Entity Recognition Errors in Football Data: When a Player's Name Collides With an Actor's

That night, while auditing the data pipeline my analysis team uses to track the transfer window, an unfamiliar file appeared in the queue. It was tagged "football." The first line mentioned a name any student of World Cup history would recognise: Oscar Bonfiglio. But the context was jarringly wrong — no club, no match, no player, no coach. Only a cast list, a producer, a premiere date and a broadcast slot. A TelevisaUnivision drama airing on Las Estrellas had slipped into our analysis room and been processed as if it were a transfer bulletin. I sat still for a few minutes, hands on the keyboard, because what had just appeared on screen was not a minor glitch. It was a systemic break point.

The story begins with two names. Oscar Bonfiglio, in that file, is an actor in the drama's cast. In football databases, Oscar Bonfiglio is the goalkeeper who played for Mexico at the 2026 World Cup and later became a coach. The same string of characters, two people, two worlds. One case alone would not matter. But the file also carried Christian Ramos — an actor, yet to anyone analysing South American football, Christian Ramos is a Peruvian international centre-back. The system matched the name strings, cross-referenced a football knowledge base, found a hit, and stamped the document "football" without ever reading what followed.

I raise this not to defend the algorithm but because it was the first time I saw an operational failure that I believe is quietly spreading through football, in the least observed layer of all: the data layer. We are used to arguing about decisions on the pitch — a penalty, a red card, an offside line. We rarely argue about whether the data we read actually describes the right people and events. And when the transfer window arrives, that data layer comes under maximum pressure.

The transfer window is the data layer's harshest test

In a transfer window, speed is everything. A name surfaces on social media at 10pm; by 10:15 it must be in a feed; by 11pm it is in a transfer round-up; by morning it has been copied across dozens of sites. No one has time to verify every detail. That is the perfect environment for errors to multiply. Based on my experience following matches and market reports, these errors fall into three groups. The first is source error: an anonymous account posts an unfounded claim that is then cited as if it were grounded. The second is logic error: a rumour that is structurally plausible about a squad but has no financial evidence behind it. The third — and by far the most dangerous — is identity error: a system or a person misidentifies a name, a person, a club.

Identity errors are the hardest to detect because they do not self-report. A false transfer rumour collapses when the window shuts. A mis-assigned name persists in databases for years, seeping into every report, every statistics table, every scouting file, until someone patient enough traces it back to the source. And in an era that treats data as king, the number of patient people keeps shrinking.

The file that night showed how far one identity error can travel. It was not merely wrong about a player. It was wrong about the entire nature of the document. A television drama's promotional copy — written to sell an entertainment product, containing no analysis and no counter-argument — had been classified as a football source.

A name is not a person

What troubled me most was not the error itself but the assumption it exposed, one the whole football industry quietly shares: that a name is a unique identifier. In reality a name is just a string of characters. Its meaning depends entirely on context, on time, on country, on competition.

I saw this while working with referee data. At the 2026 World Cup in Moscow, I spent weeks logging the numbers of 64 matches, and I was pedantic about cross-checking referee names against match codes. A referee can appear in two different matches on the same matchday, and if a table is mis-keyed we get two referees sharing a name officiating two different games, or one referee credited with back-to-back matches. Such mistakes are quiet, but they erode the entire value of a statistics system.

Football has one of the highest name-collision rates in sport. Consider how many players carry the surnames Rodríguez, González, Martínez, Silva, Ramos or Sanchez worldwide. A Peruvian defender named Christian Ramos can exist alongside a Mexican actor of the same name, and nothing stops an automated system from merging them if it relies only on character strings.

The mechanism behind the error is named entity recognition — the automated step that extracts people, organisations and places from raw text and links them to a knowledge base. If the string "Oscar Bonfiglio" matches a football entity and the document lacks source metadata strong enough to contradict it, the system defaults to treating it as football content. It does not read further. It does not ask. It simply matches.

The irony is that this happened precisely when football prides itself on its highest-ever level of technical sophistication. We have semi-automated offside technology cutting the offside error margin from 0.4 metres to 0.1 metres. We have VAR with dozens of camera angles. We have predictive probability models. Yet we still have no operational standard for the most basic data layer of all: how to be sure a name points to the right person.

