One Fictional Salah: The Labeling Error That Let Football Data Fool Itself

On 22 September, at a film premiere in New York, 29-year-old actor Shaheen Jafargholi spent a few minutes talking about the second season of the HBO series Heated Rivalry. He spoke about playing Fabian Salah, about being a fan of the source material long before he auditioned, about a small, welcoming and safe crew.

Not one sentence mentioned football.

And yet, as that interview passed through an automated classification layer, it came back tagged “football” — not a faint ranking hint, but a topic label solid enough to push the piece into the very data stream I read every morning.

I sat with it for a long while, looking for what I had missed. A club? No. A match? No. A manager, a transfer, a goal? Nothing. The only thing in the whole text that breathes sport sits in the last two syllables of a fictional character’s name: Salah.

What is being said

Heated Rivalry is an HBO production adapted from Rachel Reid’s Game Changers novel series; the adapter is Jacob Tierney. The cast includes Justice Smith, Charlie Gillespie, Emily Hampshire and Jafargholi. Season two is scheduled to air in spring 2027. The interview was captured by PEOPLE and relayed by The Express Tribune. That is the entire factual set.

In football, “Salah” is the surname of one of the most widely recognised players on the planet. Any entity-recognition system that scores proper nouns by frequency will grab that string and reach for the football drawer. Add “Shane Hollander”, add “Ilya Rozanov” — names that read like athletes. Add a book series titled Game Changers. Three weak signals, no strong one, and the system chose football.

What actually happened

An entity-recognition system does not understand football. It counts names. When a string matches the name of a famous figure, the model drags in the entire semantic field attached to that name — and football, one of the most frequently mentioned topics on the internet, is the heaviest semantic field in the room.

The label then does not sit still. It flows into dashboards, into entity graphs, into the aggregated wire copy that many small newsrooms buy. A false node is born, and nobody removes it, because nobody sees it.

A topic label is not a fact about content. It is a claim that requires verification, and in a data stream with no verification gate, one duplicated name is enough to contaminate an entire database.

I came to this through a working habit. In 2026, when the leagues stopped for the pandemic, I dug through The Football Database to see what disappears when the stands are empty. I found a number: across 2026–2026, away teams in the Premier League scored 43% of their goals in the final 15 minutes. From then on I set myself a rule: every claim must be checked against at least three independent sources.

That rule now has to widen. Numbers need three sources. Names do too.

In the black notebook I still carry from training sessions in Nha Trang, I keep a separate section for names filed in the wrong place. In 2026, I sat taking notes at a training session and was sneered at for praising the split passes of a young midfielder. Three months later he was called up to the national youth team. Some talents are buried under contempt, and I have watched them bloom. Some stories are buried under a wrong label.

The wrong label here buried the most interesting part of the interview: an actor saying he was a fan of the source material before he auditioned, that he never expected to land the part, that he is still filming. Those are classic authenticity beats in a promotional interview — but at least they are real content, spoken by a human being.

Behind that sits a trend more relevant to anyone working in sport. Sports rights are flowing hard into scripted drama. Novel series about athletes are being bought for adaptation, multi-part documentaries about competitions have become flagship products, and fictional characters increasingly carry names that look exactly like real players. Data gave me the number, but the empty stand gave me the question. The question here is: as more fictional athletes enter the same vocabulary as real ones, which system will tell them apart?

Where I could be wrong

The label might have been applied by a human editor rather than a machine. The error might be harmless because a downstream filter already dropped the piece. I might be inflating the importance of a single false node in an enormous graph.

The worry is not the false node itself. It is that nobody found it until a human sat down and read manually. And a system only gets fixed when someone is willing to sit down and read manually. If this error rate is low enough that nobody bothers to check, we are running a machine that is more confident than it deserves to be.

What is worth waiting for

A verifiable prediction: before Heated Rivalry airs in spring 2027, at least one more sports fiction title will be filed into the real sports news stream.

One small proposal: a “football” label should only stick when it comes with at least one real-world attribute — a club, a competition, or a match date. With none of the three, it is a name claiming to be football.

People call me a contrarian. I call myself someone who finds things. This time I found an article that does not belong to football, and an entire system that believed it did.