When Tennis Data Turns Out to Be a Stock Ticker

2:47 a.m. in Hai Phong. I opened an item in my feed tagged tennis and read the first line: the KSE-100 index gained 830.43 points, or 0.48 percent, on volume of 773.59 million shares worth 26.45 billion Pakistani rupees. Below it sat the refinery group PRL, ATRL, NRL, CNERGY. Further down were international oil prices, de-escalation signals between the United States and Iran, and an IMF mission reviewing Pakistan's seven-billion-dollar loan programme.

No player. No court. No ATP, WTA or ITF, no Grand Slam, no set, no scoreline.

The only word connecting that text to my world was points.

I sat there another forty minutes, not to write, but to understand what had just happened to the system I trust every night.

I have worked in this trade for twenty-eight years. In 2026 I joined Sports Illustrated as a fact-checker, cross-referencing hundreds of figures a day before they went to print. That job taught me something I have carried ever since: most errors in a story do not live in the sentences, they live in the labelling. Someone classifies an event wrongly, and everything downstream drifts with it.

In 2026, at the 29th SEA Games in Kuala Lumpur, I was the only woman in the athletics press area. In 2026, my tactical piece took me to the World Cup in Russia as a broadcast analyst. A year later I sat in a hotel room and cried for 48 hours because I had mispronounced the name Luka Modric three times in the first half of a semi-final.

Then came 2026, when the pandemic wiped out the calendar. My Dinh Stadium stood silent for 214 days. I left Hanoi for Hai Phong, closed the door of my reading room, reopened my master's thesis in sociology, and started a newsletter called The Empty Track - one legendary race each week, set inside the social context that produced it. By year's end it had 3,200 subscribers, mostly coaches who had lost their training grounds. The empty track is where I hear my own footsteps most clearly.

That was also when I built my own data feed. Every night it scans thousands of articles, tags them by sport, and pushes the worthy ones into my field of view. The feed has two layers. The first classifies the subject. The second breaks out the detail: technique and tactics, data and form, tournament structure and calendar, the professional landscape, rules and governance, team and athlete management, risk, media narrative, and the value chain of the whole sport. I designed it around a single principle: never invent. If a dimension lacks data, it must say so plainly - insufficient information to assess.

That night, layer one returned the label tennis. Layer two, exactly as designed, refused to conjure a player out of nothing.

The output was a table of nine dimensions, and all nine returned the same sentence. Technique and tactics: empty. Data and form: empty, despite very specific figures on hand, including a level of 172,232.51 points. Tournament structure and calendar: empty. Professional landscape and player standing: empty. Rules and governance: empty. Team and athlete management: empty. Risk: empty. Media narrative and expectation: empty. Tennis industry transmission chain: empty.

I read that table and felt lighter. An honest data system is measured by how often it dares to say I do not know, not by how often it dares to assert. Some data does not need to be loud; it only needs someone patient enough to read it.

But stopping there would make this a minor technical glitch. The real story sits elsewhere.

Look at the keywords that fooled the tagger. Points. Gains. Rally - a recovery, and in tennis a long exchange. Upper circuit - a price limit, and in sport a tour structure. Sector - an industry group, and in some sports a competition area. Every one of those words is legitimate tennis vocabulary. They were simply in the wrong place.

I have seen that exact mechanism in my own trade. In 2026 I found that Nguyen Thi Oanh won the women's 1500m with a negative split: her first 800m was 2.3 seconds slower than her closing 700m. When I pitched the tactical breakdown to an editor, he laughed and said women do not understand pacing. I did not argue. I spent three weeks rewatching every tape, drew my own charts, and published it on my personal blog. It reached 50,000 views in 48 hours and was shared by the national team's head coach.

The lesson was not to believe in myself. The lesson was that when someone mislabels your data, you have to reopen the raw table with your own hands. People look at the rankings; I look at what the rankings hide.

A year later, in Moscow, I mispronounced Luka Modric three times in one half and took a wave of online criticism for it. I withdrew to my hotel, cut off contact, cried for 48 hours - and still rewatched all five Croatia matches. Moscow had snow, but Modric had a way of melting it with a pass. Across that tournament he covered more than 90 kilometres and created 14 chances from his passing, a figure later cited by Croatia's Sportske Novosti. From then on I set a three-source rule for every name, and shifted my writing from narrating events to narrating meaning.

Mislabeling works the same way: it is not frightening because it produces one bad article, but because it quietly flows downstream.

For three nights afterwards I counted my own feed. Across more than four thousand items, I logged a few dozen carrying a sports label without a single sports entity inside - no tournament name, no athlete, no governing body. The rate was small. But it was steady, and steadiness is what worries me.

Picture a sports prediction model receiving this item. It reads 830.43 points and files it under match score. It reads upper circuit and classifies it as a tournament round. It reads PRL, ATRL, NRL and CNERGY as four clubs. Nothing explodes immediately. The error accumulates, silently, until one day someone looks at a statistical table and finds a number that looks perfectly reasonable and is entirely hollow.

In sports data, the deadliest errors always live in the middle layer - in classification, in entity checks, in the gates nobody wants to build because they are not glamorous. People love to talk about AI writing commentary, about goal-prediction models. Very few want to talk about a fence that checks whether a player's name is real before labelling an oil-price story as tennis.

I cross-checked once more against the VuaBong.vn database. No tennis entity matched. No player, no tournament, no match. The label stood alone with nothing behind it.

Three signals are worth tracking if anyone wants to turn this incident into serious work: how often sports-labelled items contain no sports entity; whether an entity-verification gate exists before classification; and the clusters of keyword collisions between financial language and arena language. Fix those three and the quality of an entire pipeline changes.

There is one more layer, and it belongs to the business of sport. Data platforms are selling each other auto-aggregated information packages, and the end buyers are bookmakers, analytics firms, and newsrooms that have cut their verification staff. When the middle layer rots, the loser is not the machine but the reader who trusts the number. Data does not generate trust on its own. Trust comes from someone who took the trouble to sit down and check.

At 44, after all these years, I believe the opposite of what most people are saying. They fear machines writing instead of journalists. The real problem is machines labelling instead of people, with nobody checking the labels.

Most assume a bad data item just needs deleting. I think a bad data item is worth more than the correct story next to it. It is a case study: it shows where the mechanism broke, at which layer, and how.

There is another counterintuitive reading. Layer two's refusal to analyse is precisely the strength worth learning from. Inexperienced writers fill gaps with inference. Experienced ones let the gap stand and tell readers that this part is unknown. That is a form of courage few applaud, because it produces no handsome headline.

Rebellion does not have to be loud; sometimes it is quietly rearranging the numbers.

I still keep that strange item in a folder of its own, named 830.43. Not to remind myself of an incident, but to remind myself that trust in sports data has to be built from small bricks: one source cross-check, one entity-verification gate, one time daring to say there is insufficient information to assess.

Elite sport is the art of repetition - and of breaking repetition. Sports data is the same: it lives on the times we patiently read again, and the times we dare to strike something out.

So when did you last check, with your own hands, the label your system put on your data?