Mislabeling an Energy File as Football: The Data Gap Vietnamese Analytics Has Not Faced Head-On

7:00 a.m., a cafe on Tran Phu Street, Nha Trang. I open the data package my automated collector pushed overnight: more than two thousand records in twenty-four hours, mostly match reports, possession metrics, injury lists, and a few unverified transfer items. The record at the top carries the tag “Football.”

I click it. There is no team inside. No player, no formation, no scoreline, no coach. The content is the minutes of a Pakistani government meeting on the technical audit of electricity distribution companies, co-chaired by Federal Minister for Economic Affairs Ahad Cheema and Federal Minister for Power Division Awais Ahmad Khan Leghari. The file breaks down into twenty-one information points. Points related to football: none.

I sit still for a while, long enough for the coffee to go cold. Twenty-five years ago, a mislabeled record cost me ten minutes of rereading and nothing else. Now a mislabeled record can flow straight into a forecast model, into an internal index sheet, into an analysis a head coach reads before kickoff. Tactics do not save a team, but they tell you where you died — and the team in this story died somewhere with nothing to do with football.

I log every phase of play as a witness, not as a fan. A wrong label is a phase of play too. It goes into the notebook.

What the data pipeline actually runs on

A football analytics room in Vietnam, even a well-regarded one, usually has three to five people. Nobody has enough staff to read every record. So the work splits into two tiers: a machine tier that collects and applies provisional labels, and a human tier that re-checks the labels that matter.

The machine tier works off a dictionary. That dictionary was written by people, and it mirrors how people think about the world. When a headline contains the phrase “technical audit,” the system tokenizes it, matches it against a topic catalogue, and picks the topic with the highest score. In that catalogue, “technical” maps to the Vietnamese “ky thuat,” a word buried deep in football vocabulary: individual technique, tactical technique, technical coaching. “Audit” maps to “kiem toan,” a word that appears regularly in club finance reporting. “Companies” maps to “cong ty,” and most professional clubs are registered as companies.

Three keywords, three bridges pointing toward football. The result: an energy-sector document filed on the sports shelf.

Volume makes error close to inevitable. One V.League weekend generates hundreds of records: match reports, statistics, commentary, transfer items, press-conference transcripts. One European matchday generates tens of thousands. Most Vietnamese sports outlets take international wire copy, translate it, and push it into internal systems. Every translation builds another bridge between two languages, and every bridge is a place where something can collapse.

In a four-person room, if each person carefully reads twenty records a day, the room handles eighty. The rest goes straight into the store with a machine-applied label. The manual review rate at many outlets sits below one percent. That is my estimate after years in such rooms, not a published figure.

What worries me is not the error itself. Isolated label errors appear in every system, including expensive ones. What worries me is that downstream models assume upstream labels are correct. A tactical forecast model only reads records tagged tactical. A club finance model only reads records tagged financial. When an energy-sector record gets tagged financial — because it discusses circular debt and losses — it quietly enters the training set of a football finance model.

Nobody notices, because nobody rereads. The label itself becomes the evidence.

Dissecting a file that belongs to the power sector

I rebuilt the file against the same eight-dimension framework I use to read a match: tactical analysis, club finance and the transfer market, results and the opinion cycle, league landscape and team positioning, rules and compliance, management and the dressing room, risk profile, and media narrative.

All eight dimensions came back empty.

The tactical dimension has no system, no formation, no playing style. The finance dimension has no broadcast revenue, no wage bill, no transfer valuation. The results dimension has no table, no form line. The league dimension has no league. The rules dimension does not touch a single clause of football law. The management dimension has no owner, no sporting director, no coach. The risk dimension has no injuries, no suspensions, no fixtures. The media dimension has no transfer rumor worth verifying.

What the file actually contains belongs to an entirely different field: a technical audit of all distribution companies, tracing technical and commercial losses, addressing circular debt, and curbing power theft. Two utilities are named for solarisation pilots, Pesco and Qesco. The delivery mechanism involves an expression of interest, independent expert engagement, and terms of reference expected within the following week. The two named individuals, Ahad Cheema and Awais Ahmad Khan Leghari, are federal ministers co-chairing a government committee.

