Gameweek 4's Biggest Fantasy Surprises — and Disappointments
A five-goal Brighton rout, a Palace collapse, and a Leeds win that put GW3's biggest miss back on top of the list — this time from the other side. All ten GW4 matches now final.
Model v2 — the first gameweek run entirely on the corrected model. What changed and why, below.
Update, GW4 complete: Leeds beat Newcastle 4-1 in the gameweek's last fixture, and it moved the list — Jayden Bogle's goal, clean sheet and 3 bonus points push him to the second-biggest surprise of the round, bumping Kevin Schade out of the top five (he's now sixth). Everything else below is unchanged from the original post. Also unchanged: this ran on a same-day model fix caught the day this post first went up — a scoring bug that had been quietly inflating some players' predictions by 10-15+ points a gameweek — so GW4 is the first gameweek scored entirely on the corrected numbers.
Ranked among 239 players who played at least 30 minutes in gameweek 4, by actual points minus predicted xP. Average surprise across that pool: +0.06 (std. dev. 3.53).
Photo: Bryan Berlin, CC BY-SA 4.0, via Wikimedia Commons
Box score: 1 goal, 2 assists, clean sheet, 3 bonus · xG 0.82, xA 0.19 · ICT not yet finalized by FPL
A goal, two assists and a clean sheet as Brighton put five past Coventry away — the model had him as a solid pick against a newly-promoted side, not a season-defining haul. Goal involvement in a rout tends to cluster on whoever's already central to the buildup, and Groß was that player three times over.
Photo: Timmy96, CC0, via Wikimedia Commons
.jpg){kind=link}
Box score: 1 goal, clean sheet, 3 bonus · xG 0.41, xA 0.05 · ICT not yet finalized by FPL
A goal, a clean sheet and 3 bonus points as Leeds beat Newcastle 4-1 at home — the same player, the same 'the model didn't see this Leeds defender's day coming' pattern as his own GW3 story, just flipped. That one was about Brighton's attack rating swinging to an extreme off two small-sample results, not really about Bogle at all; this one is closer to a proven first-choice right-back and an established Leeds side simply outplaying a moderate-on-paper fixture read. GW3's single biggest miss is GW4's second-biggest hit.
Box score: clean sheet, 3 bonus · xG 0.00, xA 0.00 · ICT not yet finalized by FPL
A clean sheet plus two saves away at newly-promoted Sunderland, with 3 bonus points on top — a start_prob discount from Arsenal's midweek Champions League fixture (new this gameweek, see the model notes below) had already shaved his prediction down slightly; the clean sheet itself was still well above what a mid-table away trip usually returns.
Box score: 1 goal, 2 assists · xG 0.52, xA 0.19 · ICT not yet finalized by FPL
A goal and two assists from left-back as newly-promoted Ipswich won 3-2 away at Crystal Palace — exactly the kind of result our price-prior model is built to under-rate for a £4.0 defender at a promoted club with no attacking track record yet this season.
Box score: 1 goal, 1 assist, 3 bonus · xG 0.29, xA 0.04 · ICT not yet finalized by FPL
A goal and an assist in a 2-2 home draw with Hull — a reminder that goal involvement points don't need a win behind them, and that our opponent-difficulty read on a newly-promoted side's away form is still thin this early in the season.
Four of this week's five surprises tie to a newly-promoted club one way or another — Groß's Brighton put five past Coventry, Raya faced Sunderland, Davis plays for Ipswich, João Pedro faced Hull — the same thin-early-season-data pattern GW1-4 have all shown. Bogle is the exception, and a fitting one: Newcastle are an established side, so this reads less like a data-thinness problem and more like a proven player and an in-form Leeds side just having a very good day. It's also a nice bit of symmetry — Bogle was GW3's single biggest miss, in a match where the model's fixture-level read, not Bogle himself, was wrong. A week later, in a different match entirely, he's back near the top for a different reason.
Gameweek 4'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: Ardfern, CC BY-SA 4.0, via Wikimedia Commons
.jpg){kind=link}
Box score: — · xG 0.00, xA 0.00 · ICT not yet finalized by FPL
The GW3 star of our positive list (15 points) came straight back down to earth: Palace conceded twice at home to newly-promoted Ipswich, and a big price-informed prediction built on last week's form met a genuinely bad afternoon for the whole back line.
Box score: — · xG 0.00, xA 0.00 · ICT not yet finalized by FPL
The same sub-60-minute story as his GW2 appearance in this exact section — substituted at 58 minutes of a game his team drew 2-2, still short of both the clean-sheet-bonus cutoff and, this time, an actual clean sheet to be short of. Hull's press-worthy squad depth cuts both ways: real chances for fringe players, and a defense that hasn't settled into one reliably-nailed back line yet.
Box score: — · xG 0.00, xA 0.00 · ICT not yet finalized by FPL
Sent off after 33 minutes in the same home defeat to Ipswich — a red card is the single most model-invisible outcome in football, and Disasi's finished the match with the worst score of anyone in our qualifying pool this week.
Photo: Kieran Clarke, CC BY 2.0, via Wikimedia Commons
_(cropped_x2).jpg){kind=link}
Box score: — · xG 0.00, xA 0.00 · ICT not yet finalized by FPL
Our GW1 top-five riser (14 points) played the full 90 away at Chelsea but conceded twice with no attacking return — the price-prior blend that correctly read his early promise hasn't yet seen him have a genuinely quiet away day at a top side.
Box score: — · xG 0.00, xA 0.00 · ICT not yet finalized by FPL
Substituted at 36 minutes as Palace chased the game against Ipswich — the third Palace player in our worst five from one match, the mirror image of Ipswich's Leif Davis appearing in our top five from the same 3-2 scoreline.
Where the model should improve
Today's fix, in one sentence: FPL's own defensive_contribution field is a cumulative raw action count (clearances + blocks + interceptions + tackles) despite its name, not points — and a bug was treating that count as if it were already the flat 2-point bonus FPL actually awards for clearing a position's action threshold in a match. A high-action defender's ~14 actions per 90 was landing as roughly +14 expected points from this one term alone, instead of the real +2. It's fixed now: a linear ramp toward the real 2-point cap as a player's per-90 rate approaches the threshold, verified against real players before shipping.
Two smaller, unrelated changes landed the same week: a UEFA Champions League rotation-risk discount (a club that played in Europe a few days before a Premier League match is now derated on start probability, since squad rotation after a midweek European trip is common and previously invisible to the model), and early-season per-team shrinkage in the ratings fit — the fix for exactly the failure mode behind GW3's Jayden Bogle miss, where one team's attack rating had swung to an extreme on just one or two small-sample results.
All three changes are retroactive: GW2 and GW3's posts have been updated with the corrected numbers (shown as struck-through old values next to the current ones), since the underlying model changed, not just this week's data.