A Tennis Report With No Tennis Player: When Sports Data Gets Mislabeled

Two in the morning in Nha Trang, and the analysis file lands on my laptop. The label at the top says one word: tennis. I scrolled down expecting a match, a court, a player with a serve drifting wide down the line. What sat inside was a tax document from Pakistan.

The withholding rates were listed in full: 6%, 7%, 12%, 14%, 15% and 20%. The document takes effect on 1 July 2026, issued by Pakistan's Federal Board of Revenue, applying to doctors, lawyers, architects, accountants and software engineers. Not a single player. Not a single tournament. Not a single court.

Beneath it were nine pre-built tennis analysis dimensions: technical and tactical, form and data, tournament systems, the professional landscape, rules and governance, team management, risk, media, industry transmission. All nine tables carried the same line: N/A. Attached was a cold note: off-domain, insufficient information to analyse.

My first reaction was laughter. My second reaction is what deserves writing down: in this trade, an empty file is the most honest file I have received in months.

Modern sports runs on labels. A single professional tennis match generates thousands of data points: first-serve speed, points won on second serve, spin direction, foot placement on the return. Across five sets the raw volume grows past the point where anyone reads it by eye. Machines read it instead, and machines read by label.

A label is how a system answers three things: what sport is this, who is this, what happened here. Answer correctly and the item goes straight to the editing desk. Answer wrongly and a Pakistani tax document walks calmly into the tennis pipeline, and nobody stops it at the door.

The cause is usually the dullest thing imaginable: keyword collision. A serve is a service, and a professional service is also a service. Advancing in a draw is advance, and an advance payment is also advance. A tennis court is a court, and a court of law is also a court. A short acronym like FBR is enough to drag an entire financial document onto a tennis court. No conspiracy lives here, only an overconfident filter.

What held me longer was the downstream effect. The label travels first; editing follows. A wrong label at ingestion gets read by an editor as fact, becomes a headline, becomes a reader's belief, and three months later nobody remembers where it started.

I learned this early, joining Sports Illustrated as a fact-checker. The work was unglamorous: re-reading every name, every date, every scoreline. But that is where I understood verification is not the last step before publishing. It is the first step, taken before the file is even opened.

In January 2026 I was seventeen, in my final year of school in Nha Trang, and I started a fan page called Phong Thay Do Nha Trang during Vietnam's run to the AFC U23 Asian Cup final. We lost that final 1-2 to Uzbekistan in extra time, after Nguyen Quang Hai had put us ahead. That night I logged 387 comments with the timestamp of every roar. The page reached 2,500 followers after the Russia World Cup.

Changzhou taught me that some heartbeats carry far without a goal. It also taught me something less romantic: the heartbeat of a stand only holds value if I label it correctly, which minute, which half, which emotion. Mislabelling one timestamp wipes out the meaning of the whole diary.

In 2026, V.League stopped for more than four months. The stands held nobody. I built a dataset of 124 matches involving Khanh Hoa FC and other V.League sides across the 2026-2026 seasons, and found home advantage collapsing: the home win rate fell from 38% to 23%. Khanh Hoa scored 0.7 goals per match before distancing and 2.1 after the restart.

Before those two rates existed, I had to do something nobody sees: reclassify all 124 matches. Which were played behind closed doors, which with limited attendance, which with registered spectators. Classify one match into the wrong group and the whole dataset drifts. Tonight's file taught me the same lesson at a larger scale.

A wrong label at ingestion does not die on the spot. It survives, puts on statistical clothing, and reaches the audience as a fact.

When the stands fall silent, I listen to the pitch through xG and find that data can tremble too. But I have to state what many data writers skip: xG shows where a shot came from, and never explains why we stayed out in the rain singing. Statistics answer what happened. They do not answer why we remained.

In 2026 I wrote my thesis on emotional statistics during Vietnam's final round of qualifying for the 2026 World Cup. After the 0-1 defeat to Japan on 11 November 2026, I collected 4,700 comments across three platforms to build an optimism index measuring deflation. On 1 February 2026, Vietnam beat China 3-1 and that index rose 212%.

That 212% taught me a methodological point: the loudest group is not the largest group. Had I labelled the two hundred most heated comments and called them the voice of a whole community, I would have committed exactly the error the system made with the Pakistani tax file: taking a convenient sample and naming it the entire truth.

There is another temptation for anyone writing with data, and I am not immune to it: turning statistics into poetry too early. My fix is mechanical. Write the bare number first, 23% is 23%, 0.7 goals is 0.7 goals. Only once the figure stands on its own do I allow a layer of feeling on top. Data can tremble, but only when we do not distort it.

Fear is pouring toward machines inventing players who do not exist. The larger risk sits on the opposite side: a system with no button marked I do not know. An engine trained to always answer will always answer, even when the answer is assembled from another country's withholding tax, thousands of kilometres from any court.

Tonight's file went against the current. It filled all nine dimensions with N/A, recommended quarantining the input, and stated plainly that any attempt to read it through a tennis frame would be fabrication. As a product, that is a failure: there is nothing to publish. As a craft, it is correct behaviour.

Imagine the other scenario. The system receives a Pakistani tax file, forces it into a tennis frame, and returns a fluent two-thousand-word analysis of tempo, endurance and tie-break pressure. It gets published. It gets shared. It gets cited. Nobody catches it, because it reads too smoothly.

Fluent and wrong is the most dangerous output in my trade. It makes no noise going through the door. It leaves only sediment in a reader's memory, and that sediment piles up over the years.

The same logic governs how I handle dressing-room information. In late 2026, tracking Khanh Hoa FC's transfer window as the club prepared for promotion, the agent of young midfielder Nguyen Minh Hoang agreed to hand me exclusive details of a loan move to Ha Noi FC. I wrote it, but lowered expectations instead of inflating them, and spent half the piece on the risks facing a 19-year-old at a big club.

Dressing-room access does not come from speed. It comes from having cross-checked two sources before pressing send.

Fans do not need a golden trophy; they need a reason to sing together in the street. A wrong reason is worse than no reason at all, because it makes people sing off-beat, then turn around and ask why they ever believed.

Starting today, every dataset I build gets one more column: where the source came from, who applied the label, when, and what verified it. That column will be slower, uglier, and it will never make the front page. But it is what keeps the rest from collapsing.

An empty file can be a sign of a broken system, or a sign of a mature one. The difference lies in whether anyone reads the notes carefully.

Next time a statistic arrives with total confidence, try one simple thing: ask who applied the label. Do we have the courage to open an empty table and tell the audience we do not know yet?

Keep the rhythm.