Wrong Labels and the Craft of Fabrication: Data Discipline Inside a Sports Analytics Room

I opened the file tagged "tennis." The first line inside was about the Brent crude price. The third line was about talks between Washington and Tehran. The fifteenth line was about crude oil export volumes through the Strait of Hormuz. I read all thirty information points, read them a second time, then a third time far more slowly. Not a single player's name. Not a single court. Not a single set. Not a single first-serve percentage.

There were about two seconds—I did glance at my watch—when my brain automatically searched for a thread linking oil prices to tennis. Players' travel costs. Flights going up because fuel goes up. Tournament revenue. A transmission logic long enough to sound clever. Then I stopped, because I realized I was doing exactly what I had spent nine years warning myself against: pulling a thread from where there is no data to where I happen to need data.

This article is about that pause.

Every day, sports newsrooms run on automated classification infrastructure. An article is pushed into the system, the system assigns a topic label, and the piece is routed to the right expert. Tennis goes to the tennis person. Football to the football person. Finance to the finance person. In theory, this chain saves thousands of labor hours a week. In practice, it creates a new category of risk few people name out loud: the risk of fabricating facts out of blind faith in the label.

A label is an assumption, not a fact. But when the label comes from a machine, people stop questioning it. That is the psychological weakness any classification system can exploit. I have seen this take many forms: a stats table labeled with the wrong tournament, a metric assigned to the wrong season, a player recorded in the wrong position. In most cases, the receiving expert "corrects" the error by telling a story that fits the label, rather than calling the label wrong.

I understand why. In a newsroom environment, an empty analysis looks like a failure. An analysis with a conclusion—even a hollow one—looks like a productive day. The gap between those two things is the gap between a researcher and a text-producing machine.

I once stood on the wrong side of that line. In 2026, at sixteen, I built an Excel algorithm to predict SHB Da Nang's V.League results based on 120 prior matches. I published a "breaking the defensive meta" model on a forum, urging a back three and a high press. In the next two matches, the team conceded seven goals. The online community mocked me. I did not delete the post. I wrote another two thousand words defending the thesis.

To this day I keep that line: I was wrong about school football data, and that is the most accurate finding I have ever had. Because the lesson was not that I predicted wrong. The lesson was that I forced a formula onto a dataset that never contained the variable the formula needed. I did to SHB Da Nang exactly what I almost did to the Strait of Hormuz: take a beautiful analytical frame and stuff it with something irrelevant.

The difference between good analysis and sophisticated fabrication comes down to one question: does the data I am using actually contain the variable I am talking about? If the answer is no, every conclusion after that is literature, not science.

That is why I call my method data-crossing—and why I have to draw a red line around it. Data-crossing is legitimate when two domains share a real variable. When I connect the possession rate of Japanese youth teams to the misplaced-pass rate of grassroots sides, I cross domains through exactly one variable: pass efficiency under pressure. Data-crossing is illegitimate when I connect oil prices to player form through an imaginary variable—"cost," "psychology," "motivation"—and then paint it with a confident tone. The second thing is not analysis. It is an argument disguised as numbers.

At the 2026 World Cup, aged seventeen, I watched Japan beat Colombia 2-1. I counted fourteen crosses but only two touches inside the opponent's box. The old way of seeing called that waste. I wrote three thousand words proposing a "dead-ball cross" model—crosses that need no touch, existing only to stretch the defense. A major fanpage shared it, twelve thousand reads in two days.

Looking back, I know I was lucky. If Colombia's back line had not collapsed on cue to turn those fourteen crosses into a real threat, I would have told a story from a number that looked meaningful but was hollow. And I still remember that feeling—the feeling of someone who just read a correlation and mistook it for causation. It was not that Japan played beautifully; they simply exposed a formula the whole world overlooked. But I have to be honest: sometimes the formula I see is a formula I drew into the picture myself.

That is exactly the line I have to hold today, when a file tagged tennis contains thirty information points about oil, nuclear talks, and refined fuel markets.

If I were a content-producing machine, I would write this: "Oil prices rose more than two percent today, directly threatening the economics of tennis tournaments as players' travel costs spiral." That sentence sounds reasonable. It even has a number. And it is completely unproven by any line in the original article. The trap is that a wrong claim wrapped in right numbers outlives a bare wrong claim, because no one bothers to check a number that already has a source.

I learned this from another failure. In 2026, when the pandemic turned stadiums into rows of empty seats, I was nineteen, and I started a Telegram group called "Non-Governmental Football" with 47 members, testing match analysis through the sound of players' applause—because there was no crowd. By Euro 2026, my group predicted Italy would win based on a low-risk passing index. But I opened too many topics at once: tactics, finance, psychology. The Euro 2026 debate room collapsed because I thought every idea deserved a hearing. The group dissolved in three weeks.

