Wrong Labels, Contaminated Data: The Limits and Promise of Blockchain Verification in Football Analytics Pipelines

Monday, seven in the morning, Penang. The first file I opened on my laptop was named stage1_output_football_2026-02-09.json. The folder is my football analytics folder. Inside were sixteen information points, and the domain label read, plainly, football. Every line, though, was about wildlife. More than one hundred and seventy species, more than sixty birds, more than thirty mammals, bioluminescence, iridescence, base jumping, billiards, scuba diving. At the end, a date: 19 November 2026.

It was Game Informer's report on Rockstar Games' Grand Theft Auto 6. Sitting in my football pipeline. It even names Michael Kane, Rockstar's vice president of dynamic art, mentions the fictional characters Jason and Lucia, and refers to a fictional region called Leonida. Not a single football word.

My first reaction was laughter. Then I recognised the error. It is the same machine that drops a rumour into a squad-planning model. The only difference is that a rumour at least belongs to football. This did not even belong.

What the pipeline actually does

Between modern clubs, media houses and markets sits an invisible factory. Its first stage is sources: club statements, an agent's call, a reporter's tweet, a federation document, a broadcaster's graphics feed. The second stage extracts information points from those sources. The third attaches a label to each point: football, cricket, entertainment, gaming. The fourth joins information according to the label. The fifth pushes the joined information into models, scouting reports, broadcast telestrators, even prices.

The labelling stage still runs largely on linguistic guesswork. Word overlap, entity presence, the shape of numbers. That is where the trouble starts. "More than 170 species, 60-plus birds, 30-plus mammals" reads exactly like a stat line. Swap species for sprints and nobody would blink. "Legendary" means a legendary striker in football and a rare creature in gaming. "Release date" means a transfer window in football and a launch in gaming. When an automated tagger finds football-shaped numbers and football-shaped verbs but no club, no player, no league, no coach, its embedding vector genuinely loses its bearings. The label stays football; the content becomes a video game.

I have been inside that factory twice. In 2026, an eighteen-year-old student in Penang, I skipped a statistics lecture to watch Penang U19 beat Perak U19 at USM Stadium. Sitting behind the goal, I hand-coded every defensive action for ninety minutes. Penang's 4-4-2 mid-block forced eighteen high turnovers and allowed seven entries into the left half-space. I still have the 2026 notebook; the 4-4-2 mid-block wrote itself in pencil. That day I learned that data is not only numbers. Data is what I counted and what I looked away from while counting.

The second time was 2026, at the Russia World Cup, as a volunteer data-logger. In Moscow I tracked Croatia's 2-1 extra-time win over England. Luka Modric covered 14.3 kilometres, Croatia shifted from 4-1-4-1 to 4-3-3 after the sixtieth minute and pinned England's 3-5-2 wing-backs, and England began losing second balls after the seventieth. The next morning the data team asked me to present my map. A volunteer sees the World Cup from the stairwell; Croatia 2-1 England taught me the noise before the goal. That noise never reaches a pipeline, because it is not an entity. It is a wave.

In 2026 the stadiums emptied. Penang FC beat Kelantan United 2-1 and I was a volunteer analyst at a closed-door match. With the stands silent I recorded ninety-six coach commands and thirty-one pressing cues in the first half alone. Penang's 4-2-3-1 press was triggered by a backward pass to the goalkeeper, not by a loose touch. The assistant coach confirmed it at half-time. The empty stadium taught me that the most valuable information is usually written down nowhere.

Where the error is born, and why it spreads

The file reached me in the middle of a transfer window. Every day brings thousands of claims, denials, counter-claims, medical dates, release-clause structures, wage bills, all floating in the market at once. In that crowd a wrong label goes unnoticed, because nobody returns to the source any more. Everyone decides from the output of the previous layer.

The error is born in three places. First, at source: the same RSS feed carries sport and entertainment together, and nobody draws the border. Second, at extraction: the script pulling the information knows patterns, not football. Third, at verification, and this is the widest gap. There is no entity-based domain validator that says: this document has no club, no player, no league, no coach, so it cannot be football.

Chain of custody and the blockchain question

This is where blockchain enters, carefully. What blockchain genuinely does well is chain of custody: an immutable record of every step from source to decision. Which feed an article entered from, at what moment, in which version, into whose hands as a tagger, which label it received, who approved it. If all of that sat in an append-only log, at least the question could be asked. Right now the question cannot even be asked. The file arrived, settled in, and blended into the model.

Football has natural places for this. Player registration and transfer paperwork, especially sell-on clauses, image rights and solidarity payments, where competing claims could be settled on a shared ledger. Second, provenance for broadcast and data feeds: if you could verify which feed produced a given xG or PPDA figure, both broadcasters and markets would make fewer mistakes. Third, fan tokens and club governance voting, though my confidence there is lowest, because a ledger can record who voted without ever answering who decided the weight of a vote.

Before all of that, one thing needs saying. Blockchain can make a label immutable. It cannot make it correct.

What never makes it onto the chain

Here I should state my hesitation plainly, because it is a trade-off inside my own profession. If a ledger proves that a label was attached at a certain time by a certain person, and the label was wrong, what we have is a flawless, timestamped, cryptographically signed error. Blockchain does not understand meaning. It does not know that "legendary" means a striker in one place and a creature in another. It does not know a game's release date is not a transfer window. Placing meaning is human work, done by someone who knows football.

There is a larger danger. Once something is on-chain, people relax: it is verified. Then nobody returns to the source. Verification becomes a substitute for evidence. Football already suffers from this. When an agent's claim is printed under a big name, nobody asks whose interest produced the claim. Blockchain can harden that laziness by one more notch.

Then there is cost, latency and privacy. Live match data changes in fractions of a second; a public chain cannot match that pace in practice. Injury files, wage details, contract clauses will not be published on an open ledger, and if they are, it will be on the club's own terms. Who holds the keys? The biggest clubs, the ones who can already turn the final twenty minutes into a war of attrition with five substitutions. If they gatekeep the ledger, transparency does not arrive. A new kind of shadow rule does.

And the last point, the one I think about most. The most valuable information is usually unwritten. In the 2026 Penang FC versus Kelantan United match, not a fraction of those ninety-six first-half coach commands existed in any official feed. Nobody counted how many of those seven entries into the left half-space were actually progressive. The person in the stairwell, the volunteer reading a coach's lips from the corner flag, will not be given a node. The football intelligence built in Dhaka's night-street arguments and on Kuala Lumpur's rented pitches has not even been given a label yet. A ledger that records only the truth of the powerful records half the truth and calls the rest false.

Where the next verification begins

So my proposal is simple, and it does not begin with blockchain. It begins with an audit. Take one data feed for one week, your choice: transfer news, post-match statistics, anything. Then trace every record backwards. Who published it, when, which entities it contains, who assigned the label. Count the records with no club, no player, no league, yet labelled football. That number is the true measure of contamination in your pipeline.

My guess is you will not enjoy the number. And if that same number could sit in public, timestamped, with everyone watching who entered a record, who approved it, who objected, at least the boundary between rumour and information would be visible. Drawing a boundary and respecting it are two different jobs, and I know it. Still, that is where it starts.

In the noise of this transfer window the question matters more, not less. Is the story that sounds most certain actually about football, or does it merely sound like football? And the data entering your decisions: is the road back to its source still open, or has it closed at some point, in someone's hands, in a way nobody now remembers?