Inside my framework, both go into the “not applicable” cell. I wrote the words “not applicable” twenty-one times. That is not evasion. That is discipline. When I started writing analysis, I thought a good piece was one that filled every cell. Later I understood that a good piece is one that knows which cells must stay empty.

Keeping cells empty matters more than people think. At the point of use, an editing error and a fabricated number look identical. A reader has no way to tell a carefully computed metric from a guessed one if both sit in a well-formatted table. The only way to hold the line is to refuse to fill the blank.

One detail stopped me longer than anything else: all twenty-one information points carried no source. Even the procedural points — the timing of the meeting, the deadline for the terms of reference — had nothing to cross-check against. That means even the dry factual spine of the file is pending verification. Such a file can serve as a process sample. It cannot serve as a basis.

Three layers of error inside one label

Peel a wrong label apart and three layers of error sit stacked on top of each other.

The first layer is the labeling error at source. It happens in the dictionary tier, where people define what counts as football. This layer is the easiest to fix, because it surfaces the moment someone opens the record and reads it.

The second layer is propagation error. It happens in the model tier, where metrics are generated from a contaminated dataset. This layer is far harder to fix, because it never surfaces anywhere. One bad record does not break a model. It nudges parameters slightly, just enough to tilt predictions at the edges — where a single goal decides survival — toward the wrong side.

The third layer is belief error. It happens at the end-user tier, when a coach or a journalist opens an index sheet and assumes the numbers were checked. This layer cannot be fixed with engineering. It can only be fixed with professional culture.

Stacked together, those three layers produce what I call an accountability asymmetry. The person who applies the label does not bear the consequence. The person who bears the consequence never sees the label.

The price sits at the edges, not in the middle of the table

A mislabeled record rarely collapses an index sheet. It eats the edges.

Picture a club hunting a striker for a relegation fight. Their analysis sheet pools data from several leagues and several different recording systems. If a small share of records are tagged with the wrong competition or the wrong season, the target striker's per-ninety numbers shift slightly. That shift is not enough to drop him off the list. It is only enough to place him above another player he should have ranked below.

In the middle of the table, error makes no difference. At the edges — choosing the third name instead of the fourth, one point keeping a club up, one goal changing a season — error is the decision.

This is why I tell young coaches never to read an index sheet top to bottom. Read it from both edges inward. The middle rows are confirmed by too much data to be wrong. The edge rows are confirmed by a handful of matches, and that is where a wrong label survives longest.

Circular debt and a club's debt

One concept in the energy file made me pause: circular debt. It describes a chain of obligations that feeds itself — generators go unpaid, distributors under-collect, the state budget absorbs the shortfall, and the loop continues. The slower the response, the larger the loop.

Professional football has a near-identical structure. A club overspends, falls behind on player wages, on suppliers, on tax, then sells academy players to service old debt. The next season repeats it. What both structures share is that they never collapse in a single day. They collapse when one small link is mislabeled and nobody checks.

“Clear and obvious error” is also a label

There is one place in football where the labeling problem becomes most visible, and I have followed it for years: VAR.

People imagine VAR as an objective machine, where the machine draws the line and the machine concludes. It is not. VAR runs on two stacked labeling tiers. The first is the on-field referee, who must label a phase of play at real speed: foul, or no foul; handball, or chest. The second is the VAR team, who must label that decision: clear and obvious error, or not clear and obvious error.

The phrase “clear and obvious” sounds like a technical condition, but it is an ambiguous clause. It sets no threshold. It sets no viewing angle. It only says the error must be clear — and clarity depends on who is looking.

The space for subjective judgment inside VAR is larger than people think. I regard that as correct rather than as something to eliminate. A system that declares itself entirely objective will never be questioned, and a system that cannot be questioned is more dangerous than a system that is wrong.

What stands out is that this two-tier structure is identical to the structure of the data pipeline I just dissected. The referee is the provisional labeling tier. The VAR team is the review tier. And the crowd is the end-user tier, the one that assumes the label was right. Three layers of error, three layers of responsibility, one label.

Summer 2026 and the toughest editor I have had

In early 2026, a Vietnamese football outlet discovered my geometry file through a former colleague and invited me to write analysis during the World Cup in Russia. Russia beat Egypt 3–1, and I wrote about how Russia operated a five-four-one but pressed in six-second bursts in midfield. I pointed out that Mohamed Salah was isolated, touching the ball only four times inside the opposition penalty area across ninety minutes. The piece was shared more than two thousand times on a major football page.

