When the Spreadsheet Is Empty: How Sports Analysts Must Learn to Refuse a Conclusion

The first three matches inside the Orlando isolation zone, the GPS system returned a full set of files for each team. Enough rows, enough columns, enough player names. And empty. Not a single coordinate, not a single metre run, not a single sprint recorded. I sat in front of the screen at nearly three in the morning, hands on the keyboard, and a very familiar temptation surfaced: fill the blank cell with a number that sounds plausible.

In July 2026, I was 29, working as a data editor for ESPN, covering the MLS is Back Tournament in conditions with no spectators, no home advantage, no crowd noise. If I had pushed estimated figures into those 37 matches, nobody outside could have checked. The tables would still look tidy. The piece would still run on the front page. And every conclusion drawn from it would be wrong in the quietest possible way.

I chose otherwise. I typed "insufficient data" into the notes field, then spent two days rewatching footage of each match to rebuild by hand what the machines had dropped. A blocked analytical pipeline is not a failure of the method; it is a failure of the person who fills the gap with guesswork.

Modern sports analysis runs like a pipeline. Raw data flows in from tracking cameras, GPS vests, event data providers. Then comes cleaning, standardisation, tagging, model building, and finally storytelling. Each layer can only be as tall as the layer beneath it. When the first layer returns empty, everything downstream inherits exactly that ceiling, not one cell more.

Esports does not escape this architecture; it only renames the columns. In League of Legends, people read gold difference at 15 minutes, pick and ban rates, damage per minute. In CS2, the measures are Rating, ADR, pistol round win rate. In Dota 2, GPM, XPM, creep score. A model built on League data cannot be transplanted onto CS2 and labelled analysis. Metrics carry no universal meaning; they mean something only inside a specific rule set.

I still follow Vietnam's domestic scene. The biggest difference lies in infrastructure. A VCS team playing thirty matches a season may leave behind only a few dozen usable rows. That gap is not a gap in knowledge. It is an infrastructure gap, and infrastructure gaps are always filled by prejudice faster than by any model.

There are three distinct kinds of emptiness, and only one of them is a technical fault.

The first is emptiness caused by failed collection. Orlando 2026 belongs here. Files arrived, columns existed, values did not. The handling is clear in principle: log the fault, demarcate the missing data, and if you want to use it, rebuild it by hand. There is no cheaper shortcut.

The second is emptiness caused by bad definitions. The metric still has numbers, but the number does not measure what its name claims to measure. This is the most dangerous kind, because it does not look empty. A densely populated column can still be empty of information.

The third is emptiness caused by changed background conditions. The data is complete, the model runs smoothly, but the context is no longer what it was. Comparing a match with spectators to a match in an empty stadium, comparing a normal season to a season inside a bubble, is comparing two different things and then calling them by the same name.

In 2026, at 26, I had just left a master's chair in movement science and believed absolutely that data does not lie. I joined the Miami Herald, assigned to cover Miami FC in the NASL. My debut match was against Indy Eleven at Riccardo Silva Stadium. I meticulously logged midfielder Richie Ryan's passing: 87 touches, 74 passes, 91.9 per cent accuracy. I wrote a piece based entirely on the table, listing every parameter, and my editor killed it on the grounds that it read like dry toilet paper.

I did not argue. I quietly rewatched the entire match footage, then built an analytical frame I later called the Territorial Influence Index, combining receiving positions, passing direction and controlled space. When the second piece appeared with those same numbers, the editor ran it on the front page immediately.

The lesson was not that 91.9 per cent is a beautiful number. It was that 91.9 per cent only becomes information once I know where Richie Ryan received the ball, how he turned out of pressure, and what space his forty-metre cross-field pass opened behind the line. Raw numbers are mud; to see the truth you have to put your hands in it.

In 2026, at 27, I was at The Athletic as a data journalist. Before the World Cup in Russia, I built a prediction model on xG differential and PPDA, the number of opponent passes allowed before a defensive action. Put simply: a low PPDA means your team does not mind the opponent passing, as long as they pass it back into their own half.

I publicly predicted France would win while the analytical consensus leaned towards Germany and Spain. In the semi-final against Belgium, France's average PPDA was 7.8, extremely low, meaning they deliberately surrendered possession in order to counter. Belgium had a PPDA of 11.2 but lacked pace at the back. France won 1-0 in Saint Petersburg on 10 July 2026, then beat Croatia 4-2 in the final on 15 July 2026 in Moscow. My piece was shared more than 3,000 times. Russia 2026 is where I staked my entire reputation on the PPDA model, and I do not regret it.

But justified belief is not the same as unconditional belief. That is the boundary I have had to hold for my whole career.

In 2026, when Euro 2026 was staged late because of the pandemic, I handled data for a European football podcast. In the semi-final between Denmark and England on 7 July 2026 at Wembley, I noticed attacking midfielder Mikkel Damsgaard, who at that point appeared on no list of players worth watching. I calculated his pressing recovery rate across the tournament: 4.2 recoveries per match in the attacking third, the highest among players under 23. Against England, Damsgaard made five tackles and won all five, creating three chances from high pressing situations. Denmark lost 1-2 after extra time.

After that piece, three Premier League club scouts emailed me for more. What I remember most is not the share count, but that I had eye-checked every pressing action before publishing, because a recovery metric can be inflated simply by miscounting a duel that was never won.

Back to Orlando. When the GPS data came back to life, I had results from 37 matches: the average player ran 9 per cent less than the previous season, but sprint counts rose 12 per cent. More dead ball. Matches became more explosive in short bursts and more fragmented overall. I wrote a 4,200-word internal report arguing that the way we measure match effectiveness has to change when there is no crowd. It was later edited into a front-page ESPN piece and opened a debate about the new kind of match. Inside the Orlando bubble, the data fell silent, but the silence had an echo.

The biggest risk in this trade is not dirty data. It is a clean report. A perfectly formatted document, with every heading, every section, every conclusion in place, generates false reassurance across every analytical dimension at once. Readers cannot check which layer the empty column sits in. They only see prose that flows.

The industry's incentive structure pushes everyone in that direction. Newsrooms reward certainty, not caution. Audiences want a prediction, not a confidence interval. And when the deadline is nine in the morning, filling in a plausible number is always easier than writing two words: not enough.

Another trap is transplanting an analytical frame from one game title to another. Success in Russia 2026 does not give me the right to apply PPDA to a League of Legends dataset and call it a model. Every title is a new season; every patch is a new set of background conditions. Carrying the old frame across only produces the illusion of depth.

I have been wrong too. There was a stretch when I predicted a team would go deep at an international event based on a superior possession metric in the group stage, and that team was eliminated in the first round of the play-off bracket. The broken assumption was specific: I assumed group-stage tempo reflected knockout tempo, when the format change makes teams play very differently. Since then, every time I publish a prediction, I state which assumption is carrying the whole conclusion.

What I want readers to carry into this season is not a model. It is a habit: when you read a sports analysis, ask where the data came from, how many data points actually exist, and how many claims were verified by eye. Numbers that know when to stay silent are more trustworthy than numbers always ready to speak.