As goals flew in across the 31 games in the traditional Saturday 3pm slot across England’s top four divisions, one game promised to deliver something historic.

Coventry City, managed by former Chelsea midfielder Frank Lampard, raced into a 7-0 lead against Queens Park Rangers after just 66 minutes, threatening to break the Championship’s eight-goal winning-margin record.

Ultimately, QPR salvaged the thinnest sliver of pride, allowing no further goals and grabbing a late consolation, but while the result was eye-catching enough, it was the post-match numbers that were particularly striking.

Coventry’s seven goals came from a total expected-goals (xG) value — the metric defined by statistics company Opta as measuring “the quality of a chance by calculating the likelihood that it will be scored by using information on similar shots in the past” — of just 1.27.

This kind of wild distortion between statistics and reality is rocket fuel for xG sceptics, purported proof that the models are broken. But the explanation is more nuanced.

Here, The Athletic explains how this remarkable goalscoring anomaly unfolded.

As with any xG overperformance, our first port of call is to roll back the tape and look at the finishes themselves. Did Coventry just hit their efforts absurdly well?

Looking at the shotmap below, it’s clear that speculative sharp-shooting played a big role in their result. Two goals came from outside the box, and the overall xG per shot stood at a paltry 0.07.

Take Coventry striker Victor Torp’s brace. His first, shown below, saw him reach behind him to guide the ball through a sea of bodies into the bottom-right corner. The xG value of this shot is 0.04, and it’s pretty reasonable to estimate that only one in every 25 of these efforts goes in, given the precise nature of the finish that is required with so many players between Torp and the goal.

His second was even more spectacular and unlikely: a ferocious curler into the top-right corner. The model puts it at a one per cent chance, and again, it’s difficult to argue with that assessment.

But stringing together that many low-probability finishes in a single game is incredibly rare. By analysing the individual xG of each of Coventry’s 19 shots, the model puts the chance of scoring seven or more at 0.004 per cent — roughly 1 in 25,000.

The unlikeliness the model assigns is matched by the observed data. Across 18,631 matches in England’s top four leagues, the Bundesliga, La Liga and Serie A since the 2019-20 season, only two produced a larger single-game xG overperformance: Wigan’s record-equalling 8-0 win over Hull in the Championship in 2020 and Mansfield’s 9-2 victory against Harrogate Town in League Two in 2024.

Every model throws up extreme outliers, and this match was just one of them. Typically, xG overperformance is centred around zero, but what stands out in the graphic above is just how noisy single matches are: on 59 per cent of occasions, there is a gap of at least half a goal between goals scored and expected goals.

Single games are subject to so much randomness that no one model can fully capture it. Even the chances that feed into xG are context-dependent over 90 minutes: a high-value opportunity can come from a calamitous mistake or a fortunate bounce rather than repeatable attacking play. That variance makes huge outliers like Coventry’s possible (though still improbable).

Over the long run, this noise disappears. It’s unlikely Coventry will keep scoring like this week in, week out. As the number of games increases, a team’s average gap between goals and expected goals shrinks. You can beat xG by a goal or more in a one-off match, but sustaining that average across a season is near-impossible.

In our data, German club Borussia Dortmund came closest in 2019-20, and even they managed only to exceed xG by 0.66 goals per game.

Another contributor to single-game xG volatility is variability in goalkeeping performances. This is where expected goals on target (xGOT) can help. xGOT is a metric that estimates the quality of on-target shots, taking into account factors such as the angle from which the shot was taken and its placement within the goal frame to give an indication of how likely the subsequent effort was to find its way in.

In most big overperformances, xGOT jumps well above pre-shot xG, pointing to stellar finishing.

Here, Coventry’s pre-shot xG was about 1.06, and xGOT lifted that only to 3.00, suggesting that QPR goalkeeper Joe Walsh should have prevented four goals. While his positioning and shot-stopping were far from pristine, this feels an unduly harsh assessment of his display on Saturday, and it’s hard to fault him for the majority of the goals.

That brings us to an undeniable truth: xG is an imperfect science.

It approximates chance-quality well, but no model can fully capture every single minute detail that shapes a shot. Instead, it’s a useful tool for long-term patterns where a large number of shots behave — on average — like the model predicts.

Coventry’s overperformance was mainly down to freakishly good finishing, but the model probably did undervalue the value of their opportunities. But the very rarity of an overperformance like this points to their clinical display being a massive outlier rather than evidence of a busted model.