Wrong Labels, Skewed Trust: What a Defence Pact Misclassified as Sports Data Teaches Us

07:12 on a Monday in Sydney. My dashboard runs in the background, an automated data feed pouring in as it does every week. The label at the top of the file reads clearly: tennis. The next task is familiar — pull apart the serve, the share of baseline points won, the pulse of each set. But when I open the content there is no player. No court, no set, no tournament. What appears is a statement from Pakistan's Foreign Office about the Makkah Joint Defence Agreement between Pakistan, Saudi Arabia and Türkiye, along with developments surrounding Houthi attacks on Saudi Arabia. Twenty-one information points. Not one of them belongs to tennis. In eighteen years of watching the sports analytics industry, I learned something that seems obvious but is rarely said out loud: most mistakes in analysis come not from the maths, but from the label at the top of the file. A wrong metric can be fixed. A wrong label destroys the whole chain behind it. That day I did not write any tennis item. I wrote an internal note: the pipeline had assigned the wrong domain. But that note is the thing worth analysing, because it exposes a bigger question: how much do we trust sports data, and trust based on what. To see why a wrong label is worrying, picture how a sports data pipeline works. Everything starts from a raw event: a touch, a tackle, a serve. That event gets labelled — action type, pitch position, actor, outcome. From that label set, advanced metrics are calculated: xG, xA, pressing counts, distance covered, baseline points won. Providers such as StatsBomb or Opta sell precisely this label set, not just the final number. Anyone who uses their data inherits their event taxonomy. In other words, a number is not born on its own. It is built from a chain of classification decisions, and every decision can be right or wrong. Before you trust a number, ask where it came from. I came into this profession in 2026, aged twenty-five, doing data analysis for The Football Sack, a newly founded Australian football site. When the A-League reached round twelve, I published a three-thousand-two-hundred-word analysis of Melbourne City's pressing metrics, using GPS positional data to show that manager Warren Joyce's side was pressing in the wrong direction. Midfielder Luke Brattan ran 11.2 km per match but produced only 1.3 successful tackles. Fans mocked the piece as too dry. Three weeks later Joyce changed the pressing shape, and Melbourne City won four matches in a row. In 2026, writing for a small data blog, I published an English piece predicting Croatia would reach the World Cup semi-finals, based on their xG. Luka Modrić created 2.4 xG of chances per match in the group stage. A group of amateur coaches on Reddit called me a bookworm who did not understand football. Croatia reached the final. After the tournament, a journalist from The Athletic contacted me to ask how I calculated defenders' expected goals prevented. I spent two weeks writing Python, cross-checking against StatsBomb data, and sent back a seventeen-page breakdown. At the 2026 World Cup they laughed at my xG. This year they ask me what xG is. In June 2026, when the Bundesliga returned to empty stadiums, I was running a match-result prediction model for a data consultancy in Sydney. My model priced home advantage at 0.45 goals per match. After nine rounds without crowds, that figure fell to 0.08. I turned down a magazine's offer to write about crowdless football because I needed three more weeks of data to be sure. When I published, I stressed that I had been wrong not to include the crowd variable. Home advantage is not just geography, until it disappears. Those three stories share one denominator. Each revolves around correctly classifying a single variable. How many kilometres Brattan ran is a raw fact; calling it effective pressing or not is a classification decision. Modrić's 2.4 xG of chances per match is a raw fact; turning it into Croatia's midfield superiority is a classification decision. 0.45 and then 0.08 are raw facts; attributing the shift to crowds rather than scheduling is a classification decision. Misclassify one variable and you lose your bearings for a whole year. Back to the Monday incident. The problem was not that the data file concerned defence. The problem was that the pipeline called it tennis. One wrong label, and the entire nine-dimension framework I normally use to dissect a match — technical and tactical, form and data, tournament system, tour landscape, rules and governance, team management, risk, media narrative, industry transmission — becomes meaningless. All nine dimensions had to be filled with blanks: insufficient information to assess. That is a lesson in honesty. A decent analyst will not stuff fake player names into a defence report to make the framework look complete. They will say it plainly: wrong label, cannot assess. That restraint is not weakness; it is the condition for trust to survive. In football the consequences of a wrong label are far more visible. Take xG. To calculate it, the system must label every shot: location, angle, body part, defender pressure, the situation that led to it. Let a single cross be labelled a shot, or a long-range effort a close-range one, and the match xG shifts, and every conclusion downstream — who played well, who played badly, which team deserved to win — shifts with it. I once saw a dataset assign the same touch to two different players in two consecutive records. That dataset was not wrong in its maths; it was wrong at the label of the actor. The same happens with defensive labels. An interception and a tackle are two different actions, but in many older datasets they were merged. The result is that midfielders who read the game well — who intercept before danger forms — are undervalued, while those who react late are highlighted. A defender's expected goals prevented can differ by two-tenths if you change how one action type is classified. Over a thirty-eight-round season, that gap is enough to reorder a back line. Refereeing and VAR are where data labels collide with fan emotion most sharply. I hold to a view that has followed me for years: millimetre offside lines are slowly killing the attacking instinct. A striker learns to time a run to gain an advantage, and the system calls it offside because a toe or a shoulder leaned ahead a frame earlier. The measurement may be accurate, but the frame used to measure depends on which moment the system chooses as ball contact. Which frame you pick is a classification decision. Change the frame, change the conclusion. Referees are becoming the editors of the match, and viewers are learning that joy is only confirmed once the line has been drawn. Another labelling error fans rarely see is distance statistics. GPS measures total distance, but the label