Three matches into the 2025/26 Premier League season, FIVDA published a goalkeeper leaderboard.
Guglielmo Vicario was first.
Using FBref-derived post-shot expected goals data, Tottenham’s goalkeeper had prevented approximately 2.2 goals more than expected. Brighton, meanwhile, sat at roughly -1.0 on the same early measure. FIVDA
The numbers were accurate.
The lesson we might have been tempted to take from them was not.
By the end of the season, Opta ranked Vicario bottom of the Premier League for goals prevented at -5.5. Bart Verbruggen travelled in the opposite direction, finishing third at +5.5. Opta Analyst Premier League
That does not mean the August numbers were useless. Vicario really had saved above expectation over those opening matches.
It means three fixtures told us what had happened.
They told us much less about what would happen next.
Figure 1 — Three matches can turn a goalkeeper ranking upside down. FIVDA Premier League Opta Analyst
Vicario is the perfect warning
Vicario’s reversal looks even more dramatic because FIVDA was not alone in identifying his strong start.
Opta’s own early-season goals-prevented ranking also placed him first at +2.2, ahead of James Trafford on +1.9 and Bernd Leno on +1.8. Premier League
Fast forward to the completed season and the picture had flipped.
Opta calculated that Vicario conceded 50 non-own goals from 44.5 expected goals on target, leaving him at -5.5 goals prevented, the lowest figure in the Premier League. Opta Analyst
His conventional numbers pointed in the same direction. Vicario faced 133 shots on target, made 83 saves and finished with a 62.4% save percentage across 31 league starts. FBref
But we should resist turning that into a neat story of a goalkeeper suddenly losing his ability.
The early and final figures in Figure 1 come from different post-shot models. FIVDA’s opening analysis used FBref PSxG data, while the completed-season ranking uses Opta’s xGOT framework. They belong to the same broad family of metrics, but they are not numerically identical models. FIVDA / FBref Opta Analyst FBref
So Vicario did not “lose exactly 7.7 goals” of shot-stopping quality.
What changed dramatically was the conclusion we might have drawn about his season.
Verbruggen shows the same problem in reverse
Brighton give us the opposite case.
After three matches, FIVDA’s chart placed them at approximately -1.0 PSxG+/-, suggesting their early goalkeeping had performed below the post-shot expectation of the chances faced. FIVDA
Verbruggen ended the season third in Opta’s goals-prevented table at +5.5. Premier League
He also faced a meaningful workload: 151 shots on target, making 103 saves for a 69.5% save percentage over 38 starts. FBref
Again, the interesting point is not that the opening numbers were wrong.
It is that goalkeeping metrics can move violently when only a handful of shots sit underneath them.
A striker may take several shots in one match.
A midfielder may complete dozens of passes.
A goalkeeper might face three meaningful efforts.
One exceptional save can therefore move an early shot-stopping metric much more than we intuitively realise.
What post-shot expected goals actually measures
This is where PSxG and xGOT become useful.
Traditional expected goals evaluates the chance before the goalkeeper’s involvement is known. Post-shot models incorporate what happened after the ball was struck.
Opta’s xGOT model combines the underlying quality of the original chance with the eventual location of an on-target effort. A shot placed towards the corner receives a higher probability of becoming a goal than an otherwise similar effort directed centrally. Opta Analyst
Goals prevented then compares the xGOT value a goalkeeper faced with the number of goals actually conceded. Opta Analyst
That makes the metric much more informative for shot stopping than a simple save percentage.
Imagine two goalkeepers both save eight of ten shots.
One faced ten relatively tame efforts.
The other faced shots repeatedly placed towards the corners.
Their save percentages are identical.
Their performances probably were not.
Post-shot models try to capture that difference.
Why clean sheets can mislead
Clean sheets answer an important football question.
Did the team concede?
They answer a much weaker individual question:
How well did the goalkeeper stop shots?
Arsenal provide the clearest 2025/26 example. Their goalkeepers faced only 89 shots on target across the entire Premier League season, while the team recorded 19 clean sheets. Brighton faced 151 shots on target, Manchester United 142 and Manchester City 126. FBref
The difference in workload is enormous. Workload also shaped FIVDA’s mid-season analysis of Robin Roefs.
Late in the season, David Raya was leading the Golden Glove race with 13 clean sheets but had made only 37 saves, the fewest among the contenders in the comparison. At that stage, Opta had him at -3.1 goals prevented. Premier League
Those facts do not diminish Raya’s contribution to Arsenal.
They show why team defence and individual shot stopping cannot be treated as the same thing.
