Take Riyad Mahrez. Across the sample in one academic study he sat 14.61 goals above what xG models expected — the classic signature of an elite finisher. Then the researchers removed 17 deflected shots, 5.9% of his total. His number fell to 9.03. A third of the "finishing skill" was deflections.
That study — from the sports analytics group at KU Leuven — is the clearest statement yet of something we build around: goals minus xG looks like a finishing-skill meter, and it isn’t one.
Three reasons the meter lies
1. Almost nobody shoots enough. To show, statistically, that a player finishes 25% better than average, he needs roughly a hundred shots a season — a volume only a handful of players ever reach. Below that, a hot goals-minus-xG number is indistinguishable from a hot coin.
2. The shots aren’t all the same kind of evidence. Deflections, speculative hits, garbage-time attempts — all of them count equally in the total. That is the Mahrez problem: the metric moved by a third when 6% of the shots left.
3. The model isn’t a neutral judge. An xG model learns from every shot in its training data — and good finishers take more of those shots than average players do. The “average” the model measures against is quietly above average. The KU Leuven team showed the effect directly: feed the model more shots from good finishers and Messi’s above-xG number falls; compare him only against other high-volume attackers and it rises by about 17%. Same player, same shots — the verdict moves with the training data. Their recalibration suggests standard xG underestimates skilled finishers by roughly 3% on average.
What we saw this weekend
Our own numbers tell the same story at team level. We measured 1,237 consecutive ten-game stretches across 15 leagues: finishing above chance quality carried almost no persistence from one stretch to the next.
Then this weekend staged the perfect demonstration. On Sunday we published Inter’s finishing ledger: +11.8 goals above xG across ten games, one of the hottest runs in our data. That night they created 3.80 xG against Cagliari — thirty shots — and scored once. The deepest under-performance of their window, hours after we charted the over-performance. The ledger didn’t lie; it just never promised to continue.
So what is goals-minus-xG good for?
This is the chart we ship on every match — the ledger. How to read it: each bar is one match, its height the gap between goals scored and what the chances were worth. Violet above the line: the side banked more than its chances — a deflection that crept in, a worldie, a keeper’s bad night. Grey below: chances made, goals missing. Only the two extremes get labels, because the extremes carry the story; the rest is texture. The strip above is Inter’s — the +11.8 window, and the rightmost bar is what arrived the very next night.
And why the chart exists at all, given everything this article just said: because the gap between goals and chances is real information about the past even when it predicts nothing about the future. It tells you whether a run of results was built on chances (which travel) or on conversion (which, as above, mostly doesn’t). It is the difference between a table position you can trust and one that flatters. The ledger doesn’t grade anyone — it keeps the books.
Description. It is a faithful record of what happened: who banked more than their chances were worth, who banked less, when, against whom. That record is worth reading — it explains results tables that make no sense, it flags when a scoreline flattered a team, and over a window it shows you a side’s recent reality.
That is exactly how Broovo uses it. Every match carries a ledger strip — one bar per game, goals against chance quality, violet over the line, grey under. We’ll tell you a side has finished above its chances in nine of ten. We won’t tell you they are “clinical”, and we won’t tell you it continues tonight — because the statistics above say nobody honestly can. We read what happened. We don’t grade the finisher.
Related reading
- Two titles, two different teams — a rare case of overperformance sustained, recorded either way.
- What is xG? — the 24-second explainer behind every number here.
Sources
- Davis, Robberechts et al. (KU Leuven DTAI sports analytics lab): Biases in expected goals models confound finishing ability. The Mahrez, shot-volume and training-data figures above are theirs.
- Persistence: our study of 1,237 consecutive ten-game stretches across 15 leagues.
- Inter’s window: our served reads and final match sheets, 30–31 August 2026.