The lesson is not "don't propose ideas." The lesson is: a debate room, and an analytics room, survives on one big experiment at a time. When I crammed five ideas into one piece, I did not enrich the argument—I diluted it until nothing was testable. The same mechanism runs when someone receives a mislabeled article and tries to stuff every scattered fragment into one neat conclusion.

In 2026, I joined Sports Illustrated starting as a fact-checker. That job taught me something no statistical model can: the value of a null result. When I checked a metric and found it does not measure what people say it measures, my biggest contribution was to report that it does not measure it—not to find a pretty interpretation for it.

Machines cannot do that. A machine only returns a label. A human has to be the one to say: this label is wrong.

The transfer business is where I see this labeling-and-fabricating mechanism most clearly, except there the label is money. Transfers are not mathematics, but mathematics explains why people go mad. When a free agent signs for a club, the press calls it a "free bargain." But the signing fees for a free agent—the loyalty bonus, the agent's commission, the inflated wage—are often more toxic than a transfer fee of the same value, precisely because they slip past FFP's core scrutiny. The label "free" is true in accounting terms and false in structural terms. Anyone who reads only the label and writes "cheap blockbuster" without opening the wage bill is reselling an assumption instead of analysis.

I once felt that from the other side, as the person whose idea got taken. In 2026, at twenty-one, I found a young Moroccan midfielder, Bilal El Khannouss, then eighteen, with a 91.3 percent pass-completion rate but playing in the Spanish second division. I wrote a potential analysis and sent it to five scouts on LinkedIn. No one replied. But an anonymous account used my idea to write for a European football outlet. Instead of getting angry, I took it as proof of my early trend-spotting. But I also asked myself: if that 91.3 percent figure had been attached to a different player in a different league, would I have caught the error—or would I also have written a flowery piece based on the label?

That question led me to a view I hold fairly rigidly: the tournament structure rewards storytelling, not disconfirmation. An analyst who says "I don't have enough data" is seen as unproductive. An analyst who says "this player has title potential based on three matches" gets invited on television. The result is that the entire opinion market tilts toward noise, and the real signals—rare and dry—get drowned out.

The same thing happens in youth development. There, the label "talent" is stuck on a fifteen-year-old, and from that second the whole system starts forcing an unformed body into adult match rhythm. The boy plays well at sixteen, gets pushed to the first team, carries minutes his muscle torque cannot bear. No one calls it a mistake because the short-term result looks good. But the long-term data—injury days at twenty-three—is a sentence written from the label stuck at fifteen. This is why I do not trust hype pieces about prodigies. I trust minutes-played-versus-bone-age charts.

The same logic applies to the file tagged tennis in my hands. The label says "tennis." The thirty data points say "energy and geopolitics." If I choose to trust the label, I will produce a flashy analysis, with tables, with terminology, and completely wrong. If I choose to trust the data, I have to say something that sounds like professional failure: this topic is outside my expertise, and I have nothing to analyze.

I choose the second option. Not out of morality. Out of technique. A tennis analysis built on oil news will be exposed by anyone who opens the original article, and the credibility lost in those thirty seconds cannot be regained by thirty good articles.

This is the counterintuitive part I want to put on the table. The whole industry is pouring money into improving the accuracy of labeling models—more data, more layers, more fine-tuning. But the deadly error is not in the model. It is in the human who treats the model's label as truth instead of assumption. A pipeline that mislabels five percent of the time is still safe if the receiving expert habitually asks "does the data actually contain this variable." A pipeline that labels correctly ninety-five percent of the time is still dangerous if the receiving expert believes it absolutely.

In other words: the problem is not that the machine labels wrong. The problem is a newsroom culture that does not allow anyone to say "I don't know."

I once thought contrarianism was saying the opposite of the majority. Now I think contrarianism is refusing to tell a story before the story has bones. I believe in data, but I believe more in the mistakes that data cannot measure. A wrong label is something data cannot measure. It only surfaces when someone is willing to sit down, read slowly, and say: this doesn't match.

So what does this mean for the ordinary sports reader, the one who just wants to know who beat whom today? More than they think. Every time they read an analysis with pretty tables and a decisive conclusion, they are betting that the label behind it is right. Most of the time it is. But on the day it is wrong—the day an oil piece is tagged tennis and someone still writes it up as if it were tennis—the reader is the one being led. Not because they lack intelligence, but because they have no tool to see the skewed label.

The only tool a reader has is the writer's transparency. And transparency starts with saying when we have nothing to say.

I left the thirty oil data points where they were. I did not turn them into tennis. Not because I lacked the skill. Because I learned, through SHB Da Nang's seven goals conceded and one collapsed Telegram group, that an analyst's real strength is not the number of stories told, but the number of stories refused. This profession rewards people who talk a lot. It pays people who talk right.

And if next time a file tagged tennis actually contains tennis—a player, a court, a serve percentage—I will write it right away. But I will write it differently. I will start with the number that lies, not the number that flies.

For now, the question I leave with your newsroom: what percentage of what you read this week was written from real data, and what percentage was written from a label no one bothered to check?