The PSX Filing That Got Tagged 'Tennis': When Algorithms Confuse, Data Journalists Must Be the Final Check

On a Monday morning in August, I opened my data inbox and found a filing from the Pakistan Stock Exchange (PSX). The first line was clear enough: Kashan Hasan resigned as Chief Executive Officer and Director of FrieslandCampina Engro Pakistan Limited — a company that sells milk. There was no serve, no set, no forehand anywhere in the document. Yet my content-classification system tagged this filing as 'tennis.' I sat there staring at the screen, and a silly question crossed my mind: what does a dairy cow know about match point? FrieslandCampina Engro Pakistan Limited, or FCEPL, is one of Pakistan's leading dairy companies, formed from a joint venture between Royal FrieslandCampina, the Dutch dairy giant, and Engro Corporation, one of Pakistan's largest industrial groups. The PSX filing showed that Kashan Hasan, a manager with more than 20 years of experience across Pakistan, South Africa, the UK, the Middle East and North Africa, had submitted his resignation. The notice said the resignation was effective immediately, but that he would continue to serve during the contractual notice period. The board described the vacancy as a 'casual vacancy' and said it would be dealt with in accordance with applicable legal and regulatory requirements. This is a textbook corporate-governance event, with nothing to do with sport. But the filing had flowed into my sports-analysis pipeline. That is when I realized the real issue was not the filing — it was the classification system. I opened the Stage-1 data table, where an algorithm had extracted 17 information points from the PSX notice. Every point was corporate data: filing date, title, board member names, succession plans. There was a $450 million foreign direct investment figure. There were more than 1,300 milk collection centres, two processing plants at Sukkur and Sahiwal, and the Nara farm. There was not a single word about serving, returning or break points. No ATP, WTA, ITF or Grand Slam. No player names. No matches. As a data journalist, I established an absolute rule after the 2026 V-League season. Back then I wrote the first series applying xG to Vietnamese football. The match between Hai Phong FC and Song Lam Nghe An at Lach Tray Stadium became a turning point for me. The home side created 1.92 xG but lost 0-1 because of an individual error. The media called it a 'decline.' I called it 'random injustice' — the opposition goalkeeper made 11 saves, 3.8 times the average rate. My article was ridiculed for two weeks, until Hai Phong's head coach publicly cited my numbers in a press conference. That experience taught me that data is never in a hurry. The one who rushes is the one who is wrong. But the same experience showed me another trap: the trap of trusting classification systems blindly. If I had not read the PSX notice carefully, I might have written a tennis analysis about an event that had nothing to do with sport. What would have happened? My readers in Vietnam would have received a thousands-word article about the 'groundstroke tactics' of a dairy CEO. They would have been confused, and I would have lost the trust I spent years building. Across my time working in both the US and Vietnam markets, I learned that numbers do not lie, but people — and algorithms — can. A good data journalist is not someone who produces many spreadsheets. He is someone who knows when to stop and say: 'I do not know.' Back to the FCEPL filing: I cannot write a tennis analysis about a corporate-governance event. That would betray my readers and violate my methodology. I can say this is a business news story, a sign of senior leadership change inside a multinational group, and that the mislabeling is a warning for the entire data-journalism industry. But I will not say this is a tennis match. People remember results. I remember the conditions that shaped results — and the conditions of this filing have nothing to do with any sport. What interests me about the FCEPL filing is not Hasan's resignation itself, but the process that led to the mislabeling. Among the 17 data points, some keywords may have confused the classification algorithm. I do not have enough evidence to identify the exact cause, but I have a hypothesis: the system was trained on an unbalanced dataset, where English texts from South Asia were frequently labeled 'sports' because of the overwhelming share of cricket in the source material. This is like a match in which the referee made a poor call — it forces the whole system to review the VAR. In a data environment, an item can be mislabeled for many reasons: an insufficiently diverse training set, an algorithm that misreads context, or simply a coding bug. To trace an error, I often use a method similar to reviewing match footage: rewind, freeze each frame, and compare time references. I noticed a pattern: PSX filings contain many proper nouns and abbreviations, which makes them