of heavy runs, sprint runs and maximal sprints decides what the number means. A player who runs 11.2 km mostly at a light pace is nothing like one who covers the same distance with twenty accelerations. Same number, different story, and the difference lives in the label, not in the metres. Numbers whisper. Those willing to listen hear an entire match. There is a deeper layer the Monday incident exposed: data labels are not only technical, they are also about power. Whoever defines the labels defines the truth. In modern football, the power to define labels sits with a handful of data providers, a handful of big leagues, and increasingly with state-backed corporations that have money. When money flows into a league, it does not just buy players; it buys the data infrastructure, it buys the way people see that team. Saudi Arabia is the example sitting right next to that mislabelled file. Exactly one week before I opened the Makkah file, I had read a report on the football investment wave from that country's public investment fund, and on Cristiano Ronaldo's move to play in the Saudi Pro League. They are two different stories on the page, but the same source of money. Transfer value is a story, but data is the signature. What is worth thinking about: a joint defence agreement compared to NATO's Article 5, with a planned permanent secretariat in Saudi Arabia, and a national football league funded by state resources, share one logic. Both build infrastructure to shape how the world reads them. In both, the most important step is labelling: what is called cooperation, what is called competition, what is called intervention, what is called investment. I do not have enough data to assert a causal link between defence money and football money in the Gulf. Two correlations running in parallel are not necessarily one causation. That is why I raise it as a signal to watch, not a conclusion. A season missing detail is like a match missing stoppage time. There is a counter-argument I must put to myself, because otherwise this piece would slide into conspiracy. It goes like this: mislabelling is the everyday reality of any automated classifier, from spam filters to image recognition. A sports analytics pipeline misassigning a defence article to the tennis domain is just an operational error. The right response is to fix the label and re-run the process, not to build a philosophical lesson about power. I agree with half of it. Yes, labelling errors are mostly operational. But one part I do not concede: once a wrong label is spotted, the analyst has a duty not to stuff in fake content to fill the template. The nine-dimension framework is sitting right there; a mechanical person will find a way to put something in every box. A decent person will leave many boxes empty and write plainly: insufficient information. That difference is not in the algorithm, it is in discipline. And discipline is what sports metrics lack. We have built sophisticated xG models, transfer-valuation models, result-prediction models that beat a coin toss by a few percentage points. But we have not built a widespread habit: before arguing about a number, check which labelling system produced it, by whom, and when. The who-matters question is very practical. A pressing metric published by provider A and one published by provider B can differ substantially, because their definitions of a successful press differ. When a manager explains that a metric led him to change his line-up, the first question should be whose metric he used. When fans argue over who defends better, the first question should be what defence even means here. Without answers to those two questions, the argument is just two people talking about different things. A colleague at the consultancy used to call this the label war. He was not wrong. In professional sport, most public debate is not about data but about labels. People do not argue about how far a player ran; they argue about whether to call that behaviour diligence or waste. Whichever label wins, that story wins. One more point I want to make clear, because I remind myself of it every week. Labelling mistakes do not distinguish newcomer from veteran. I was wrong in 2026 when I left the crowd variable out of my home-advantage model. I was right in 2026 when I used xG to predict Croatia, but that rightness was partly luck, and I am grateful I did not claim more credit than was mine. A mature analyst is not the one who errs least; they are the one who keeps records carefully enough to know where they went wrong. Back to the pipeline incident. I did three things. First, I reported the domain-mislabelling error to the operations team, with evidence that twenty-one information points contained not one tennis entity: no player, no tournament, no coach, no match. Second, I logged the whole episode as a test case for future label-audit procedures. Third, I refused every suggestion to stuff fake tennis content into the analysis to make the frame look full. All nine dimensions were marked: insufficient information to assess. That was the correct answer. I record this not to boast. I record it because in this industry an honest blank is worth more than a fake full box. Discerning readers spot immediately who is filling the blanks with air. And once they spot it, they stop trusting even your correct numbers. Trust is easily lost and hard to build. So what is the biggest lesson from a defence report wearing sports-data clothing. For me it comes down to one idea: the quality of a sports analytics system is not measured by how many metrics it produces, but by the honesty of the labels it applies. A good labelling system may produce fewer metrics but be more trustworthy. A careless labelling system can produce hundreds of beautiful metrics, none of them usable. Looking to the next cycle, I see three signals worth tracking. First, major data providers will be forced to be more transparent about their label dictionaries and labelling versions, because professional users are now asking questions they did not ask five years ago. Second, heavily invested leagues will build their own data infrastructure, and that will create multiple parallel versions of truth for the same match, demanding ever greater cross-verification. Third, domain-labelling errors — like the Monday incident — will surface more publicly, and how organisations handle them will become a measure of their credibility. I will keep logging the data version at the top of every analysis. I will keep asking every number where it came from. And I will keep leaving blanks empty when the data is not enough. If you see a sports analysis with no empty boxes, as flawless as an advertising flyer, read it with one hand on your wallet and one on your reasoning. Numbers whisper, but they only whisper the truth when labelled correctly. One wrong label, and an entire match falls silent.