When the Algorithm Called an Earthquake Siren a Football Match

In the data table I opened at two in the morning, one row was tagged "football." Its content was about 23,000 earthquake-warning loudspeakers, about President Claudia Sheinbaum, about Mexico's Second National Drill scheduled for 12:00 on September 19, 2026. Not one player. Not one club. Not one matchday. An earthquake siren had just been named football by an automated classification system, and it was sitting in the queue waiting to become "sports news." If you have worked in this trade long enough, you will not laugh. You will feel a chill down your spine. What just happened was not a trivial software bug. It was a crack in the foundation of the entire sports data journalism industry — where a number placed in the wrong slot can steer millions of readers in the wrong direction. When the press room laughed at xG, I knew I was reading the right book that they had never opened. But this time, my own book was being misread by an algorithm. Every modern sports newsroom runs on a data pipeline. Raw stories pour in from thousands of sources, pass through an automated classification system, and get labeled before reaching an editor. The first stage of that pipeline does one thing: it assigns a topic label. "Football." "Basketball." "Transfers." That label decides which analytical framework a story falls into, which expert reads it, and which readers see it. For a forecasting system, a single wrong label at the input is enough to distort the entire chain of computation downstream. The problem is that the labeling system does not understand football. It counts keywords. And the story about Mexico is full of words that made it think it had found a tactical analysis: "scenarios," "protocols," "response," "regions," "containment." Five regional scenarios for the drill, plus emergency response protocols, were enough for the algorithm to nod. It did not check whether any player existed. It did not ask whether any league existed. It simply saw "scenarios" and "protocols," and slapped on the label. More telling still, the decision to keep the familiar sound of the alert — so residents would not confuse a drill with a real earthquake — is a text about public-safety communication, entirely foreign to any sports subject. This is the kind of error I call a "lexical false positive." In a model validated across 10,000 matches, you can trust the probabilities. But in a ruleset that counts keywords, you are trusting only the coincidence of language. A single number can lie, but a model validated across 10,000 matches has no reason to pretend. The same principle applies to the data pipeline: a topic label is only trustworthy when it rests on entities, not on verbs. Look at what the story actually contains, and the seriousness becomes clear. Across all 11 information points, there is not a single football entity: no club, no coach, no league, no governing body. The figures extracted are President Claudia Sheinbaum, civil-protection authorities, Mexico City, and the seismic-alert loudspeaker system. The only two notable numbers — 80 million mobile phones and 23,000 loudspeakers — are emergency-warning infrastructure, not broadcast revenue, not wage bills, not transfer values. Sheinbaum's former role as head of the Mexico City government, along with her past testing of alternative alert options, is a matter of public administration — with no analogue whatsoever in a dressing room or club governance. In quality control, a sample like this is called a "negative control": a known-wrong input used to test whether the system is sober enough to reject it. Here, the system failed. It did not reject. It accepted, labeled, and forwarded. If I had not blocked it, that data row would have entered a tactical-analysis framework, where an editor would have been forced to write about the "tactical shape" of 23,000 loudspeakers. That is not a far-fetched hypothetical. That is how hollow analyses are manufactured every day, and how reader trust is eroded bit by bit. This is why I never let an automated label determine my judgment. Based on my years of experience watching matches and monitoring data flows, I have learned one rule: before analyzing anything, verify that the central entity exists. If no club, no player, and no football governing body appears in the text, then every tactical conclusion drawn from it is fiction. And fiction in data journalism is not creativity — it is fabrication. The counterintuitive angle here is this: the greatest danger is not the misclassification system, but the reflex of misanalysis that follows it. An algorithm slapping on a wrong label is fixable. But when an analyst receives already-mislabeled data, and under pressure to produce content begins to "invent" football meaning from an earthquake bulletin — at that point the pipeline is not merely wrong, it is contaminated. Keyword correlation is not causation. The appearance of the word "scenario" in a document does not turn it into tactical analysis, just as a match with many corners is not thereby an exciting match. I have witnessed this same type of error in forecasting models. When the context shifts — as in the 2026 season played in empty stadiums — every old metric becomes noise, and models that failed to update their adjustment coefficients produced wrong forecasts en masse. The common thread: a system operating correctly in mechanism, but no longer correct in context. What needs fixing is not the number, but the conditions under which the number is produced. An empty stadium does not erase the truth. It only strips away the fog that 40,000 shouts once created. A wrong label is the same — it does not create a new truth, it merely hides the fact that there was no truth to state. The signal for the next cycle is clear. The sports data journalism industry needs an entity-based validation rule: a story may only be labeled "football" when at least one club, player, or football governing body exists in the text. No entity, no label. That is the minimum barrier to protect readers from forcibly grafted news. And one question I leave for myself, and for those who do this work as I do: if an earthquake siren can slip into a football feed without anyone stopping it, how many tactical analyses you have read are, in fact, sirens placed in the wrong slot?