The 67th Minute in the Manchester Derby: When Carrick Left a 10-Goal Striker on the Bench

The 67th minute. I sat in front of four windows open at once: a self-built xG table that had been running through the first half, a table tracking duels by the minute, a stopwatch placed beside a coffee that had already gone cold, and a live feed I kept only to hear the crowd. When Benjamin Sesko unzipped his jacket and stood up on the touchline, I did not look at his face. I looked at the clock. Sixty-seven minutes. A striker who closed the previous campaign with ten goals in fifteen appearances. Four Premier League matches this season, not one start. The man chosen for the centre-forward role was Matheus Cunha — whom Gary Lineker himself described as more a ten than a nine, and who was walking into a Manchester derby with a goals tally of zero. I have been in this trade for forty-three years, from the days of taking notes by hand in Madrid for World Sports, to now sitting in Saigon, building V-League xG models by hand, shot by shot. One lesson cost me dearly: a result is a single number, and a single number is never evidence. What interested me in the 67th minute was not which team was leading. It was the chain of decisions that dragged a person to that 67th minute, and the probabilistic price of that chain. That night I closed three windows and kept only the stopwatch. I wanted to remember the number. Sixty-seven. Not because it is elegant, but because it is late. The xG shock at Hang Day turned me from a watcher of football into a reader of data. In 2026, at Hang Day Stadium, Hanoi FC took seventeen shots, posted an xG of 2.87, and drew 1-1 against a Quang Nam side that managed two shots and 0.94 xG. I lost 180 million dong that night. What I lost was not the money. It was my faith in my own eyes. I stayed up, reviewed 112 V-League matches from round one to round fourteen, hand-calculated xG for every shot, and discovered something highlights never showed me: that team created plenty of chances but finished 23 percent worse than the league average. From that night on, whenever I look at an attack, I do not look at goals. I look at the gap between chances created and chances converted. And when I look at Michael Carrick's decision in the derby, I see exactly that gap — except this time it sat on the bench. Let me be clear about how I read a derby. A derby is not an ordinary match. It is a match in which two things compress at once: the pressure of error and the speed of decision-making. In a major national-team match, pressure comes from outside the stadium. In a derby, pressure comes from inside the players' heads, and it usually arrives fifteen minutes earlier than in a normal game. Under that compression, every managerial decision carries more weight than usual — not because it is more right or wrong, but because it is amplified. Carrick entered the derby with an attack he had chosen by a clear logic. He selected Cunha, Bryan Mbeumo, Marcus Rashford — names with the ability to move, to interchange, and more importantly, to take part in a high pressing structure. The opponent was Manchester City, a team operating with a very tight possession-control structure. When you face a possession team, there are two schools of response. One is to build a low block, concede the ball, and wait for errors. The other is to use a false nine, create numerical superiority in midfield, force the opponent to pass into crowded zones, and use mobile players to press at the source. Carrick chose the second school. On doctrine, that is an orthodox choice. The false nine is no novelty. It appeared long ago in modern football and it has its own logic. The problem with a doctrine is never the doctrine. The problem is the executor. Cunha is a technically gifted player, capable of receiving between the lines, turning, and stretching a defence. Those are precisely the qualities of a false nine in theory. But there is a detail theory does not tell you: a false nine must be able to convert the chances he himself creates. If he cannot, you have created a hole in the middle that no one fills. You gain a man in midfield, but you lose a man in the box. This season, Cunha has not scored. That is a number I cannot ignore — not because a player who does not score is a bad player, but because in a system that makes that player the centre of the final attacking phase, the number becomes a measure of the system's lag. And what I saw in the 67th minute was exactly that lag. I do not predict the future; I only read ahead the way the past keeps operating. When I look at the decision to keep Sesko out until the 67th minute, I am not looking at a personal mistake. I am looking at a model. The model works like this. Sesko closed the previous campaign with ten goals in fifteen appearances. That is 0.67 goals per match — not the rate of a substitute. Across four Premier League matches this season he was restricted to substitute cameos. In the derby he came on in the 67th minute, after his team had fallen behind. Alan Shearer said plainly that he should have been introduced at least ten minutes earlier. Ten minutes. I want to stop on that number — not to declare Shearer correct, but to measure what ten minutes are worth in probability terms. Imagine a team that must chase a scoreline in the final twenty-five minutes. Over twenty-five minutes, the number of chances a quality team creates typically falls between two and four. Shorten that window to fifteen minutes and you lose roughly forty percent of the chance volume available to your main striker. That is a simple calculation requiring no complex model. And it points to one thing: the decision was not