Anatomy of a break point

Dissecting the file as I would a match report, I found no step in the pipeline designed to catch this kind of error. The first layer is collection. The raw text was ingested with no recorded source — no journalist, no outlet, no clear publication date. In my work as a rules specialist I always begin with three questions: where, when and by whom was this written. Without answers, I treat a document as void. Here all three were empty. A sourceless document had passed the first gate.

The second layer is classification, where the error forms. The classifier has no mechanism to cross-check the topic label against actual content. It does not ask: if this is football, which club, which competition, which players actually appear. A few keyword signals are enough to assign the label.

The third layer is verification, where a human should read it back. In high-speed newsrooms that layer is often cut because it is slow and costly. And once it is cut, no one remains to notice that a television drama is being processed as a football club.

People see the red card; I see the clause written in haste. Here the hastily written clause was a classification process with no counter-argument step. And as in the laws of football, a procedural gap does not vanish — it waits to be prised open.

A pandemic does not create loopholes in the law — it only knocks on the gaps that already exist. I learned that in 2026, compiling a 15-page handbook on contract clauses affected by lockdown orders, after a dispute between a V.League club and a striker over termination for force majeure. Every crisis finds the gaps that were already there and amplifies them. This data incident was the same. The transfer window did not create identity errors. It turned a silent gap into a visible incident.

The historical trap of repeated names

One detail made me pause longer than anything else: the interfering names were not modern names but names written into football history. Oscar Bonfiglio is a real historical figure — part of Mexico's 2026 World Cup squad, the first World Cup ever played, later a coach. To a football database, a name tied to 2026 carries high historical value and is therefore carefully stored and tightly linked to other historical entities. But precisely because it is carefully stored, it becomes an attractive match point for a classifier: a rare, easily recognised, easily confirmed string.

This is a paradox I see repeatedly in analysis. The rarer the name, the more dangerous it is, because the system grows more confident when it matches. With a common name like Rodríguez, a classifier must be more cautious because the collision probability is high. With a rare name like Bonfiglio, it has almost no reason to doubt. Rarity of a name is inversely proportional to the level of checking.

I saw the same thing in officiating. At a V.League match at Lạch Tray stadium in 2026, a sending-off in the 68th minute sparked fierce controversy. I wrote my analysis that night, quoting Law 12 of the IFAB Laws of the Game verbatim, and showed the challenge lacked the danger level required for a direct red. What made the decision so damaging was not the rarity of the situation but the referee's certainty that he was right. Confidence in a rare situation is accuracy's greatest enemy. The classifier suffers the same disease. It errs not from a lack of data but from being too certain when the data matches.

When a system misreads a name, it misreads history

Let me be explicit about the consequences, because this is the part the industry underrates. An identity error does not affect one article only. It affects the entire data value chain behind it.

Imagine an automated scouting system. It scans thousands of articles, extracts player names, links them to performance profiles. If one name is mis-assigned, the correct player's profile can be blended with another's data. A club reads that profile and decides wrongly. A scout flies halfway around the world to watch a player who is not in the squad he is pursuing. The damage is not in the article. The damage is in the decision.

In a transfer window, when the wage bill becomes a survival line and when contract structure — release clauses, sell-on clauses, payment schedules — matters no less than the player himself, an identity error can push a club into a commitment to the wrong person, at the wrong price, at the wrong time. I tell colleagues that the real story of a transfer window lies in contract structure and wage bill, not in loud headlines. But to read that structure, we must first trust that the name being discussed is the right person.

This is why I regard that night's incident as more serious than a technical bug. It shows football's data layer operating without any standard for identification. We have standardised a great deal on the pitch — the Laws updated annually, rules on technical-area occupants, added-time conventions, VAR protocols. Behind it, at the data layer, we have almost nothing.

The counter-intuitive angle: data is not objective

A widespread belief holds that data is objective and that once data exists, argument must end. I consider this one of the most dangerous illusions of our time. Data is not objective. It is objective only at the last point of a chain of subjective decisions. Every step — which source, which name to extract, which entity to link, which topic to label — is a choice that can be wrong. When the chain is hidden behind a tidy interface, the reader feels they are touching truth. But that feeling is built from layers of assumption, each of which can break.

I have argued this for years as a rules specialist. In football we never say a decision is right merely because a referee made it. We check it against the clause, the evidence, the context. So why do we accept a number merely because a system output it? A number from an unverified pipeline deserves cross-examination as much as a referee's decision.

