A Non-Football Story Filed Under Football: Misclassification and the Price of Noise

At 2:47 a.m. in Guangzhou, I opened my tracking board as I have for nine years. On the screen sat a story from The Express Tribune, filed under the “football” label. Inside: Andy Cohen, the American television host, had been mistaken by a stranger for his own daughter's grandfather. No team. No player. No scoreline. Not one line about tactics, transfers, competition law or club finance. I read it three times, following professional habit: first for the facts, second for contradictions, third to be certain I had missed nothing. After three passes, only one conclusion held: the label was put in the wrong place. The moment the naked eye misses, the data never forgets. And the data here was unambiguous — there is no football signal inside a slot built for football. What kept me at the desk longer than necessary was not a presenter's family anecdote. What kept me there was the mechanism that pushed that story into position inside a feed I and thousands of others read every day. A single mislabel is not worth writing about. A mislabel repeated often enough is. Vietnamese football's information stream now runs on four layers: established outlets, aggregators, fan pages and automated distribution systems. The first three are decided by people; the fourth is decided by machines, and the fourth is the one that reaches readers most. A machine does not read the article. It reads signals: domain, original section, headline keywords, recognised entities, and the source's distribution history. When one of those signals is wrong, the article keeps travelling, because nobody stands in the middle to stop it. During a transfer window, noise multiplies. The number of stories per day rises, the time spent verifying each one falls, and the reward for publishing first is always larger than the reward for publishing correctly. A rumour about a release clause can be reposted twenty times in six hours, each pass adding a detail absent from the original. By the twentieth post, the original has vanished from the story, leaving a block of text with no traceable source. Football already has a verification layer built for this class of problem, and it is called VAR. The IFAB Laws restrict intervention to four categories: goals, penalty decisions, direct red cards and mistaken identity. From 2026, VAR began appearing in V.League 1 and quickly became a familiar talking point every matchday. What few notice: VAR does not judge emotion. It checks an incident against a framework written in advance. No framework, no VAR. No standard, no ruling. Content streams have no equivalent layer. We have editors, procedures, sometimes a morning meeting. We do not have a written framework stating which section a story belongs to, and what evidence permits it to sit there. So everything drifts on source inertia. Treat this labelling incident as a case, and dissect it the way I dissect a foul on review. On pure sporting grounds, the value is zero. No line-up, no build-up structure, no expected goals, no post-loss pressure index. On industry grounds, zero as well: no deal, no wage bill, no broadcast revenue stream. On timeliness, the value is low but not nil, since the incident occurred on September 10 and still sits inside the attention window. On reference value, zero: there is no model to learn from this story. I do not believe in luck. I believe in a number repeated a hundred times. Here, the repeated number is the count of times I had to discard, not the count of times I got to learn. The biggest trap in this trade is confusing “not enough information” with “wrong information”. The two states demand entirely different responses. With missing information, you go and collect more. With wrong information, you must pull it out of the stream before it produces consequences. Labelling errors belong to the second group, yet they are usually handled like the first: leave it, wait, then forget it. I still remember 2026, when I was sixteen, volunteering as a match statistician at a national U19 fixture. I was assigned to log forty-seven foul situations and twelve offsides. In the 78th minute I found that the referee had missed a foul inside the penalty area, and the goal arrived immediately afterwards. I built a comparison table against the IFAB Laws and sent it to the organisers. No reply came. The lesson was not “stop sending”. It was: if you want anyone to read it, record the data in a way that cannot be argued with, and accept that most errors will never be corrected. An empty stadium taught me that noise never scores. In 2026, working on a study of how crowdless stands affect match outcomes, I collected eighteen rounds of data and found home win rates falling from forty-two per cent to thirty-one per cent. I sent the report, and my supervisor advised comparing it against five previous seasons. Doing that, I understood the gap was too small to support a conclusion, because the sample was too thin and squad quality dominated everything else. From then on I set myself a non-negotiable rule: never use one season to assert a trend. That rule applies directly to this case. One mislabelled story says nothing about the quality of an entire system. But if the same source, the same section and the same error type recur, that is a trend, and a trend must be measured. I have no precise figure for the mislabelling rate in Vietnam's football feed, and I will not invent one to make this piece look more certain. That is a non-numeric grey zone, and every grey zone should be marked rather than filled with guesswork. What I can state firmly is the structure of the error. Mislabelling arises where three things meet: a source with an over-broad section, an entity-recognition system that is too sensitive, and an operator with no incentive to check. The more topics a source covers, the blurrier its section boundaries. The more sensitive the system is to proper nouns, the more likely it is to label a story on the strength of one name appearing inside it. And the more an operator is measured by traffic, the less reason there is to stop at an irrelevant article. The real cost of such errors is not the wrong article. The cost is that readers gradually lose the ability to tell a sourced story from an unsourced one. When labels stop being trustworthy, readers start trusting tone instead. Tone cannot be verified. And when belief migrates from data to tone, every football argument becomes an argument about the speaker, not about the incident. Every slow-motion replay carries its own truth. My job is to find the truth that cannot be disputed. In the VAR room, nobody debates what the referee was feeling. They place the measure against the image and read the result. A content feed needs such a measure too, however crude: what does this article give the reader, and what evidence stands behind it. Based on my experience watching matches, most controversial decisions are not decided at the moment of contact but by the observer's position. A referee standing in the wrong place sees nothing. An assistant standing in the right place sees everything. Same incident, two angles, two conclusions. Content distribution systems stand in a position worse than a badly placed referee, because they stand nowhere at all. They only read labels. Now the counterintuitive part. The instinctive response is to blame the algorithm. I disagree. The algorithm does exactly what it was told: optimise against the signals available. If the inputs are a broad section, a famous name and an attractive headline, then applying the “football” label is a rational output from a rational system. The problem lives in how we design rewards, not in how we write code. Here is the harder part: readers do not truly want verification either. They want story. An entertainment item tagged as football still generates traffic, comments and engagement. Accuracy generates none of that. Accuracy generates trust, and trust is only paid for later, when a reader needs real information to make a decision. Nobody pays for trust on the busiest day of the market. But noise wins a half, not the match. A system that lives on noise slowly loses its most valuable readers: the ones willing to pay to know the truth. That group is not large, but it stays longer, and it is the only group still reading to the final line. If I were to install a verification layer for football content, I would do three things. First, separate section labels from topic labels: an article can sit outside football even when it mentions a familiar name. Second, require every story to declare its origin, and show that origin to readers instead of burying it in code. Third, reserve a distinct label for unverified content rather than treating it as verified. None of those three layers requires sophisticated artificial intelligence. They require a person responsible enough to put the measure on the table and read the result, exactly as I do every morning. Football does not change because you look at it more closely. Football changes because you look at it more correctly. And looking correctly starts with the smallest things: a label placed in the right slot, a source written in plain sight, a number set beside its own context. The Andy Cohen story will drift out of the feed within days. The error that pushed it in will stay far longer. The question I leave for myself, and for anyone running any content stream: if your system cannot say “I do not know which section this article belongs to”, then it is lying with its own confidence.