about whether to bench Sesko. The decision was that the team had shortened its own window of risk exposure. There is a paradox here I have met many times. A manager chooses a player for tactical reasons — pressing involvement, link-up play — and then, when the match enters the phase that demands a goal, that same manager must turn back to the player he left out. This does not prove he was wrong. It only proves that tactical doctrine and the need to score are two different variables, and they only overlap when the executor of the doctrine can score. Kazan does not take revenge; Kazan only keeps the spreadsheet and waits for me to miscalculate. I once publicly predicted Germany's group-stage exit at the 2026 World Cup based on their pressing data. Their average distance covered fell 12.3 percent against the 2026 title-winning side. Their PPDA rose from 8.2 to 11.7 — meaning they let opponents pass more before contesting. When Germany lost 0-2 to South Korea in Kazan with an xG of just 0.41, and their final six shots all struck defenders, I understood something. Football does not punish anyone. It simply records the error line in the quietest of languages. And the error line in this derby sits in a very specific place: a striker with a strong conversion rate kept on the bench, while a player not scoring was handed the centre-forward role. No doctrine rescues an inversion like that, unless there is a variable I have not yet seen. I always look for the hidden variable. A data monk is not permitted to assume everything is already in the table. There are reasons a manager leaves a striker out without announcing them: load management, adaptation to pressing intensity, attitude in training, or an internal model that rates Cunha's link play above Sesko's pure scoring. I have no evidence for any of these. My confidence in such assumptions is low. But precisely because confidence is low, I am not permitted to use them to defend a decision that produced a bad outcome. What I can say with higher confidence is this: the tactical intent behind the false nine was most likely to create midfield superiority against City's 3-2-4-1 shape and to press from the front with agile runners. That intent is sound in theory. But it only holds if the player acting as the false nine converts the chances he creates. When Cunha fails to score and struggles to make an impact, the tactical rationale collapses under its own weight. One thing I learned from Kazan: when the model breaks, do not rush to blame the data. Check which variable you omitted. In this derby, the omitted variable may not be in the public data at all. It may sit in a training room, in a meeting, in a conversation no one recorded. But one thing I am certain of: a decision to send your main striker on after your team has fallen behind is a reactive decision, not a proactive one. And reactive game management always carries a probabilistic cost. Let me reconstruct that chain in numbers. Four Premier League matches, Sesko started none. He closed the previous campaign with ten goals in fifteen outings. Cunha has not scored this season. In the derby, Sesko came on in the 67th minute, after the goal was conceded. The team lost. The pundits said the attack lacked teeth and had no natural striker. If I chart those numbers, I get a very clear downward line: from a striker with a high conversion rate, to a player who does not score, to a late substitution, to a defeat. That line does not prove causation. But it proves the team chose a lower-probability option over an available one, and the cost of that choice lies not in the scoreline but in the fact that the team reduced the number of chances it could create during the decisive window. Belief is a noise variable; run the emotional regression before you place the bet. I write that sentence not about the fans. I write it about myself. There is a temptation every analyst faces: upon seeing a decision produce a bad outcome, we tend to turn it into a mistake, and turn the mistake into a list. Lists are easy to read, easy to argue over, easy to sell. But a list is not analysis. A mistake only becomes a lesson when we measure it in numbers and compare it to a baseline. What is the baseline here? The baseline is Sesko's conversion rate, 0.67 goals per match. The baseline is Cunha's goalscoring emptiness this season. The baseline is the twenty-five minutes a team chose for chasing a result instead of forty. Place those three baselines side by side and you do not need to judge. You only need to read the numbers. I have been wrong before. In 2026, when COVID-19 halted global football, the Bundesliga returned on May 16 in empty stadiums. I examined twenty-eight matches after the restart and found home teams won only five, 17.8 percent, against a historical home-win rate of 42 percent. My betting model then applied a home-factor of 1.32, and in one week I lost forty million dong. The crowd left, the model broke, and I learned to hear the breathing of an empty stand. I reviewed two hundred Bundesliga matches that season and found something: home teams still pushed forward as usual, but actual xG fell by 0.45 per match without a crowd. Within seventy-two hours I wrote "Home Is No Longer an Advantage" and rebuilt the entire system. From then on I designed a "context coefficient" — adjusting xG, PPDA and outcome projections for factors such as empty stands, weather, and travel distance. I say this to make clear that I am not standing above Carrick's decision to judge it. I stand beside it to measure it. And when I measure, I see something very clearly: that team invested heavily up front in the transfer window, yet failed to convert that investment