A Tennis Data File and Thirty-Seven Lines Without a Single Serve

On Tuesday night, a file landed in my queue with a single label: tennis. I opened it at 11:40 p.m., having just rewound three tapes of a Flushing Meadows quarter-final to check the return-points-won rate of the man who advanced. I read item one. Item two. Item three. By item thirty-seven I pushed my chair back, poured another coffee, and stared at the screen.

Not one player. Not one tournament. Not one coach, not one umpire, not one medical timeout, not one line about the 25-second serve clock. Thirty-seven information points, and every one of them concerned the KSE-100 index, crude oil prices, the Pakistani rupee, the meeting between Trump and Xi Jinping, and money flowing into artificial-intelligence stocks. A stock-market report wearing a tennis label, and it nearly walked through the door.

I have worked this trade for twenty-eight years, nineteen of them at the Daily Mail, and in my early years at Sports Illustrated I was placed on the fact-checking desk. There they taught me something I still keep today: a false fact lives longer than a missed one, because nobody issues a correction for something they never said.

Every morning my queue takes in dozens of files. Ranking tables, transfer dossiers, match reports, short wire items. Each file arrives with a domain label typed in by an operator, and after twenty-eight years I have learned that the label is the least reliable part of the package. Labels are typed by humans, usually at 11 p.m., after twelve hours of work. Entities, by contrast, are declared by the document itself, and documents do not get tired.

In 2026 I wrote a three-thousand-word analysis of Mohamed Salah after Liverpool paid 42 million euros to Roma for him. Colleagues doubted his body would survive the Premier League. I published the finding that his shooting and box-entry metrics sat in Europe's top five percent, and concluded he would score more than thirty goals. Salah scored thirty-two. In the same piece I declared that Gylfi Sigurdsson would dominate Everton's midfield after a 45-million-pound move, and he faded all season.

In 2026 I used xG to dismiss Croatia's run to the World Cup final as luck. The crowd pushed back, I retreated for a month, rewatched every penalty shootout and found that Croatia's keeper dived to his right 2.3 times more often than to his left. I built a private index out of it and dropped the word "deserved" from my vocabulary. Croatia won inside a sequence of events with a probability of roughly 18 percent, and the rest was something my data could not yet reach.

Since then I never stake my reputation on a single metric. I do not trust labels either. With every new file, I count entities first and read conclusions second.

On that Tuesday night, I ran exactly that protocol.

Across the thirty-seven information points, the entity chain read like a listing board: KSE-100, the Pakistan Stock Exchange, Topline Securities, MARI, PPL, HUBC, FCCL, LUCK, BAHL, FFC, MCB, and MSCI. The transmission chain the document described was consistent from start to finish: oil prices cooling after US-Iran tensions eased, dragging inflation expectations down, improving the external account, then pushing money into equities. A complete chain, internally coherent, and sealed inside finance.

Tennis transmits along an entirely different axis: tournament prize money sets entry density, entry density sets ranking points, ranking points set commercial value, commercial value sets team budgets. The two chains do not touch at a single link. Hand this file to a tennis model and it will not throw an error. It will run smoothly and return a meaningless answer, neatly formatted as professional judgment.

I cross-checked three times in three different ways. Counting entities across the full text: not one player appeared. Scanning for competition-rule keywords — serve, medical timeout, shot clock, line-call handling: nothing. Checking document structure: opening and closing patterns matched a market report, nothing like a match narrative.

Three layers, one answer. The probability that this file belongs to tennis, on my estimate, sits below one percent.

A null result, recorded properly, is worth more than a wrong conclusion presented neatly. In sports analytics we usually admit two states: win and loss. There is a third, rarely discussed — a data sample outside the domain under review — and that third state protects everything else in the system.

The document also lacked a concrete time anchor. No date, no trading session, no week. For a market report, a missing date destroys nearly all value. For a tennis report, the same holds. I logged that gap as a separate defect, distinct from the labeling error.

And I refused to interpret the file. I did not invent a player. I did not attribute an index decline to anyone's form. Fans watch with their eyes; I watch with a probability distribution — but a probability distribution only means something when you know precisely what you are measuring. The truth sits deep under the table of numbers, where headlines never reach; and sometimes the table of numbers has nothing to do with the story attached to it.

The biggest risk here is not the wrong label. A wrong label is trivial, one line to fix. The real risk sits downstream, where a loose gate lets the file flow into a model, and the model dutifully weaves a tennis story out of securities data.

I have seen that happen. I once watched a poor week of results explained through "dressing-room psychology," when the actual cause was three matches in six days plus a night flight. Correlation is not causation. A model without a domain gate cannot tell the two apart, and it will not bother trying.

This is also a resource-allocation problem. Clubs, federations and sports organizations spend on prediction models, scouting algorithms, polished dashboards. They spend very little on the first gate — the step that confirms the document entering the system is the document the system needs. Such a gate costs less than a model, and it blocks more disasters.

There is a parallel with the officiating story I have tracked for years. When a decision goes unexplained on court, the crowd becomes the forgotten party, and transparency is reduced to a slogan. A mislabeled data file produces the same effect: the end reader receives an unsupported conclusion with no way of knowing where they were led astray.

If my reading is right, the genuine tennis article was swapped during ingestion, and a Pakistan Stock Exchange report landed in the wrong slot. My probability for that scenario is roughly 80 percent — high enough to question the operating step, not high enough to indict a person.

The fix is clear: an automated entity gate running between the two processing layers, matching names of people, organizations, tournaments and seasons before the file moves on. Low cost. Build time measured in days.

I do not write about football; I only transcribe scripture from data. Thirty-seven lines, not one serve — and that turned out to be the most valuable information in the entire file.