A goalkeeper behind an elite defensive structure can contribute to many clean sheets while rarely being asked to make difficult saves.
A goalkeeper playing behind a more open defence may concede more often while still preventing more goals than the quality of shots suggests.
The full-season leaderboard looked completely different
Once the sample had expanded across the season, Opta’s top five contained Senne Lammens at +6.5 goals prevented, Gianluigi Donnarumma at +5.7, Verbruggen at +5.5, Jordan Pickford at +3.6 and Matz Sels at +3.6. Premier League
Figure 2 — Who actually prevented the most goals in 2025/26? Premier League
Even that list needs context.
Raw goals prevented is cumulative.
A goalkeeper cannot prevent a goal from a shot he never faces.
That means higher workloads create more opportunities to accumulate both positive and negative shot-stopping value. Opta therefore also uses a goals-prevented rate to account for the quantity of opportunities a goalkeeper receives. Opta Analyst
At one checkpoint, for example, Lammens had prevented +5.5 goals since his Manchester United debut and recorded a goals-prevented rate of 1.23. Emiliano Martínez was at 1.27 and Verbruggen 1.25, showing how an efficiency measure can order keepers differently from the raw cumulative total. Opta Analyst
So even “goals prevented” needs a second question:
Against how much work?
A full season helps. It does not solve everything.
Three matches are clearly unstable.
Surely 38 are enough?
Better, certainly.
Permanent, no.
None of the five leading goalkeepers for Opta goals prevented in 2024/25 appeared in the top five for 2025/26. Dean Henderson had led at +5.6, followed by Vicario +4.6, Mark Travers +4.5, Ederson +4.1 and Mark Flekken +2.8. Premier League
A year later, all five names had disappeared from the leading group. Premier League
Vicario is again the extreme example.
He moved from +4.6 in 2024/25 to -5.5 in 2025/26 using Opta’s goals-prevented framework. Premier League Opta Analyst
That does not mean goalkeeper performance is random.
It means one season still contains injuries, changes in defensive structure, different shot profiles and ordinary performance variation.
Repeated evidence is stronger.
Goalkeeper skill can persist
Jordan Pickford offers one example.
He recorded +5.47 goals prevented in 2023/24 and returned to the 2025/26 top five at +3.6. Premier League Premier League
Nick Pope provides a longer-horizon example. Across his first 67 Premier League matches for Newcastle, Opta calculated that the xGOT he faced exceeded the goals he actually conceded by roughly six goals, placing him among the competition’s more sustained positive performers. Premier League
That is the distinction worth making.
Goalkeeper ability is real.
Our certainty about it should increase with repeated evidence.
So how much evidence is enough?
There is no magic number.
But the hierarchy is useful.
Three matches can describe form.
A full season offers a much stronger estimate.
Multiple seasons tell us whether the performance survives different opponents, defensive structures and runs of finishing variance.
And even then, shot stopping is only one part of the position.
FBref’s advanced goalkeeper framework separately tracks launched passing, goalkeeper distribution, goal-kick length, crosses stopped and defensive actions outside the penalty area alongside PSxG measures. FBref
A goalkeeper who is merely average against shots might still transform a team by claiming crosses, sweeping behind a high defensive line or beginning attacks with his distribution.
There is no single number that captures all of that.
A better way to evaluate goalkeepers
The lesson from FIVDA’s August leaderboard is not that early analysis should be avoided.
It is that we should be clearer about what early analysis can tell us.
Vicario genuinely had an excellent opening three matches by the supplied post-shot measure. FIVDA
Verbruggen genuinely began below expectation on FIVDA’s early Brighton figure. FIVDA
Neither observation deserved to become a stable verdict.
A better goalkeeper evaluation starts with shot stopping, but adds workload, box control, sweeping, distribution and tactical fit.
Then it adds time.
That last ingredient may be the most important.
Football encourages us to rank players immediately. Three matches arrive and we want to know who is best. A run of clean sheets arrives and we want to know who deserves the Golden Glove. One spectacular save becomes evidence of reflexes; one mistake becomes evidence of unreliability.
Goalkeepers make that temptation particularly dangerous because their decisive actions are relatively rare.
FIVDA’s first leaderboard captured August accurately.
The season taught us why August was nowhere near enough.
Let us know your thoughts in the comment section below, we’d love to hear them!
Stay tuned for future posts and please send us a message if there is a specific topic you would like to see covered. Additionally, if you have any ideas on how to improve these articles, please send them our way too.
As always, thanks for taking the time to read these posts!
JC
Sources for the data presented are referenced within the article.
The main image is from Elio Santos on Unsplash with the article title overlay.