An Empty Cell Is More Dangerous Than a Wrong Number

In June 2026, a data file from a partner landed on my machine just before midnight. It had a league name, a "football" label, and all the right columns: date, matchday, team, player, minute, shot coordinates. I opened it and every cell was empty. Not one event row. Not one name. Not one scoreline. The only thing alive in that file was the classification label — the word "football," stuck there like a signboard hanging in front of a room already emptied of furniture.

I called the sender. The answer was short: the system completed the classification step, but the extraction step returned nothing, and nobody had placed a checkpoint in between. The empty file went through. The summary was still drafted. The headline was still written.

Three months later I was still thinking about that file. In this trade, the most dangerous thing has never been a wrong number. A wrong number can be argued with, corrected, pinned on someone. An empty cell treated as a number cannot be argued with, because nobody knows where to begin.

A match passes through four silent layers

To understand how an empty file gets through, you have to look at how football data is produced. A single V.League match passes through at least four layers: the event logger at the stadium, the organiser's data entry system, an intermediary data provider, and finally the analyst's model. Each layer can fail silently. No siren. No red light. Only empty cells drifting downward, and at the bottom, someone has to decide: write "missing data," or write "0."

In 2026, when I built my own xG model for 14 V.League clubs, I worked with exactly that risk. The first xG table I ever wrote was by hand on a coach bus, back when nobody called it data. I sat there logging shot by shot, and the first lesson was not the xG formula. The first lesson was learning to recognise a match with too little data to compute. That match had to be flagged "insufficient sample," never assigned a default value just to make the spreadsheet look full.

That same year my model produced a result many people called reckless: Phan Van Duc, then 20 years old, carried an xG per match of 0.48 — higher than the average for foreign strikers in the league — despite scoring only five goals. I wrote that he would become a national team mainstay within three years. People accused me of being deluded by numbers. But the more important part of that story is that I nearly had no prediction to publish at all: of the 14 clubs, two were missing shot-location data for almost half a season. Had I filled those gaps with guesswork, the model would still have run, the table would still have looked neat, and the conclusion would still have been printed. It just would have been wrong, and nobody would have caught it.

Four failure modes, one common signature

Looking back at that empty file from 2026, I can sort the failures into four types. First, the source document never reached the extraction unit — the system ran, but it ran on nothing. Second, the document arrived but was empty, or was an image, or was video without a transcript, or had a broken encoding. Third, extraction returned a malformed result and the layer below silently coerced it into a default. Fourth — and the most alarming — the system returned its own instruction template instead of data. It printed its own homework.

The common signature across all four: the classification label survived and everything else vanished. A record left holding a single category tag is a record that never existed, yet it looks exactly like a completed record. On the pitch it is the equivalent of a match report with a competition name and a matchday printed on it, but no scoreline, no line-ups, no cards. Nobody would publish a report like that. But inside a data pipeline, it passes through every day.

I have seen this exact mechanism in a VAR room. An offside call gets reviewed frame by frame, the picture gets cropped, the lines get drawn, and in the end the decision still provokes argument — more argument, in fact, than before VAR existed. To me, the problem is not the technology. It is the inevitable result of moving an argument off the pitch and into another room where the law still has grey zones. Data behaves the same way. When an empty cell is filled with a guess, the argument does not disappear. It just transfers to whoever did the filling.

Contamination travels downstream

The biggest risk of an empty file is not the file itself. It is that the file gets passed on. A blank record that clears one more layer becomes a valid input. One layer further and it becomes a fact. At the final layer it becomes a confident conclusion.

During a transfer window, this mechanism runs many times faster. A rumour with no source gets reposted by three accounts; by the fourth it has "a source." By the tenth it has become a specific number: the fee, the contract length, the wage. Nobody in that chain lied. Each person simply filled one empty cell, because leaving it empty means the piece does not run.

The irony is that the market rewards the filler, not the person who leaves the cell blank. The instinct of this trade is to fill a gap with a plausible story, and that is precisely the moment when data is sold cheapest. Elsewhere, a loan with an obligation to buy operates on the same logic: a payment that has not happened is booked as though it had, so that a small club's financial plan looks fuller than reality for one season.

Absence of evidence is not evidence of absence

Here is the counter-intuitive part. When a record comes back empty, our reflex is to conclude that nothing happened. Wrong. An empty record says only that the recording system failed. It says nothing about the match.

That is why I always devote a paragraph in every analysis to stating my sample size, my confidence interval, and the variables my model cannot measure. Based on my experience watching matches, the crowd watches the move; I watch 22 numbers in motion — and wait patiently for them to tell a different story. But I also know that when the count is too small, the best number to state is no number at all.

In 2026 the stands were empty, yet every pass still fell into a cell in my model, and I understood that data never keeps company with a pandemic. With six months and no matches, I dug back through the V.League archive from 2026 to 2026 and found a pattern: clubs that changed president mid-season saw their win rate fall by 23 percent over the next five matches. After that retrospective series ran, a club executive called to thank me for helping them avoid a sacking decision at the wrong moment. There was nothing miraculous in it. It was just a sample large enough to be left alone for ten years, instead of being filled in by feeling.

The checkpoint, and the most valuable cell

The technical lesson is short: every data pipeline needs a hard checkpoint between layers. If a core field is empty, the system must stop and raise an error, not carry on. Better a process that breaks loudly than one that runs quietly and hollow.

The professional lesson is longer. My model does not cry and does not celebrate, but after every match it owes me a lesson. And the biggest lesson from that June 2026 file is this: the most valuable cell in a table is not a full one. It is an empty one, honestly labelled. Readers will forgive a model that says "I don't know yet." They will not forgive a model that speaks with certainty while holding nothing.

The next chapter of this story will not be written by a new algorithm. It will be written by whoever is willing to be the first to type the words "insufficient data" onto a page that needs traffic.