into playing time for its biggest asset. That is a structural problem, not an emotional one. Now, the financial structure of a big club. I do not hold its broadcasting revenue, commercial revenue, wage bill or net debt. Those numbers are not in the dataset I am reading, and I will not invent them. But there is one number I do hold: the heavy investment up front during the transfer window, noted by Shearer, alongside his observation that the club did not strengthen defensively. That is an imbalanced structure. And an imbalanced structure has a very concrete efficiency consequence: each additional dollar spent on attack lowers its marginal utility, because you are adding to an already crowded area while leaving another empty. In economics, that is diminishing marginal returns. In football, it is the story of a team that scores plenty but loses the matches that matter because the lines behind cannot hold their shape. There is a second consequence, purely financial. A big contract sitting on the bench generates no on-pitch return. It generates only amortisation. If an expensive attacking asset does not play, its market value tends to fall, and the original investment becomes a latent loss. This is why big clubs must weigh immediate tactical needs against asset-value preservation very carefully. I do not have a specific fee for the transfer in this dataset. If the player referenced is a striker once pursued by several big clubs, a competitive premium is conceivable. But I cannot confirm that link from the data I am reading, so I record it at low confidence. What I can say with medium confidence is this: if a club spends heavily up front without balancing the defence, pressure under the Premier League's financial rules will grow, and the gap between fan expectation and actual results will widen. But I do not want to stop at finance. Finance is a variable, not a story. The story is in the people. Look at the player on the bench. A 22-year-old striker who once scored ten goals in fifteen appearances, entering a new season with enormous expectation, then sitting out four matches in a row. I do not know what he thought during that time. But I know one thing about the psychology of young strikers: silence on the bench accumulates in a way unlike any other position. A defender on the bench knows he may be chosen when the team needs to defend. A striker on the bench knows he is chosen when the team needs a goal — meaning he exists only in a state of emergency. And a state of emergency is a poor state for performance. It gives you no time to read the game, to find rhythm, to adapt to the opposing defence. It gives you twenty-five minutes and a single demand: score. That is a probabilistically unfair demand, and it is one every substitute striker in the world faces. When I watched Sesko enter at the 67th minute, I did not see a player. I saw an unmeasured variable. His conversion rate is 0.67 goals per match. But that rate was measured in starting conditions, with time, with rhythm. Put a player on at the 67th minute and you change the sample. You are no longer measuring the same variable. That is what I mean by the difference between absolute data and contextual data. A figure of 0.67 goals per match means nothing unless you say when it was measured, over how long, against whom. I learned this from the Hang Day shock. I learned it again from Kazan. And I learned it again from the empty Bundesliga stands. Each time my model broke, and each time I had to burn the original scripture. The day the model breaks is the day the data monk must burn the original scripture. And the original scripture always begins with the same line: no figure speaks for itself. Now let me address what I consider the biggest blind spot of this whole story. It is the confusion between two different kinds of question. Question one: was choosing a false nine sound? Question two: was sending the main striker on late sound? These are entirely different questions, and merging them is an analytical error. A manager can make a correct tactical choice at kick-off and an incorrect game-management choice mid-match. Those two do not exclude each other. Conversely, a bad outcome does not prove both were wrong. I stress this because I see many football analyses fall into the causation trap. When a team loses, every decision becomes a mistake. When a team wins, every decision becomes genius. But football does not work that way. Football works as a probabilistic system, where a correct decision can produce a bad outcome and a wrong decision can produce a good one. The analyst has a duty to separate the two. In this derby, there is one thing I can say with medium confidence: sending Sesko on in the 67th minute, after the team fell behind, was a reactive decision. The evidence is Lineker's own remark that the absence of a natural striker was glaring, and Shearer's remark that Sesko should have come on at least ten minutes earlier. Two independent pundits, one conclusion. That is not causal evidence, but it is a strong signal. The second strong signal is this: if the false-nine doctrine had meaning, it had to prove that meaning in the first half, when the match was level and the team still had time to adjust. If the intent produced no chances, then what needs changing is not the timing of the striker's introduction, but the entire doctrine. There is an image I cannot forget. It is the moment the manager stands on the touchline, clipboard in hand, and behind him, on the bench, a young striker is pulling on his boots. Neither man speaks to the other. Neither says a word. But the distance between them is precisely the distance the model has not yet measured. I