Semenyo's Two Goals and the Gap 0.14 xG Cannot Measure

The 43rd minute at the Etihad. The ball spilled off a Sunderland defender's boot and rolled slowly toward the edge of the penalty area. Nobody in the stands stood up. It was that kind of dead ball that makes people turn to the person beside them, pour another cup of tea, glance at a phone. Antoine Semenyo did not turn to anyone. He planted his standing foot on the grass, paused for half a beat, and swung his right leg. The ball travelled flat and level, threaded through a forest of white-and-red striped legs, and settled in the net. The scoreboard jumped to 3-2.

Fourteen minutes later, in the 57th minute, he scored again, this time from a Sunderland defender's mistake just outside the box, a fierce strike stretching the lead to 4-2. Then came a pass laid on a plate for Rayan Cherki. Two goals, one assist, and the Premier League Man of the Match board with 33 percent of the vote.

I stayed seated long after the final whistle, looking at the stats board still glowing on the screen, and asked myself: if the model credits those two finishes with only 0.14 expected goals, where is the largest part of the story?

To understand why an afternoon like that deserves writing about, it has to be placed in its proper spot on the calendar. Manchester City hosted Sunderland in a Premier League matchweek on September 20, 2026. This is the stage of a season before the table has taken shape but after habits have already shown themselves. Sunderland — a club newly returned to the rhythm of the top flight — came into the match with a low, dense defensive block and a very clear plan: cede the ball, seal the central corridor, wait.

City were the opposite. The hosts controlled possession, pushed both full-backs high, and looked to stretch the opposition back line with sideways passes at moderate speed. The first half played out like a handball match: five goals were scored before the referee blew for the break. Neither side held its structure for more than ten minutes.

In that setting, Semenyo was the one man who kept his composure. He did not join the duels in midfield. He stood off to the side, waited, watched. And when the ball spilled out — as it always spills out in a match with five first-half goals — he was the man in the right place.

The ball rolls through my years, and I copy it down in verse — even when that verse has to be written in three decimal places. Based on my experience following matches over nearly five decades, afternoons like that handball match do not create stars. They only create early arrivals and late arrivals.

This is the part that needs slowness. Semenyo finished the match with two goals and one assist. His advanced statistics, per official Premier League data, read as follows: Expected Goals 0.14; Expected Assists 0.17; Expected Goal Involvements 0.31.

Anyone familiar with this kind of data will immediately spot the anomaly. A player scoring twice from 0.14 xG has done fourteen times what the model predicted. That is the level of overperformance analysts usually call noise, error, something that will correct itself within a few matchweeks.

But let us slow down a little. xG measures the quality of a shot, calculated from position, angle, number of defenders blocking, and the body part used. It does not measure the amount of time a player spent getting to that position. It does not measure the ground he covered to leave the crowded zone before the ball arrived. It does not measure the fact that he stood still while nineteen other players charged toward the ball.

The first goal in the 43rd minute is an example. The ball spilled outside the box. In that instant, every Sunderland midfielder turned back toward their own goal. Semenyo turned away. He retreated two beats, created separation, positioned himself at the ball's landing point, and struck it low across goal. From that spot outside the box with a wall of defenders in front, the model gives the shot a very low scoring probability. What the model cannot see is this: only a player standing in exactly that spot could take the shot, and only a player who had already anticipated the spill would be standing in that spot.

The second goal in the 57th minute followed a different route but the same logic. A Sunderland defender mishandled the ball just outside the box. The mistake by itself does not make a goal. Turning a mistake into a goal requires a player at the right reaction distance, the right pressing angle, the right speed of release. Semenyo was there. The shot was hard and precise, leaving the goalkeeper able only to watch.

Then came the assist for Rayan Cherki. The 0.17 xA says that pass, in the model, belongs to the modest-chance category. That is also true by the model's own logic: the value of a pass depends on the shot that follows, and the shot belongs to Cherki. People often forget that a pass only becomes an assist when someone puts it in. And Semenyo, before his name appeared on the board, had already twice chosen the right moment to keep the ball alive.

The three figures 0.14, 0.17 and 0.31 add up to a story about the gap between prediction and reality. And inside that gap, I see football.

The data analysts will have an answer ready. They will say: wait ten more matchweeks, these numbers will return to where they belong. Afternoons like September 20, 2026 are exceptions, and exceptions should not be used as a foundation for judging a player.

I do not object to statistics. I object to the habit of using them to freeze memory. The blind spot lies elsewhere. xG models are built on an implicit assumption: that a player's position is a given that exists beforehand, that where a player receives the ball is the business of the pass, and the player simply shoots. Semenyo breaks that assumption. He does not receive the ball in a high-probability position. He creates that position himself, within about two seconds, just as the opposition back line is turning away.

People call players like that lucky for three matchweeks, then call them durable when they repeat it all season. Labels do not live in a player's feet.

The Man of the Match vote is notable in another way. 33 percent is the highest share in the official Premier League poll, but it also means two-thirds of voters gave their ballot to someone else. On an afternoon when one player scored twice and assisted once, people still found a more compelling story elsewhere. That is evidence that the human eye is still hesitating in the face of everything the human ear has heard about data.

I once wrote that stoppage time is a debt I owe to time, and I must pay it every night. Modern football borrows in the same way. It borrows models to understand matches faster, and then must repay with sleepless nights, when a player does exactly what the model said could not be done.

That night I shut my laptop and walked a loop around the neighbourhood. There was nothing left to verify. The three figures were there, and one player had done the opposite of what they said. I write for that player, who does not know my name but keeps me up late.

The season is long. And the thing most worth waiting for is perhaps not whether those numbers return to their expected place, but whether anyone will bother to spend two seconds looking closely at the moment the ball spilled out, before it became a goal on a scoreboard.