When an HIV Drug Story Gets Tagged Football: Pipeline Error and the Cost to the Trade

On 25 September I opened a file sitting in my editing queue. The classification tag read, in two words: Football. The first line mentioned Cofepris, Mexico's federal health-risk regulator, and a decision to grant marketing authorisation to lenacapavir, a long-acting injectable used for HIV pre-exposure prophylaxis. I scrolled through all twenty-five information points in the file, then sat back. Not one team. Not one player. No match, no contract, no table, no minute of stoppage time.

Memory pulled me back to an afternoon in June 2026 in South Korea, when I was a final-year sports management student sent out as a field reporter at the U20 World Cup. In the quarter-final between U20 Vietnam and U20 France I mispronounced the name of Jean-Kévin Augustin three times in a single first half. Viewers called in to complain. The name I got wrong that year remains the most expensive lesson journalism ever handed me. Only today did I see the industrial version of that mistake: a content pipeline mispronouncing an entire industry, with nobody calling in.

In fifteen years of watching this trade I have never seen sports content produced at today's volume. One newsroom in Shenzhen runs a morning with as many items as a European desk produces in a week. Most of them are not written by people. They pass through several automated layers: one reads the raw text and assigns a topic label, one extracts entities, one builds the article skeleton. The editor only touches the last stage.

Labelling is the cheapest stage, and the most lethal. A classifier needs only three overlapping keywords to feel confident. Our trade shares an entire vocabulary with medicine, law and public administration — words that, read without context, drag a pharmaceutical story onto the pitch.

The first keyword is "registration". In football, registration is the procedure that enters a player into a competition, a transfer window, a squad list. In that file, registration is a marketing authorisation issued by a health regulator. The machine saw the word, saw Mexico, and its memory held thousands of articles about transfers in the Americas. Match.

The second keyword is "regimen". A regimen is a drug schedule. The pipeline reads it as "plan", and from "plan" the bridge to "tactics" is one step long. One more step is enough for the system to believe this document describes how a team operates.

The third keyword is "trial". In medicine, a trial is a clinical study with a code, a protocol and an enrolment number. In sport, a trial is a friendly, a run-out for a formation. The original article contained a trial named PURPOSE 2 with 2,180 participants and a 96 per cent reduction in incidence — very specific, very numeric, exactly the kind of data a sports analytics system is trained to hunt.

The fourth keyword is "commission". Disciplinary commission, league commission, referees' commission. The machine saw a commission issuing a ruling and assumed it was a football governing body.

Those four collisions together were enough for a medical file to wear a sports file's clothes and reach my queue. What chilled me sits somewhere else: had I skimmed, it would have passed.

This is where our trade is changing nature. Ten years ago a bad story had to get past a sceptical editor. Now it only has to clear a confidence threshold — and that threshold is set by the very system that produced it.

Colleagues know me as a verification addict. In 2026, while the whole desk fixated on Kylian Mbappe staying at Paris Saint-Germain, I noticed something small: Erling Haaland's representative had hired a law firm based in Manchester to handle image rights. I called a source close to the Borussia Dortmund coaching staff, confirmed the sixty-million-euro release clause had been triggered, and published forty-eight hours before the club announced. That process had three steps: build a source file, cross-check the agent's transaction history, and always write "according to a source close to the situation" instead of asserting certainty.

I have also read enough heat maps to know they are becoming a new form of fortune-telling. A beautiful heat map, a striking pie chart, an expected-goals figure presented without context — all of them can conceal a player's real role inside a tactical system. They are not wrong about the data. They are wrong about the story people attach to the data. This morning's HIV file is the crudest version of the same disease: correct data, wrong label, story imposed from outside.

Where could I be wrong? Three possibilities, and I put them on the table before concluding.

First, the mislabel may be deliberate. If this file was pushed through as a test of whether an editor would notice, my writing this piece proves the test succeeded — and also proves it failed, because I lost half a day to a file that should have been blocked at the gate.

Second, the problem may not be systemic. A single error inside a large corpus says nothing about the whole pipeline. If this is one case in ten thousand files, the error rate is low enough to accept, and turning it into a professional lesson is exaggeration.

Third, and most importantly: am I even doing my job by writing about a medicine? I have no expertise to judge how well lenacapavir works. I only have expertise to judge that it was labelled wrongly. That border is thin, and I have to remind myself not to cross it.

The silence after the whistle is the passage I most enjoy writing. In this file, the whistle never sounded.

My prediction, and I accept that it can be checked: before the next major tournament closes, at least one sports outlet will publish a transfer story or a line-up breakdown built entirely from a non-football source that was mislabelled. It will not sit in an editor's queue the way this morning's file did. It will sit in your feed, with a smooth headline and an illustrative photo.

I do not write to be loved; I write to make people stop. If you have read this far, you have stopped once for a file nobody was supposed to see.

The pitch never lies — only I once misheard a name. Next year's problem is that somebody will mishear an entire industry, and nobody will sit back the way I sat back this morning.