Gameweek 2's Biggest Fantasy Surprises
A Bruno Fernandes hat-trick headlines the round, but the more interesting story is what changed in the model's own inputs one gameweek in.
Model v2 — all five predictions revised again: a same-day scoring-bug fix (defensive-contribution points were badly overcounted) plus new team-rating shrinkage and fixture-aware save/assist rates now apply to every gameweek with real current-season minutes behind it, GW2 onward. What changed, in the GW4 post.
Same method as Gameweek 1 — every player below played at least 30 minutes, and is ranked by actual points minus PitchMetric's predicted xP for that fixture. The difference this time is that the model is no longer working from a blank slate: every returning player now has one gameweek of real 2026-27 output behind their goal and assist rates, and it shows in who tops this list and why. See the methodology for exactly how a player's rate is computed from real per-match history rather than an assumption.
Ranked among 231 players who played at least 30 minutes in gameweek 2, by actual points minus predicted xP. Average surprise across that pool: +1.19 (std. dev. 3.52).
Photo: Bryan Berlin, CC BY-SA 4.0, via Wikimedia Commons
.jpg){kind=link}
Box score: 3 goals, 1 assist, 3 bonus · xG 2.02, xA 0.48 · ICT 26.1
The biggest single surprise across both gameweeks. Fernandes played the full 90 minutes in Gameweek 1 but didn't register a goal or an assist in it — so despite years as one of the Premier League's most productive playmakers, this season had given the model nothing to go on yet, and it priced him at a generic 2.14 points, barely above a squad player's floor. What actually happened was a hat-trick plus an assist in a 5-2 win over Ipswich Town — and 2.02 expected goals means the underlying chances very nearly justify the goal tally on their own. This is the good kind of model miss: not a fluke the numbers can't explain, just one quiet opening game that happened to come right before this one.
Photo: Sebalston, CC0, via Wikimedia Commons
Box score: 1 goal, 1 assist, 3 bonus · xG 0.79, xA 0.04 · ICT 8.6
Gibbs-White has been one of Nottingham Forest's most important attacking players for a couple of seasons now, and he played every minute of Gameweek 1 too — he just didn't score or set one up, so there was no current-season product for the model to price him on beyond a modest 1.72 points. Then came a goal and an assist away at Liverpool, in a match that finished 2-2, worth 13 points. The 0.79 expected goals behind his own finish is a reasonable, presentable chance rather than a wild outlier — the assist is the bigger surprise, with only 0.04 expected assists suggesting the pass that created it wasn't a high-probability chance-creation moment by the underlying numbers.
Photo: Bryan Berlin, CC BY-SA 4.0, via Wikimedia Commons
Box score: 2 goals, 3 bonus · xG 0.35, xA 0.55 · ICT 17.2
The model's read on this one moved after we fixed a real gap: it used to count goals and assists per start only, so Cherki's 2 assists in 27 minutes off the bench in Gameweek 1 — real, meaningful output — simply didn't count toward anything, because a substitute appearance wasn't a start. Weighing by minutes actually played instead (v2, above) values that cameo as real evidence, not nothing. He started this game, played 81 minutes away at Crystal Palace in a match City won 4-1, and scored twice himself. The 0.35 expected goals behind those two finishes says this was more clinical finishing than a string of clear-cut chances — genuinely hard to have called exactly, even now.
Photo: Bryan Berlin, CC BY-SA 4.0, via Wikimedia Commons
Box score: 2 goals, 3 bonus · xG 0.66, xA 0.00 · ICT 12.0
The most expensive player on either list, and still a double-digit surprise. Even the league's most prolific striker has games where nothing goes in — Haaland played all 90 minutes of Gameweek 1 without scoring, and with no goals on the board yet this season, the model had nothing personal to build on and started him from the same generic 2.17-point floor as everyone else. Two goals away at Crystal Palace, in the same 4-1 win as Cherki above, from just 0.66 combined expected goals is a big finishing overperformance on the underlying chances, not a case where the model missed obviously high-quality opportunities.
Photo: Bryan Berlin, CC BY-SA 4.0, via Wikimedia Commons
Box score: 1 goal, 1 assist, 3 bonus · xG 0.05, xA 0.11 · ICT 10.4
Groß only returned to Brighton in January, back from a spell at Borussia Dortmund, and has quickly become the club's main penalty and set-piece taker — exactly the kind of role that can produce a goal or an assist out of nothing. None of that reached the prediction, though: he'd played the full 90 in Gameweek 1 without a goal or assist, so the model had no personal scoring history to draw on, and its 2.60-point prediction (the highest of this week's five, but driven by Brighton's own fixture rather than anything specific to him) still wasn't close to enough. Brighton conceded four goals away at Chelsea and lost 4-3, but Groß personally returned a goal and an assist worth 13 points on his own. The 0.05 expected goals behind that goal is about as low as the underlying numbers get for an actual finish — the kind of moment a box score just can't explain, whatever it looked like on the pitch.
Compare the underlying expected-goals numbers across this list and two different stories emerge. Bruno Fernandes’s 2.02 xG very nearly explains his hat-trick outright — his performance was mostly signal, a genuine chance-quality outburst the model just had no prior data to anticipate. Haaland (0.66 xG for two goals) and Groß (0.05 xG for a goal) leaned much more on finishing than on chance quality — closer to variance than to a repeatable pattern. Rayan Cherki’s case is the odd one out for a different reason: it’s the one entry on this list whose prediction has genuinely moved since we first published it, once the model stopped discarding a substitute appearance’s output outright. None of the rest is a flaw exactly — it’s what a model honestly built from observed history looks like in only its second week of data.
Gameweek 2'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.
Photo: Bryan Berlin, CC BY-SA 4.0, via Wikimedia Commons
.jpg){kind=link}
Box score: — · xG 0.00, xA 0.00 · ICT 10.2
A week after topping our GW1 surprises list (17 points), De Cuyper's own price and recent form had the model expecting another big game — instead Brighton conceded four away at Chelsea and he came away with nothing. The same price-prior mechanism that correctly reads “player in form” can't see a back four collectively having a shocker coming.
Box score: — · xG 0.00, xA 0.00 · ICT 9.6
Same match, same story: four goals conceded away at Chelsea erased any defensive or attacking return. Two of Brighton's back line make our worst five this week — a single bad away day at a top-six side, not two independent player failures.
Box score: — · xG 0.00, xA 0.00 · ICT 8.8
A yellow card on top of the same four-goal defeat took him below zero. Three Brighton defenders in our bottom five from one match is as clean a demonstration as any that a team-level collapse doesn't average out — it stacks.
Box score: — · xG 0.00, xA 0.00 · ICT 5.4
Hull actually won this one 1-0 away — the frustrating part is Mendy kept a clean sheet on the pitch but was substituted at the 58th minute, two short of the 60 FPL requires to actually award the clean-sheet bonus. A real defensive shift that simply fell one minute short of counting.
Box score: — · xG 0.00, xA 0.00 · ICT 3.1
The same story again, further down Hull's bench: on for 31 minutes of a match his team won 1-0 away, but not enough of it to bank the clean-sheet points a start would likely have earned. Two Hull players in one week's worst five, both undone by the same 60-minute cutoff rather than anything they did wrong.