What worries me more is that this incident involved a document with no source, no author, no outlet — the general pattern of much material flowing into football data pipelines every day. If a television drama's promo copy can slip into a football analysis room, any noise can. The problem is no longer one stray file. It is an enormous data resource being used without any verification standard.

I recall my frustration at Euro 2026, when I wrote a series on VAR errors and showed the VAR error rate was around 1.8 times higher than at the 2026 World Cup. What annoyed me was not the number but the public reaction: many accepted that technology can err, yet few demanded a process to stop it erring again. We react to each error, forget it, and wait for the next. Fixing one law takes ten minutes; admitting the law was wrong takes ten years. With data it is longer still, because we have not even admitted we have a set of data laws to fix.

An operating procedure: an evidence filter for the data layer

As someone who trusts structure over inspiration, I do not want to end on a mere warning. I want to propose a procedure that can be used immediately. Rule one: a document without a clear source is not a document. Before ingesting anything, I require three minimum facts — publisher, author or responsible unit, and publication date. Missing any one, I discard it.

Rule two: a topic label must be verified by content, not keywords. If a system labels a document "football," it must point to at least one club, competition or match. If it cannot, the label is void. Rule three: a name must be linked to at least two independent attributes. For a player, nationality and year of birth, or current club and position. A single name string is never enough to assert identity. If Oscar Bonfiglio is linked by name alone, we put two entirely different people into one entity. Rule four: every unverified document must be clearly flagged, so the reader knows what has been verified and what awaits cross-checking.

My checklist has four questions, answered in order: where does this source come from; does the content match the topic label; has every name been linked to at least two independent attributes; and if the first three are answered yes, will this information change any real decision. Four questions take under two minutes but prevent incidents that take months to repair.

Who is responsible when a name is mis-assigned

Before writing anything I always ask: who will implement this, and what will they do. The first implementers are data operations teams, who need a set of standards before the next transfer window opens. The second are journalists and editors, who need training to recognise the signs of an unverified source, not merely to write faster. The third are federations, which hold the power to set system-wide standards. And the final, most important implementers are us — the readers.

When a report says a club is chasing a player, we must ask who that player is, not only how much the fee is. When a statistics table appears, we must ask where it was built. This is not negative scepticism. It is respect for the complexity of information. That night's incident changed no match, cost no team points, took no goal from anyone. On football's emotional scale it is almost meaningless. On a system's scale, it is an alarm bell.

The lesson is not in the data file

I have spent years arguing about controversial refereeing decisions, and I have found that those arguments are driven more by emotion than by law. Fans think about their club. They think about what they want to believe. This incident is the same, on another layer. We want to believe data is right, because the world becomes simpler when there is a source of truth that never argues back.

But a name is not a person. A topic label is not a nature. A number is not a fact. All of them must be cross-examined, queried, placed in the light of context. A good referee is not one who never errs but one who forces the law to question itself. I would add a similar line for those working with football data: a good system is not one that never fails but one that forces its operators to question themselves. That night forced me to question years of my own work. It showed that while I spent all my energy cross-examining decisions on the pitch, I had come to trust the data layer behind them almost absolutely — a layer where no one checks, no one cross-examines, no one shows a card.

What is this if not a break point in the whole data law we have never written. Which is why I believe entity-identification standards will become a mandatory part of professional football management within a few years, much as VAR protocols became mandatory in major competitions. Clubs will stop asking only whether they have data and start asking whether they can trust it. And when that question becomes standard, we will look back on the night of the mislabelled file as a small but necessary milestone.

Football always learns from mistakes — usually mistakes broadcast live to millions. This one happened in silence, in a data queue, between two identical names. If we do not talk about it, it will recur. And one day it will not stop at a television drama slipping into an analysis room — it will affect a contract, a squad place, the career of a real person.

A shared name is not a tragedy. Our lack of tools to tell them apart is the problem. And that problem is not in the algorithm. It is in our habit of trusting without cross-examining. I learned this after years of working with laws and footage, and I believe it holds for the numbers flowing across the data layer's screens. The progressive conclusion is not to abandon data but to treat it as we treat referees: with respect paired with cross-examination, with trust paired with process, always leaving open the possibility that we are wrong. Football has never been perfect on the pitch, and it will never be perfect in its data. But if each pipeline learns to question itself, the gap between a name and a person becomes a checkpoint rather than a trap.