A Pakistan Industrial Statistics Bulletin Landed in a Tennis Database: A Lesson on Information Filters

Wednesday. 9:14 in the morning. I open a file in the sports desk's data repository. The label on it: tennis. The assignment was to skim it and write a short piece on hard courts in the North American swing. I open it. Not a single player. Not a single court. No rankings, no draw, no serve. Just an agency called the Pakistan Bureau of Statistics, releasing provisional data on the Large Scale Manufacturing index. Two headline figures: up 3.03 percent year-on-year, up 9.51 percent month-on-month. Automobiles. Textiles. Pharmaceuticals. Chemicals. Furniture. Tobacco. Then I see it, sitting modestly among the industrial categories: "other manufacturing (football)." Four characters. Enough for a classification engine to nod and stamp it "sports." From sports, it slid into tennis. I sat still in front of the screen for a while. I have met mislabelled files many times in this trade. Never before had I seen so clearly the moment when noise defeats signal — and that moment occurred inside the very system we use to classify the sporting world. We are in the transfer window. Anyone in this job knows this is when the noise peaks. Every day, thousands of snippets cross the desk: a player changes his profile picture, an agent posts a photo from an airport, an anonymous account claims the deal is done. Readers drown in rumour. Practitioners drown in raw data. To handle that volume, most modern newsrooms rely on automated ingestion. Documents from newspapers, from social media, from statistical agencies, from club press releases — all flow into one pipeline. At the intake, a classifier reads the text and assigns a domain label: football, tennis, athletics, swimming, sports business. The file I was reading is the output of one such step. And it slipped through. In fairness, the document itself is not at fault. The Pakistan Bureau of Statistics is Pakistan's national statistical agency, the body responsible for publishing the QIM — the Quantum Index of Manufacturing, which measures manufacturing output volume against a base year. This is public data, with a methodology, with a publication cycle. It simply carries the wrong label. But a wrong label is worth discussing for anyone who works in sport. We mislabel all day long. We stamp "young talent" on a twenty-year-old and then read every pass he makes through that label. We stamp "finished" on a thirty-two-year-old tennis player and then ignore every metric showing he is still at peak fitness. The label precedes the data. When the label precedes the data, the data behind it becomes an appendix. In modern football, that label is even more powerful. A wide player today is taught to drift inside, to become a second threat in the box, to run into the space between full-back and centre-back. We still call him a "winger" — but that label describes a position that no longer exists in its traditional sense. The pure touchline winger has all but disappeared, and that disappearance was never recorded in a single statement. It simply happened, quietly, through thousands of small tactical decisions. There is a line I often write in my notebook: People look at the table; I look at what the table hides. With that file, what was hidden was its entire actual content — hidden behind the label "tennis." Before going further into the classification failure, I want to do what I always do with any report: check the arithmetic. The QIM for July 2026 stands at 119.13 points. The same month a year earlier: 115.62. Divide 119.13 by 115.62 and you get 1.03035 — a rise of 3.035 percent, rounded to 3.03. It matches. For June 2026, the index stood at 108.78. Divide 119.13 by 108.78 and you get 1.09515 — a rise of 9.515 percent, rounded to 9.51. It matches. Both headline figures reproduce exactly from the three published index levels. For a statistical bulletin, that is a real quality signal. Many reports I have read do not reach that level of consistency. Had I stopped there, I could have filed a short piece and gone home. But I read on into the sector detail. That is where the picture cracks. The automobile sector is mentioned twice with two different growth rates: 57.01 percent and 57.77 percent. No line specifies which period each belongs to. Furniture appears twice as well: 22.69 percent and 10.10 percent. More than double apart. One is very likely the sector's growth rate, the other its weighted contribution to the headline index — but the document does not distinguish. Chemicals: 0.25 percent and 0.50 percent. Tobacco: 35.82 percent and 0.55 percent. Sixty times apart. Then a corrupted line: "non-metallic mineral products posted a growth of 6.52 percent 4.25 percent." Two numbers glued together with no connective. Most likely 6.52 percent growth and 4.25 percent contribution — but a reader has no way to confirm. This deserves a longer pause, because it is the easiest thing to miss. In a report, people usually check only the headline. The headline reconciles, and that is that. But the headline is the most polished part, the most reviewed, placed in the most prominent position. The data that actually carries errors sits a layer down — in the sub-lines, the minor categories, the places the eye skims past because it assumes they do not matter. Some data does not need to be loud; it only needs someone patient