When the Data Table Returns Zero

In a Seoul sports broadcast control room, the match-stats monitor went blank in the 63rd minute. No xG. No progressive passes. No distance covered. The technician said the feed from the data provider was congested and would return in fifteen minutes. The lead commentator nodded, turned to the audience and said something I recorded verbatim: "Well, nothing worth mentioning has happened in the second half anyway." I was sitting in the third row, my spreadsheet already open, and I realised what had just happened was more dangerous than a technical fault. A region of data vanished, and it was instantly replaced by a judgement — that nothing was there. Nobody in the room checked. Nobody asked whether that silence was real silence, or a signal that had been cut.

That is why I am writing this. Not about a specific match, but about the reading habit I consider the most dangerous error in sports analysis: turning missing data into safety.

I have been reading sports data for more than fifteen years, from the newsroom of a small channel to deep analysis desks covering esports for the Korean market. My first principle came out of a failure. That mistake taught me that data never lies; only the reading of it is wrong.

In 2026, aged thirty, I was a mid-level staffer at a new sports channel. For Korea versus Iran in the World Cup qualifiers I was assigned the pre-match analysis. Using xG and progressive passes, I argued the national team should play possession football rather than counter-attacking. The coach kept a 5-4-1, the match ended 0-0, and Korea needed luck on the final matchday to qualify. The next day a male colleague said my piece was "a woman who does not understand football, clinging to numbers". I did not argue. I went home and pulled all thirty-eight qualifying matches from five confederations to re-analyse.

What I found was not in any number. It was that I had treated a dataset of three recent matches as sufficient to conclude something about a national team. That dataset was not wrong. It was incomplete. And I had read the incompleteness as sufficiency. Since then every analysis I write carries a section my old editor hated most: notes on error margins and data coverage.

My method now is multi-layer cross-verification. A conclusion may only exist when there are at least two independent original data sources, one verification layer from someone on the ground, and a list of the things I do not know. That last list is the most important part. It is a reminder that an empty data table does not mean "nothing happened". It means "we have not seen it yet".

One habit I have kept for years: every report of mine carries a line that reads "insufficient information to assess". That line is not cover for laziness. It is a declaration that I know my limits, and that a wrong conclusion is worse than an admitted gap. Readers may skip it, but editors do not. They start asking me better questions: how many matches does this data cover, who collected it, and what was left out.

An empty field is data; an empty field read as zero is fabrication.

In statistics this is called the false-negative trap. You find no evidence of risk, so you conclude there is no risk. But "not found" and "not existing" are different stories, and the distance between them is where every analytical error is born. I have seen this trap operate on three levels.

The first is technical. In 2026, when COVID-19 suspended the K-League indefinitely, I worked remotely and analysed FC Seoul's first ten matches to predict which teams would survive relegation. I found the squad averaged only 98.7 km per match, third-lowest in the league, with a rising rate of tactical fouls in their own half. I wrote a critique of the coach's tactics. The desk refused to publish, saying it was a sensitive moment and criticism was inappropriate. I kept the piece and dug deeper into five seasons of player fitness data.

When I reopened that table six months later, I saw something else. The matches where I had complete fitness data were precisely the home matches, where the league's tracking system ran reliably. The away matches, where the team performed noticeably worse, had gaps in the running data. I had unknowingly compared two samples with different coverage and called it a conclusion. Between the transfer numbers lies a story nobody writes in the report — and here, between the fitness numbers lay a gap the system never recorded.

The second is human. In 2026, at the World Cup in Russia, I held an official press pass. After Korea lost 0-1 to Sweden, I went to the mixed zone and struck up a conversation with a Belgian agent. He described a young Senegalese player in the Belgian second tier he had watched with his own eyes for two years. I checked the player's data: top speed 34.2 km/h, 61% successful dribbles, but very poor pressing numbers, with only 18 touches in the final third per match. I told him flatly that the weakness was counter-pressing. The agent was stunned, because I had never watched the player once.

What I did not tell him was that in the table I used, the pressing column had forty per cent of its values left blank. The provider only logged pressing actions during continuous live play; dead-ball situations and fast counters were skipped. That Senegalese player was in a counter-attacking side, where most of his pressing actions occurred precisely in the situations the system ignored. What I called "poor pressing" was partly the provider's error, not his feet.

