GW3 forecast report

The forecasts frozen at the GW3 deadline, scored against official FPL points. The forecast baseline stays fixed when the model changes later.

Final

Average miss1.80 ptsmean absolute error

Within 2 points72.5%of players who played

Bias−0.25scored less than forecast on average

Players scored459recorded minutes

Headline figures cover only players who played, since forecasting zero for a benched player is easy and would flatter the numbers. Across all 652 players the average miss was 1.31 pts.

By position

Position	Players	Avg miss	RMSE	Bias	Within 2
Goalkeepers	45	2.06	2.38	−0.95	44.4%
Defenders	162	2.06	2.74	+0.07	67.3%
Midfielders	200	1.60	2.23	−0.28	81.0%
Forwards	52	1.48	1.98	−0.54	80.8%

Calibration applied to the following Gameweek: GKP −0.15 · DEF −0.03 · MID −0.15 · FWD −0.12 pts.

Closest retained calls

Closest calls from the retained GW3 top-eight forecast sample; the complete player-level GW3 export was not retained.

Player	Pos	Forecast	Actual	Gap
Haaland	FWD	6.0	9	+3.0
Mbeumo	MID	5.4	8	+2.6
Guehi	DEF	4.5	8	+3.5
Szoboszlai	MID	4.6	3	-1.6

Largest misses

Player	Pos	Forecast	Actual	Miss
Mitchell	DEF	2.9	15	12.1
Bogle	DEF	2.5	14	11.5
Vuskovic	DEF	3.1	12	8.9
Barnes	MID	3.2	12	8.8
George	MID	1.8	10	8.2
Isak	FWD	4.8	13	8.2

What these measures mean

Average miss (MAE): the typical gap between forecast and actual points, ignoring direction.
RMSE: like the average miss, but big misses count for more.
Bias: actual minus forecast on average. Positive means players outscored their forecasts.
Within 2: share of players whose actual score was within 2 points of the forecast.

Snapshot and data source

Model version: 1.2.1-r1
Frozen: 4 Sept 2026, 17:02 UTC, before the deadline
Report generated: 7 Sept 2026, 05:12 UTC
Snapshot fingerprint: ec9658bb43d084fcaf736cd12979710bd7fd19874e2f7045fff08ad1d611a8ba

This report is historical evidence for GW3, not a test of the current model. See the live GW10 forecasts.