Gameweek 6's Biggest Fantasy Surprises — and Disappointments
A Jordan Henderson brace from a midfielder who hadn't played until last week, 12 points from 34 minutes of Jérémy Doku, and a bottom five in which every player had already had a double-digit week earlier in the season.
Same method as Gameweeks 1–5 — every player below played at least 30 minutes and is ranked by actual points minus PitchMetric's predicted xP for that fixture, using the predictions our daily snapshot captured before the gameweek (taken on Saturday 10 October, minutes before the first kickoff, so no match is contaminated by its own result). Gameweek 6 was the first after the international break, and like Gameweek 5 this post went up late, written after the fact. Predictions ran about half a point too high across the pool on average (−0.58), which is worth keeping in mind for the bottom five below.
Ranked among 233 players who played at least 30 minutes in gameweek 6, by actual points minus predicted xP. Average surprise across that pool: -0.58 (std. dev. 3.19).
Box score: 2 goals, 1 assist, 3 bonus · xG 0.14, xA 0.05 · ICT 14.1
Two goals, an assist and 3 bonus points in Chelsea's 5-1 home win over Bournemouth — from just 0.14 xG, so this is finishing rather than chance quality. He hadn't played a minute in the first four gameweeks and was making only his second start of the season, so a 4.45 prediction was a reasonable read of very little evidence; an 18-point week from a £4.9m midfielder is simply a long way out in the tail.
Box score: 1 goal, 1 assist, 3 bonus · xG 0.05, xA 0.09 · ICT 11.3
A goal and an assist inside 34 minutes at Liverpool in a 3-0 City win — 12 points from half a match. He'd started none of City's games and played 49 minutes all season (15 in GW5, 34 here), so the 1.66 prediction was mostly a start-probability call, and both returns came from very low-probability chances (0.05 xG, 0.09 xA). City's Cherki scored in the same game and just missed this list (+7.24, #6).
Box score: 1 goal, 1 assist, 3 bonus · xG 0.33, xA 0.00 · ICT 7.3
A goal, an assist and 3 bonus points in Coventry's 4-1 home win over Newcastle on the Monday night — a second Coventry player in our top five in two gameweeks, after Dasilva. He'd played just 30 minutes all season before this and never started, so the model was right to expect little from him — and wrong about how little. Newcastle's Lewis Hall is at the other end of this match, below.
Photo: Bryan Berlin, CC BY-SA 4.0, via Wikimedia Commons
.jpg){kind=link}
Box score: 1 goal, clean sheet, 3 bonus · xG 0.25, xA 0.00 · ICT 5.5
A goal, a clean sheet and 3 bonus points in Brighton's 2-0 win at Sunderland — his third appearance across these lists and the most whiplash of any player: our #1 in GW1 (+15.5), a −9.9 bust in GW2, and back in the top five here. His team-mate Lewis Dunk is at the other end of this same match, below.
Box score: 1 goal, 1 assist, 3 bonus · xG 0.40, xA 0.06 · ICT 15.2
A goal, an assist and 3 bonus points in Everton's 2-2 draw at Hull. After an 11-point opening day he'd scored 2, 2, 2 and 3 in the four gameweeks since, so the model had already marked that first haul down to 3.84; a 13-point week was still well beyond it.
Every player in this week's bottom five had a double-digit gameweek earlier in the season — Bruno Fernandes 23, Davis 14, Kayode 13, Hall 13, Dunk 12 — on between one and three goals each, and every one of them was then predicted at somewhere between 6.6 and 9.1 points. Fernandes and Gibbs-White showed the same pattern in Gameweek 5. On the positive side, the pattern runs the other way: Henderson, Doku and Cherif each had very few minutes before this, which is exactly where a model that knows little predicts little.
Gameweek 6's biggest disappointments
Every gameweek also has an equal and opposite story: players our model rated highly who delivered far less. Same method, same 30-minute cutoff, ranked the other way — the five biggest gaps between what we expected and what actually happened, on the low side.
Box score: — · xG 0.00, xA 0.08 · ICT 4.1
A yellow card and two goals conceded in a 2-2 draw at Aston Villa: 2 points for playing, minus 1 for the booking, minus 1 for conceding twice, zero in total. The number to look at is the 8.40 prediction for a £4.6 defender away at Villa — it traces back to a 13-point opening day that he hasn't come close to since (2, 1, 1, 6).
Box score: — · xG 0.00, xA 0.00 · ICT 2.2
Brighton kept a clean sheet in a 2-0 win and his team-mate De Cuyper took 15 points from it, but Dunk played only 30 minutes — short of the 60 FPL requires for clean-sheet points — and got 1. Why his afternoon was so short isn't in our data (Struijk was sent off after 56 minutes in the same match). What is: a 9.07 prediction, the highest in this bottom five, that assumed he'd play the full 90.
Box score: — · xG 0.02, xA 0.04 · ICT 4.6
Four goals conceded in a 4-1 defeat at Coventry — the other side of Cherif's #3 above. Hall was our #7 surprise the previous gameweek (+8.7: 13 points, with a goal), and the 6.58 prediction came straight after it.
Box score: — · xG 0.05, xA 0.01 · ICT 1.9
A yellow card and a goal conceded in a 2-1 home win over Fulham: no clean sheet and no attacking return. Our #4 surprise in GW4 (14 points, a goal and two assists), and the same hot-week-then-big-prediction shape as the others in this list.
Photo: Bryan Berlin, CC BY-SA 4.0, via Wikimedia Commons
.jpg){kind=link}
Box score: — · xG 0.02, xA 0.14 · ICT 5.5
Second week running in this list: a yellow card and a goal conceded in a 1-1 home draw with Spurs, for 1 point, after 8.07 → 2 in GW5. Outside his 23-point GW2 hat-trick, he has scored 1 or 2 points in every gameweek this season — yet he's still being predicted at over 7 a game.
Where the model should improve
Five of the ten entries across the last two gameweeks share one cause: the model treats a single big gameweek as a new baseline. A defender who scores once in his first five games has an observed rate of a goal every five matches, and our rate estimate blends that with what players at his price normally do — but the price-based prior counts for only about four matches of evidence, so one lucky goal still lifts a £4.0–£4.6 defender to a 7–9 point prediction, week after week, until enough ordinary games dilute it.
This was already on our list as a known weakness after the Semi Ajayi case in GW4; this gameweek is more evidence for it, not a fix. Raising how hard we shrink those rates would move every player's prediction on the site, so it needs a proper backtest across the whole pool first rather than a quick patch.