The editor called to praise it, then left me a line I still remember: “You are too accurate, but readers need to see a face, not just lines.”

I understood I had to learn storytelling. I also took something else from that match: four touches in the box only means something when paired with the right question. Four touches is the fingerprint of a striker whose supply line has been cut. If someone labels that data point “ineffective striker,” they reach a completely wrong conclusion from a completely correct fact.

The error is not in the counting. It is in the interpretive label.

Lessons from the empty-stadium summer

In March 2026, world football stopped. I was fifty-three, doing tactical research for a first-division club. Stadiums opened with no spectators.

The concept of home advantage I had used for twenty years suddenly meant nothing. The crowd, the noise, the psychological pressure on referees, the habit of attacking toward the packed stand — all of it vanished at once. I spent six months reviewing one hundred twenty European matches played in empty stadiums, measuring the average distance between centre-backs and goalkeeper when the home side trailed. The result showed home teams pushing their line roughly eighteen percent higher than usual, generating more dangerous counterattacks for opponents.

The empty-stadium summer taught me that applause is only a coat of paint. It taught me a second thing, less often said: old data does not merely lose value when context changes, it becomes toxic when used without relabeling. A home-advantage metric computed in 2026 is still arithmetically correct but semantically wrong. It carries an expired label.

The energy record on my desk this morning sits on the opposite side of the same problem. It is new, it is live, its timing is beyond reproach. Its error is the label.

A tactical diagram is like a landslide map — it shows you where not to stand. A landslide map labeled “beach” walks people straight into the ground.

Based on my experience watching matches

Based on my experience watching matches across many seasons, I have noticed that most referee arguments and most data arguments share a single root: nobody argues about the definition, only about the outcome. Two people can watch the same slow-motion replay, agree it was a collision, and then disagree sharply over whether it was a foul. The disagreement lives in the definition, not the image.

In football, that definition sits in a thick rulebook anyone can consult. In data analysis, that definition sits in the labeling dictionary of the analytics room itself, and almost nobody consults it. That is why I keep the habit of writing an “underlying assumptions” section at the end of every analysis. Readers have a right to know which assumption I stood on when I stated my conclusion.

My rule is simple: verify first, conclude second. Every claim comes after a chain of re-checking. If the chain is unfinished, I write an empty cell and wait.

The blind spot: people fear missing data, not mislabeled data

A colleague's first reaction when I tell this story is usually to blame the algorithm. I disagree, and I think that blame hides the real blind spot.

The algorithm did not invent the dictionary. It copied the dictionary people wrote. When the system tags an energy document as football, it is doing exactly what it was taught: technical, audit, companies — all three belong to football in the mind of whoever wrote the dictionary. The error belongs to people, not to the machine.

The real blind spot in Vietnamese football analytics is that we fear missing data more than we fear mislabeled data. Missing data is a visible fear: it shows up as an empty cell, as a silence in the copy, as the sentence “no figures available yet.” Wrong data is silent. It fills the cell, it makes the piece look complete, and it never speaks up.

In a content environment that runs at match pace, verification is the first thing cut, because it is the only step that produces nothing visible. Nobody gets praised for catching a wrong label. People get praised for publishing fifteen minutes before someone else.

I have sat in windowless meeting rooms where every decision was made under time pressure and nobody could look outside to check the weather. A windowless room has no view, so I write in order to see what I am saying. The same applies to data: unless the labeling process is written down, we will never see where it is wrong.

What I carry into the next matchday

Before every matchday I keep a list of things to do. From today, I add one item at the top: for every figure I am about to use, ask three questions.

Who applied the label to this figure? When did they apply it? And if the label is wrong, who catches it before it reaches a coach?

Those three questions do not slow me down. They save me from having to redo the work.

At fifty-nine, I understand that winning matters less than explaining why you won. A team can win by luck and still believe it was right. An analytics room can be right by coincidence and still believe its process is trustworthy. The only way to tell the two apart is to check the label before checking the result.

Next matchday, when I open the index sheet, I will start at the top. Not with the row of numbers. With the small line of text that names the topic.