easy to confuse with esports news flashes. But that is only a hypothesis. I need more verification before reaching a final verdict. Look at the $450 million FDI figure. In sports analysis, this number means nothing. But if someone accidentally used it to write about 'investment in a tennis training centre in Pakistan,' the story would become dangerously distorted. This is why I always ask my collaborators to cite the source of every figure, no matter how small. A spreadsheet does not generate data by itself. It only reflects what humans put into it. And if humans enter wrong data at the beginning, every analysis that follows will collapse. I once analyzed a series of physical data from the 2026 football season — the season when stadiums closed due to the pandemic. Many people claimed that statistics without crowd factors would be unreliable. But I saw it as the cleanest laboratory of modern football. No noise, no crowd pressure; players performed in a near-pure mental state. The results confirmed several tactical hypotheses. So even a classification error like the FCEPL filing can become an accidental experiment that reveals the limits of our systems. For Vietnamese sports journalism, this story is not unfamiliar. We consume a massive amount of international information every day. Machine-translation algorithms, topic classifiers and content recommenders are gradually replacing the editor's role. But algorithms cannot ask the question, 'Is this really sport?' or 'Is this source credible?' In an environment where speed is the top priority, a corporate-governance story being tagged as tennis is not an exception — it is a symptom. The FCEPL story also makes me think about how ready Vietnamese sports media is to embrace content-classification technology. In many newsrooms, editors still make the final decision. But the pressure to be first, to publish before rivals, is pushing humans out of the feedback loop. When an algorithm can generate headlines, summaries and even analysis, the role of a data verifier becomes even more crucial. I often tell my younger colleagues: every article is a hypothesis. Data is how we test it. But to test, we must first know exactly what we are testing. A shot recorded as a 'winner' in a tournament that does not exist is nothing but a phantom number in a spreadsheet. A goal scored in a match with the wrong timestamp is equally meaningless. People remember results. I remember the conditions that shaped results. And the first condition for a valuable sports analysis is to be sure that its subject is actually a sporting event. Someone might say, 'What is the big deal about a mislabeled tag? Just fix it.' But in data journalism, a mislabel is not like a typo. It is a systemic fault. Like a missed penalty in the 88th minute — it rarely comes from poor kicking technique, but from psychological pressure and a fracture in the process. If I had not scrutinized this filing, I would never have known that my data pipeline was leaking. And a leaking pipeline can ruin the best articles, the most accurate analyses, if it is not detected in time. Here is the paradox: if the FCEPL filing had never been tagged as tennis, I would never have spent time reading a notice about a Pakistani dairy company. This mislabel opened a door I did not plan to open. It forced me to look at an industry — the dairy industry — with the same rigor I apply to a tennis match. I discovered that Pakistan's dairy industry also has its own 'players': plants, farms, collection centres, and executives like Kashan Hasan. They also face pressure and experience ups and downs, but no one calls that a 'backhand.' In football, people are used to seeing a team struggle at the start of a season. A data analyst does not rush to a conclusion. He asks: does the poor result come from randomness, from a brutal fixture list, or from a genuine problem in play? Similarly, I need to determine whether this classification error is a random incident or a systemic syndrome. This is not a criminal investigation, but it requires the patience of a data detective. Next week I will review the entire list of items tagged as tennis over the past three months. I will look for strange entries — stories about milk, corporate governance, anything unrelated to balls and rackets. I will compare them against the list of actual sports events in the same period. I may find many surprises. But I will not jump to conclusions. Data is never in a hurry. The one who rushes is the one who is wrong. When a filing from a Pakistani dairy company can slip into a tennis-analysis pipeline, anything can happen. But that is exactly why my profession exists: not to chase every shot, but to ensure that every published number stands on a foundation of verified facts. The truth of the FCEPL filing is simple: a dairy CEO resigned. Everything else is noise. And in a world full of noise, the data verifier is the one who keeps the light on. So — have you ever asked yourself what your own system might be hiding?