want to tell another story, so you understand why I am obsessed with such distances. In 2026, after graduating from the Journalism Academy, I began my career at Football Newspaper, also working as a correspondent for World Sports in Madrid. Back then I did not know what xG was. I had a notebook and a pen. I sat in the stands, logged every phase, and tried to understand why one team beat another. I had no data. I had only my eyes. But my eyes taught me something data later confirmed: football is not in what you see, but in the gap between what you see. The gap between the shot and the goal. The gap between the pass and the chance. The gap between a striker on the bench and a match in progress. Those gaps do not appear in highlights. They appear only when you sit down and measure. Being 59 gives me a perspective: every cycle is a loop with a remainder. I have watched big clubs decline through financial imbalance. I have watched talented managers fail because a system did not fit the people. I have watched analytical models collapse from a single omitted variable. And in all those cases I noticed one thing: the remainder never disappears. It only moves from one place to another. In this derby, that remainder moved to the bench. It sat there for sixty-seven minutes, wearing a jacket, waiting to be called. Let me now consolidate the chain of evidence the way an analyst must — not to conclude, but to pose questions for the next round. First, on personnel selection, there is a gap between the substitute striker's conversion rate and the chosen player's goalscoring emptiness. That gap, viewed purely as resource optimisation, leans toward starting the striker with the higher conversion rate. Second, on game management, the timing of the 67th-minute introduction creates a shorter window of risk exposure than the potential window. In probability terms, this reduces the volume of chances the team can create in the decisive phase. Third, on squad structure, heavy investment up front without defensive reinforcement creates an imbalance. This imbalance lowers the marginal efficiency of each further attacking investment and raises the team's vulnerability against high-quality opponents. Fourth, on finance, an expensive attacking asset not regularly playing generates no on-pitch return and risks market-value depreciation. However, I lack sufficient figures to confirm the degree of this risk. Fifth, on hidden variables, factors may exist that were not reported — load management, adaptation to pressing, training attitude, or an internal model rating link play above scoring. I record these at low confidence. These five points do not form a verdict. They form an open dataset. And an open dataset is only valuable if you keep tracking and updating it. So what should the next round measure? I will measure three signals. The first is the starting minutes of that striker in the coming matches. If those minutes rise, my model was correct: the team corrected its error. If they stay low, there is a variable I have not seen, and I must return to the original scripture. The second signal is the position of the player chosen for the false-nine role. If he continues to be deployed there and continues not to convert chances, then the problem is not the player but the model: the team is putting a ten into a nine's role, and that error will repeat as a cycle in the truest sense. The third signal is the transfer structure in the next window. If the club keeps investing up front without reinforcing the defence, the imbalance will accumulate, and the price will not be paid in one derby but across a season. There is no such thing as a sweet bet; there is only mispriced probability sold as fair value. I am not saying that team will lose. I am not saying that manager will be sacked. I am only saying that probability is leaning in a certain direction, and the number 67 stands in the middle, as a boundary. There is one last thing I want to say, and it has nothing to do with the spreadsheet. That night, after the match ended, I turned off the screen and sat in the dark for a long while. I thought about that young striker on the bench. I thought about the moment he pulled on his boots, stood up, and was called on in the 67th minute. I thought about the sound of the crowd then — a crowd in a full stadium, not an empty stand, yet in my head it sounded at the same volume. The crowd left, the model broke, and I learned to hear the breathing of an empty stand. That breathing is not the lament of a disappointed crowd. It is the breathing of a system waiting to be re-measured. After every match, every decision, every error line, that system is still there, silent, waiting for the next person to read it correctly. I am 59 this year. I have kept notebooks since 2026. I lost money at Hang Day and rebuilt my credibility at Kazan. I have learned that football never answers your question with a single number. It answers with a sequence. And that sequence never ends at the 90th minute. The 67th minute in the Manchester derby is a number. But it is also an open question. And an open question, to a data monk, is worth more than any closed answer. Tomorrow, when that team walks out, I will open four windows again. I will press the stopwatch again. I will measure again. Not because I believe in the spreadsheet, but because I believe in the way the spreadsheet forces me into humility before what I cannot yet encode: a movement, a glance, a breath. Numbers lead. People are the final destination. And between the two there is always a gap the model has not measured — a gap I choose not to fill with judgement, but to leave open, waiting for the next round to replicate.