbench	Hull City	0	0	0
Guéhibench	Man City	90	0	6
Walle Egelibench	Ipswich Town	0	0	0
Kusi-Asarebench	Fulham	0	0	0

Transfers O'Reilly, B.Fernandes → Palmer, Lacroix

Cumulative after this gameweek: 289 · captured 15 Sept, 09:04 UTC · 4 squad captures

Where these numbers come from

Predictions are the last model run before each gameweek's deadline (first kickoff minus 90 minutes). 'seed' marks gameweeks that predate run retention, scored against the committed pre-season bundle.

Population scored: predicted E[minutes] > 60 (n=205 of 654 player-fixtures).

This page is a slice of the same /performance/live payload that drives the full record — one derivation, so a gameweek page and the season page cannot disagree. The projections themselves are published in full at /data/gw/4.

The exact rows this gameweek was scored on — the last run before the deadline, as served — are at fpl-quant-api.fly.dev/performance/live/gw/4/predictions with sha256 a0497656…c5bd. The digest is of the rows array alone as compact JSON (no whitespace), rows sorted by player then fixture — the response says so beside it. Recompute any number on this page from them.

What this record measures, on which rows, against which comparators and at which checkpoints was fixed in advance: the pre-registration (sha256 27307797…ba4f, checkpoints GW10, GW19, GW38).

Every settled gameweek has a page: GW1, GW2, GW3, GW4.