I tell this not to flagellate myself. I tell it to show that even when I win an argument with data, I may be winning with half the truth. I do not trust intuition; I trust numbers that talk after being asked the right questions — but the right questions include questions about what the numbers do not say.

The third is systemic, and here esports faces a bigger problem than football. A football match lasts about ninety minutes with dozens of fixed cameras. An esports match generates thousands of events per minute, produced by the game server itself and relayed through at least four layers: the publisher's API, third-party data providers, broadcast platforms, and the tournament dashboard. Each layer can go silent in its own way. An API can return an empty array without raising an error. A provider can interpolate missing values with league averages, making every metric look smoother than reality. A broadcast platform can display stale numbers for minutes because of buffering.

Esports does not need luck; it needs people who read the meta faster than the server itself. And when the server goes quiet, the reader must be the first to notice the silence, instead of filling it with a story.

This is where I want to be blunt about a habit in analysis circles. When data is full, we are careful: we ask about sample size, about season, about boundary conditions. When data is empty, we relax instantly. Nothing to challenge. Nothing to doubt. And we write lines like "the team has no discipline problem" merely because the card-tracking table has not updated. That is the moment analysis stops working and starts decorating.

In 2026 I followed Leicester City as they sat second from bottom in the Premier League. My model flagged an anomaly: Leicester's xG was higher than predicted, yet actual goals conceded far exceeded expected goals conceded, a gap of 7.8 goals after just fourteen rounds. The cause was not luck but individual errors in defence; centre-back Wout Faes made mistakes leading to goals in three consecutive matches. I wrote that manager Brendan Rodgers needed to switch to a back three to cover for pace. A European football site republished it. Three weeks later Rodgers was sacked, and Leicester did switch to a back three under Dean Smith — but still went down.

Notably, the data on individual errors in that piece did not come from the stats provider. It came from me rewatching the footage. The provider's table gave Faes an average tackle success rate, a number that looked entirely normal, because positioning errors were not coded as a separate column. Had I only read the table, I would have seen nothing. The betting market is not wrong; it merely reflects a truth you have not yet seen — and conversely, a clean-looking data table may be hiding a truth it was never designed to see.

In 2026 I scanned data from forty-nine European domestic leagues looking for centre-backs for Korean clubs, and stumbled on Isak Hien, then twenty-four, a Swedish defender of Ethiopian descent at Hellas Verona. Hien posted 2.9 successful tackles per match, but what stopped me was that in over two-thirds of his matches he completed forward passes into the next line, a marker of build-up ability. I wrote a deep comparison with Virgil van Dijk at the same age. When I suggested national-team scouts look at him, they declined for lack of direct sourcing. Four months later Atalanta signed Hien, and he became a pillar of their 2026 Europa League title.

I tell this for a different reason than before. Here my data was right, but it lacked a layer: the credibility of someone who had watched matches in person. Open data can point to a player, but it cannot replace the eye that sat in the stand. I began tagging every judgement with a confidence level, and contacted video analysts in Europe for an extra verification layer. A judgement with data but no witness is incomplete. A judgement with a witness but no data is too.

This is the counter-intuitive part I want read carefully. We usually think the biggest risk is wrong data. In my experience, the bigger risk is incomplete data presented as complete. A wrong number can be caught by comparing it with another source. A gap filled with an assumption cannot be caught, because it does not exist to be compared. It sits quietly in the report, looking exactly like a conclusion.

With esports this becomes more severe because of speed. The meta shifts with every patch, and each time, old prediction models lose part of their footing. In the transition window, new data fields are not yet fully collected, and that is precisely when people are most prone to reading gaps as stability. Whichever team wins its first three matches of a new season gets called a contender, even though the sample is three games, the opponents are weak, and the patch is unpatched.

I have no conclusion about any team in this piece, and that is deliberate. A piece about how to read data should not end with a prediction, because prediction is the easiest thing to be fooled by. What I want to leave behind is a habit.

Before trusting a data table, count how many cells are empty and ask why they are empty. Before concluding a team has no problem, check whether the system was designed to see that problem at all. And before calling a silence peaceful, remember that in any feed, the most frightening silence is the one that raises no error.

Every season is a ritual, and the analyst is merely the scribe of its omens. Omens lie in what appears, and in what should have appeared but did not. Whoever reads the gap will understand the match sooner than whoever reads the number. And in a major season, when the whole world stares at one scoreboard, the one who understands the gap is the only one not led by it.