The Blank Analysis File and the Discipline of Verification in Football Data

The night before a matchday, I opened the positional data file the system sent over. The file was empty. No player names, no scoreline, no passing metric or distance covered. Only a template frame waiting to be filled — a sign any analyst recognises at once: the extraction failed before it ever reached the source data.

The first reflex of a newcomer is to fill the gap. I have seen reports padded with guesswork, with memory of last week's match, with the feeling that this team still plays the same way. But after eighteen years of note-taking, I learned something that sounds paradoxical: an analyst's work does not begin by producing answers; it begins by refusing to answer when the data has not arrived. A blank file is not a finding. It is a silence that must be recorded exactly as what it is.

In modern football, data has become a hidden structural layer. Every match in the top leagues generates millions of data points: player positions measured by GPS or optical systems, pass counts, pressures, expected goals (xG), and dozens of derived metrics. The suppliers are many — StatsBomb, Opta, Hawk-Eye, alongside each club's in-house systems. Each has its own definition for each metric. One provider's successful tackle may not count as successful for another, simply because the ball-contact threshold differs.

That creates a familiar trap: the more numbers there are, the easier it is to believe the truth lies inside them, waiting to be read out. But data does not generate itself. It is produced by cameras, by tagging algorithms, by people re-checking it. Numbers whisper, and whoever listens will hear an entire match — but only if that person knows where the number was born, how, and by whom.

I learned my first lesson about this from an article that was mocked. In 2026, when I was twenty-five, I did data analysis for a new Australian football site. The A-League was at round twelve. I published a piece of more than three thousand words on Melbourne City's pressing metrics, using GPS positional data to show that manager Warren Joyce's side was pressing in the wrong direction. The consequence was that midfielder Luke Brattan had to run an average of 11.2 km per match yet produced only 1.3 successful tackles. Fans called my piece dry, reading like a talking spreadsheet. Three weeks later, Joyce changed the pressing structure. Melbourne City won four straight matches.

I do not tell that story to praise myself. I tell it because it taught me that good data does not need florid prose to be right — but it is only right when the data itself is right. If the GPS file that day had a sync fault, if player coordinates were mis-tagged in the first ten minutes, the conclusion about pressing direction would collapse — and a manager might have reshaped his team because of a broken number.

In 2026, I wrote an English-language piece predicting Croatia would reach the World Cup semi-finals, based on Luka Modric's chance-creation xG in the group stage. A group of amateur coaches on an international forum called me a bookworm who did not understand football. Croatia reached the final. After the tournament, a journalist from The Athletic got in touch to ask how I calculated defenders' defensive xG prevented. I spent two weeks writing Python code, cross-checking with StatsBomb data, then sent back a seventeen-page analysis. At World Cup 2026 they laughed at my xG. Today, when xG has become everyday television language, the question I receive most is still: where does this number come from.

And in 2026, when football returned to empty stands, I was the one who was wrong. I was running a match-result prediction model in which home advantage was priced at 0.45 goals per match. After nine matchdays of the Bundesliga without spectators, that figure fell to 0.08. When the stands fell silent, home advantage faded with them. A magazine asked me to write an explainer on football without crowds, and I refused. I needed three more weeks of data to be sure the nine-matchday sample was not random noise. When I finally published, I opened with an admission that my model had missed a variable.

Those three stories share one structure. An anomalous number appears. The next question is not what the number says, but how it was made. Before trusting a number, ask where it came from. With GPS data, the question is sampling frequency and sync error. With xG, the question is which probability model, trained on which dataset, whether it accounts for goalkeeper position and defender pressure. With defensive metrics, the question is who decided a duel was successful.

My job in Sydney is sports data analysis for the Australian market. There, a bad analysis can spread fast, because it enters bulletins, television commentary, fan decisions and sometimes club planning. Mispricing one variable is like losing your bearings for a whole year. So in every piece I state the data version, the supplier and the retrieval date. That is not bureaucracy. It is how readers can verify for themselves, rather than having to trust the writer's reputation.

There is a counter-intuitive point I want to state plainly. When an analysis comes back blank, the correct response is not to reason your way into filling it, but to suspend judgment. The silence of data does not equal the absence of risk. In sports analysis this is the most dangerous and least visible error: turning a data-pipeline failure into a nothing-to-report conclusion. If I received a blank file before a match and wrote that the two teams had no issues, I would commit two errors at once — inventing a conclusion and denying the possibility that my own tool was broken.

The same family of errors appears elsewhere on the pitch. An offside line drawn by semi-automated technology, measured to the millimetre, can be geometrically correct yet stifle a team's attacking instinct — when a striker is forced to wait for a signal instead of running on reflex. A goalkeeper's reflex metric can look fine on a stats sheet yet say nothing about distribution under pressure. Every measuring tool carries a hidden assumption, and that assumption usually goes unrecorded in the report.

Correlation is not causation. It sounds old, yet in football data it is violated every week. Winning teams usually have high xG, but high xG does not always lead to a win. Teams that press a lot usually create chances, until opponents learn to play long through the pressing line. Transfer value is a story, but data is the signature. And a signature can be forged if the writer does not read the source carefully.

That is why I add to every piece a section colleagues once saw as self-deprecating: Assumptions That May Be Wrong. There I state what my data lacks, how small the sample is, and which conclusions would reverse if the underlying assumption failed. For a rigorous reader, this creates a sense of being respected rather than manipulated by absolute numbers. A season missing detail is like a match missing stoppage time — people still know the result, but not how it was produced.

This is how football operates, if the analyst is patient enough to read it. Football runs on successive chains of causation, and an analyst can only see part of that chain within a finite data frame. Caution in conclusions is not intellectual weakness. It is the inevitable consequence of understanding that every model is a map, and a map is never the territory.

The night before that matchday, when the data file came back blank, I chose not to write. I re-sent the extraction request to the supplier, and waited. The next morning, a second file arrived, complete, and the story of the match emerged as clear as it actually was. Those twelve silent hours produced no article, but they produced something more important: a conclusion worth trusting.

In the major-tournament cycle ahead, when every number will be thrown onto screens alongside the pulse of the crowd, what is worth watching is not which metric looks best. What is worth watching is which data supplier is failing, which model is going stale, and which analysis is quietly being filled in with guesswork. Keep an eye on the blank files. They often say far more than the full ones.