enough to read it. The third group of problems is subtler, and in my view the most important. Scattered through the list are some very small values: 0.01 percent, 0.03, 0.04, 0.11, 0.18, 0.21, 0.27. Read under the label the document gives them — "growth" — this would mean that in a month when the headline index rose 3.03 percent, some sectors crept up by one-hundredth of one percent. That is self-evidently absurd. Read under another meaning — weighted contribution to headline growth — everything reconciles. PBS publishes two tables in parallel: each sector's growth rate, and each sector's contribution to total growth. A sector growing 12 percent but carrying only 2 percent weight contributes 0.24 percentage points to the headline. That is an entirely different story from a sector growing 0.24 percent. The two numbers tell opposite things. One says: this sector is booming. The other says: this sector barely moves the economy. This is metric conflation — and it is the most common error in every kind of data report, sports reports included. I meet it every week. Gross transfer value versus net spend. A club buys a player for fifty million and sells another for forty. The headline is "spent fifty million." Net spend is ten. Two numbers, two stories, and the second is the story of the club's financial health. Expected goals versus actual goals. A striker has an xG of 14.2 but has scored nine. Is he playing badly, or playing well with bad luck? The answer depends on which metric you read and over how many matches. Distance covered versus sprint count. One midfielder runs 11.5 kilometres per match but sprints above 25 km/h seven times. Another runs 10.2 kilometres but sprints eighteen times. Who ran more? It depends on how you define "ran." Tennis is the same. A high average serve speed does not mean an effective serve. One player serves at 205 km/h but wins only 62 percent of first-serve points; another serves at 188 and wins 76. The louder number is not the truer number. I once analysed a match in which the winner scored fewer total points than her opponent. She won because she won the points that mattered most. Read only the total — the most visible metric — and you reach entirely the wrong conclusion. The sports rights industry falls into exactly this trap. Streaming platforms buy rights on loud numbers — subscriber counts, viewing hours, engagement — and then report losses. Subscribers are not paying subscribers. Viewing hours are not revenue. Engagement is not loyalty. When the rights bubble peaks and the next contracts are renegotiated, people discover they have been reading the wrong metric for a decade — precisely as an industrial bulletin can be misread because of one wrong label. Elite sport is the art of repetition — and of breaking repetition. Back to the label. What makes me think hardest is not the error itself but how it happened. Across the whole document — a statistical bulletin running through dozens of categories — exactly one token smells of sport: the word "football," in brackets, after "other manufacturing." A keyword-based classifier grabs that token and labels the document sports. At the next layer, the sports label is refined into a specific discipline — and somehow it lands on tennis. I do not have enough data to describe the exact mechanism. But I have enough to say one thing: that system never checked whether the document contained a person, an event, or a sports governing body. Had it checked, it would have stopped. There is no name of any person in this document. Not one proper noun referring to a human being. Every subject named is a commodity category: automobiles, furniture, tobacco, leather, textiles, paper and board, iron and steel, beverages, rubber. For anyone writing about sport, this is a worthwhile reminder. A real sports story has people in it. No people, no sport. Only products. An industrial index can be right to the decimal point and still be soulless, because there is no one inside it to fail, to get up again, to endure. Arithmetic accuracy is not the same as human meaning. The second notable point: at the entity-extraction layer, the system behaved correctly. It found no sports entity and returned an empty result rather than inventing a name. Had it invented — had it attached some tennis player to this document — the consequences would have been far more serious. An empty file is merely useless. A wrong file is harmful. The fault lies in classification, not extraction. That is a small distinction with large consequences, because it points to where to fix: the gate, not the reading machine. I have worked this trade for twenty-eight years. I started in fact-checking — work no one sees and no one praises. For years I sat at the back of the newsroom, cross-checking names, dates of birth, shirt numbers, records. That work taught me something I still carry: most errors do not come from people lying. They come from people hurrying. Hurrying through a headline. Hurrying a label. Hurrying a conclusion. In the transfer window, that hurry has a price. A false rumour can nudge a club's share price, move ticket prices for a match, force a young player to issue a denial. I have seen articles written in forty minutes and corrected over the following forty hours. With that data file, the consequences were smaller. The cause was identical. In 2026, at the SEA Games in Kuala Lumpur, I was