A Cargo of Electric Buses Labeled as Tennis: Notes from the Data VAR Room

At two in the morning I opened a record in the analytics system. The classification field carried one word: tennis. The body below described 179 electric buses built in China arriving at the port of Karachi, Pakistan, with Karachi Port Trust as the receiving authority and Margalla Heavy Industries of Taxila Cantt as consignee. I scrolled to the end. No player. No set. No ranking, no ATP, no WTA, not a single line of service rules.

I sat still for about thirty seconds, long enough to wonder whether I was reading the wrong column. Then I did what anyone in the VAR room has to do: rewind to the first frame, read line by line, take notes on paper.

There are offsides nobody sees, but the camera never blinks. This time, the thing that blinked was the labelling layer.

What the record actually held

Eleven information points, and all eleven concerned vessels docking, cargo discharge, port authorities and consignees, plus two statements from the port body itself. The goods originated in Yingkou, China. Quantity: 179 electric buses. The Karachi end was handled by Karachi Port Trust. The consignee was Margalla Heavy Industries, Taxila Cantt.

The only time reference in the entire text was the word “Wednesday” — a day of the week with no absolute date attached. The source was recorded generically as “KPT”, with no news organisation named and no publication date. Two of the eleven points were the port authority's own assessment: the port has the capacity, Pakistan is moving toward clean transport. A self-praise ratio of 18 per cent in a single item is the fingerprint of a press release taken verbatim, not of independently verified reporting.

To a sports editor, this is a worthless record. To someone who does verification work, it is a valuable specimen: a clean mislabel, clear enough to dissect without guessing.

Vietnamese readers follow international tennis almost entirely through aggregators. No newsroom here has the staff to cover every Challenger or every qualifying draw in person. We live on the data feed that flows in. When that feed is contaminated, the person who pays is the reader who believes he is reading sports news.

Our process runs in two layers. The first reads the text, extracts entities, assigns a domain label. The second — where I sit — builds the full analytical frame: technique and tactics, data and form, tournament structure, tour landscape, rules and governance, team organisation, risk, media narrative, and the transmission path down into the industry. This record passed the first layer carrying the tennis label, and landed on my desk.

Why three letters fooled an entire system

KPT. To anyone reading Pakistani business news, that is Karachi Port Trust. To a keyword matcher that sees only a string of characters, it is a three-letter combination that could map onto several things. The phenomenon has a name: abbreviation collision. One acronym, multiple real-world entities, and an entity-extraction layer without a dictionary rich enough to tell them apart.

I have seen the human version of this error on the pitch many times. An assistant referee who watches the colour of the shirt instead of the line of the run. A referee who reads the crowd instead of the position of the foot. Same mechanism: the system takes the most legible signal, not the correct one.

What struck me was that the labelling layer never hesitated. It assigned tennis with high confidence to a text with zero tennis content. That confidence chilled me more than the wrong label itself.

Nine analytical dimensions, nine times “insufficient information”

Sitting down at the desk, I had two options. One was to stretch a few details until they fit the template: treat 179 buses as player data, Yingkou as the Chinese market, Taxila Cantt as a training base. The other was to record exactly what was there.

I chose the second, and the result was nine analytical dimensions and nine entries of “insufficient information”. No player means no playing style. No match means no first-serve percentage, no return points won, no break-point conversion. No tournament means no seeding, no draw, no calendar. No tennis governing body appears anywhere in the text, so there is no integrity question to raise.

There is a principle in refereeing that gets misread more than any other: overturn a decision only when there is clear and obvious evidence. Outsiders hear “be very sure before you change your mind”. People inside the job hear “without evidence you do not guess, even when the guess turns out right”.

In 2026, in the round of sixteen at the World Cup in Russia, Spain against Russia, minute 42, I missed Gerard Piqué's handball inside the penalty area. The referee reviewed it and awarded Russia a penalty. The match ended 1-1 and Russia won on penalties. I blamed myself for three weeks, quietly rewatched all 64 matches of the tournament and wrote notes on every VAR incident.

The lesson I took was not “see better”. The lesson was speed. That room moved faster than human cross-checking could follow. I proposed a process that pauses two seconds longer before a decision, and I still hold that proposal today — only the target has changed. The labelling layer needs those two seconds too: a consistency gate between label and content, running before a record is pushed downstream.

A millimetre changes a team's fate; I have learned to live with that. But a millimetre only means something when there is a line to compare against. Without the line, the number is just an illusion presented neatly.

In 2026, in the AFC Cup group stage, Hai Phong against Ceres-Negros, minute 78, I spotted that the visitors' striker Fidelis Ikiri was 0.3 metres offside before he scored the equaliser for 2-2. I sent the signal up to the officials. The goal was disallowed. The match finished 2-1 to Hai Phong. The coaching staff never knew I had intervened, so no one thanked me. I tell the story to make a point about verification: corrections tend to be silent, while misses always make a sound.

Based on my experience tracking and cross-checking thousands of VAR incidents, most errors do not come from weak eyesight. They come from systems designed to always produce an answer.

When a wrong label flows downstream

A mislabelled record does not die where it was created. It travels. Into model training sets, into aggregate tables, into the weekly report sent to editors. Three months later a reader in Vietnam opens a tennis piece and finds a paragraph about electric buses, or worse: finds a “trend” assembled out of rubbish.

I have watched this kind of contagion for years in the transfer market. An