A 'Tennis' Label Stuck on a Tax Document: The Night I Audited the Sports Data Vault

On a Wednesday night, I sat in front of an old laptop in a small apartment by the Cam River. On the internal data board, one entry appeared under the label “tennis”. I clicked on it, expecting a match, a player, a surface. Instead I found an administrative document about taxation in Pakistan: the Federal Board of Revenue (FBR), a newly inserted sub-section of Section 25, and a re-audit process carried out by a cost accountant, alongside a revaluation of inventory. Not a single player. Not a single tournament. Not a single serve. Just a wrong label sitting inside the system, like a pebble in a running shoe, waiting for someone curious enough to bend down and pick it up.

Nguyen Thi Oanh is not a name — it is a life running forward. I wrote that line years ago and still keep it as a reminder: behind every label stuck onto a person or an event lies a whole world that cannot be reduced to a few characters. A “tennis” label stuck onto a tax document hurts no one. But it reveals a crack in the way we build sports knowledge vaults.

I came to this profession from the track, not from the server room. In 2026, I was a former youth athlete of the Hai Phong athletics team, sitting in the far corner of the press room at the national championship, and I misspelled Nguyen Thi Oanh's name three times in an 800-word piece. The national team coach texted me a correction that same night. From that small fall, I learned a professional rule: verify the name, verify the number, verify the context, before touching the keyboard. Years later, working with data tables and automated classification systems, I realised that rule is not reserved for writers. It is the rule of every data pipeline.

This story sits inside a larger current. Sports information today travels through three layers: upstream is youth development, equipment and venues; midstream is athletes, tournaments and tour systems; downstream is broadcasting, sponsorship and derivative markets. Between those layers, before any analysis happens, there is always a first gate: classification. An algorithm reads a text, assigns it a field, and pushes it into the matching vault. If that gate works correctly, a football story lands in the football drawer. If it fails, a Pakistani tax document lands in the tennis drawer.

The entry I opened that night belonged to the second case. Its content, read carefully, is quite clear and holds its own value within its own field: the FBR issued instructions to its field formations, allowing a commissioner to require a re-audit of a registered person's accounts by a cost accountant, together with a revaluation of inventory; the instruction was issued during the week, and the text mentions a reasonable opportunity of being heard before the power is applied. It is a state-administration document, neutral, intended to inform. It has a subject, entities and a system of rules. It is just that the subject, the entities and the rules have nothing to do with tennis.

This is where I want to pause a little longer, because it touches the core of the matter. A data entry can only be confirmed as belonging to a field when at least one recognisable entity of that field exists within it. For tennis, the minimum entity list includes: players, tournaments, ranking systems, surfaces, coaches, and governing bodies such as the ITF, ATP and WTA. In that tax document, not one of those entities appears. One could search forever without finding a player's name, a scoreboard, or a draw bracket. That absence is not a small detail. It is negative evidence, and negative evidence is still evidence.

The trap lies here: once an entry has been mislabelled, the pressure of content production pushes people to fill the template with anything at all. I have seen it happen. A bulletin needs nine sections, a form needs nine boxes, and so someone starts speculating: surely it relates to some match, surely this is data about an unheralded player. There is no evidence for those speculations. But they sound smooth, and smoothness is the enemy of truth. The safest rule I have learned, and the one I apply to myself, is the null-value rule: when there is not enough information, state clearly that there is not enough information, instead of inventing a conclusion that sounds plausible.

The consequence of a single mislabelling does not stop at that entry. A sports knowledge vault is a network of entity links. When a tax document slips into the tennis drawer, knock-on effects begin: search algorithms may return it to fans looking up a tournament; language models may learn from it; prediction models may swallow junk data. On a smaller but more intimate scale, a reader in Hai Phong looking for information about a young player may be led to a tax directive. No one dies. But trust erodes, bit by bit, in the dull, persistent way of a ligament injury.

I think about what I have written about transfers. Transfer data models tend to overrate young potential and underrate dressing-room chemistry. The same logic error repeats here: the algorithm matches keywords, sees a few familiar characters, and concludes. It overrates surface signal and underrates context. In football, I have also written that inverted wingers are homogenising the game, and that traditional wingers have been written off wrongly. Automated classification tends to homogenise in a similar way: it wants everything to belong to one drawer, as tidy as possible, and it dislikes loose entries, texts that refuse to sit still inside a frame.

But this story does not stop at the algorithm. The algorithm does the job it was taught. The real question is where people place the checking gate. In many newsrooms, and in many data systems, that gate is mistaken for a procedural step. It is handed to the machine and then forgotten. Meanwhile, a few individuals working slowly, reading every entry by hand and cross-checking manually, are precisely the ones who catch the errors. I carry one line with me always: The old laptop taught me: slow does not mean late, it only means telling the story another way. That old laptop, and the way I had to press every key, every line, every save, taught me that speed is not the only virtue. In sports data, publishing speed is often celebrated. But speed without verification only spreads the error faster.

This is the counterintuitive point I want to press: the things dismissed as outdated, slow and obsolete in sports information systems are often the safety net. A paper glossary gone yellow. A minimum entity list for each sport. An editor willing to read a long document to the end before tagging it. A hard rule: if there is not at least one recognisable entity, the entry does not enter the vault. These things are not glamorous. They do not generate beautiful growth numbers. But they are the boundary between a knowledge vault and a neatly arranged rubbish dump.

What I took from that Wednesday night is not a verdict on any specific system. I do not have enough evidence to conclude on the root cause, and I will not rush to conclude. What I have is an observation, a process and a reminder. The observation: that “tennis” label is an error, almost certainly belonging to the classification layer rather than editorial intent. The process: every sports data entry should pass through a domain-validation gate, based on the presence of entities, before it is stored. The reminder: in sport as in data, saying I do not know has never been a failure. The failure is saying I know when in fact you do not.

There is another line I often think of when deciding what should stay and what should be removed: An arena is not large because of its seats; it is large because of the stories that dare to stay. A sports knowledge vault is an arena like that. It does not grow by stuffing in more entries, right or wrong. It grows by keeping the right ones, and daring to remove the wrong ones. One refusal to tag, one re-tagging into the correct field, one explicit note that information is insufficient — that is not a step backwards. That is the real work.

Sport is a common language. People watch a match in Hai Phong, in Ha Noi, in Sai Gon, and understand each other without translation. But a common language is only useful when its words are used correctly. Mislabelling a Pakistani tax document as tennis is a mispronunciation, a small one, but it can spread. The job of people in my trade is to catch those mispronunciations and correct them before they become habits. Not because I fear technology. But because I believe a sports knowledge vault is only trustworthy when every entry in it deserves to stay.