covering the women's 1500 metres. Nguyen Thi Oanh won that day, and I found something no one in the press room mentioned at the time: she had run a negative split — her first 800 metres slower than her final 700 by 2.3 seconds. That was a deliberate distribution of effort, and it explained why over the last 200 metres she was still accelerating while the others had faded. I took that analysis to my editor. He laughed and said women do not understand pacing. I did not argue. I spent three weeks rewatching the footage, drew the charts myself, and published it myself. The piece drew fifty thousand views in forty-eight hours, and the national team's head coach shared it. The lesson was not "I was right." The lesson was: correct data can still be rejected purely because of the label attached to the person presenting it. That label — "woman" — was stronger than the data. Exactly as the label "tennis" was stronger than the content of an entire statistical report. Rebellion is not necessarily shouting; sometimes it is quietly rearranging the numbers. There is a concept in data handling that sports people should know: the ability to return an empty result. A good system is one that can say "I found nothing." A bad system is one that always finds something — because it was optimised for finding, not for finding correctly. In the file I was reading, the extraction layer returned empty. That was correct behaviour. The classification layer has no such ability — it must pick a label, and it picked wrong. The same problem appears everywhere in sports media. When a player loses form, someone must explain it. There must be a reason: injury, age, a dressing-room rift, lost motivation. Sometimes the real reason is: he played normally and the ball did not go in for seven games. But "no reason" is not a sellable answer. So a reason is invented. When a team loses, someone must be blamed. When a match ends with an odd scoreline, a tactical cause must be found. Sometimes the real cause is: two shots hit the post. I am not saying analysis is useless. I am saying analysis should not be an obligation. In 2026, after the Kuala Lumpur piece, I was invited to Russia to commentate for a new sports platform. In the semi-final, I mispronounced Luka Modric's name three times in the first half. Social media reacted harshly. I locked my phone, retreated to the hotel, and cried for two days. But during those two days I still rewatched all five of Croatia's matches. I counted more than 90 kilometres of Modric's running across the tournament, and 14 chances created from his passes — passes the eye does not record because they do not end in a shot. The portrait I later wrote of his "invisible work" was shared by a Croatian newspaper. Moscow had snow, but Modric had a way of melting it with a pass. From then on I set myself a rule: three sources for pronunciation, before every broadcast. And a larger rule: move from narrating events to narrating meaning. Find the link between the smallest detail and the largest picture. A mistake can become material, if a person dares to face it rather than bury it. Now to the counterintuitive part. We tend to read a classification failure as a sign of a weak system. That reading misses something more important. A system that mislabels a document has a loose gate. True. But a system that dares to return an empty result when there is nothing to extract has discipline at a deeper layer. Both properties coexist in one pipeline, and the second is worth more than the first is worth worrying about. The paradox is that the gate is easy to fix. Add one check — does this document mention a person, an event, or a governing body. One question. If the answer is no, the document stops there. Discipline at the deeper layer is far harder to build, because it requires the system to accept that it may have no answer. This is where I think our trade is heading the wrong way. We build machines that always answer. We measure quality by speed and coverage. We praise a system for handling many documents, not for refusing in the right place. And I wonder whether the writing trade is following the same road. Every day we produce hundreds of pieces on the same event, all of them correct, all of them with numbers, and most of them saying nothing more. They are correct but soulless. They resemble an industrial statistical bulletin labelled tennis: formally accurate, fundamentally lost. I have no complete solution. I have only a small belief, formed over many years: a writer's value lies not in the number of pieces published, but in the number of times she declined to publish a piece merely because it was good enough to publish. In the end, I kept the file. I did not delete it. It sits in a separate folder, alongside a few dozen others of the same kind — documents that are factually accurate, numerically accurate, and out of place. I kept it because it reminds me of something this trade easily forgets: every label attached to a document is a judgement, and every judgement can be wrong. Readers do not need to know how many times we were wrong. They only need us not to be wrong the time they read. And perhaps the best thing we can give readers is not a verified conclusion, but a way to verify for themselves. A reader who knows to ask "what does this number actually measure" will never be fooled by a label — whether a machine applied it, or a newsroom did.