When the Data Runs Empty: A Season Stolen by the Silence of the Spreadsheet

On 12 July 2026, I sat in an office in Liverpool with a data sheet in front of me that contained not a single number. It was the fourth time in fifteen years of work that I had received a completely empty data package. No player name. No tournament name. No scoreline. Not one usable statistical line. Only blank fields labelled "insufficient information" and a null-value protocol that obliged me to state the truth: there was nothing to analyse. For a sports data analyst, that is the closest sensation to losing your hearing in the middle of a final.

In tennis, an empty data package is not a minor inconvenience. Every metric we use is indexed to an entity: a player, a tournament, a surface, a round, a point in the season. A player's first-serve points won percentage only means something when you know whether he is playing on hard court in Indian Wells or clay in Monte Carlo, in March or in May. When that layer of entities disappears, the entire analytical system collapses from the root, not from the branches. I spent that week re-checking the process: the failure sat in the source-text extraction layer, not in the analytical layer. But the larger lesson lay elsewhere: we have grown used to believing that data always exists and merely awaits processing. The truth is that data may never have existed in the first place, and an honest analyst must be able to say so without fearing being deemed useless.

A professional tennis season generates millions of data points every week, yet only a small fraction of them carry genuine diagnostic value. This sounds paradoxical in an era when every serve is captured by radar. But look at the actual structure: an ATP 250 event can produce thousands of statistical rows, most of which merely restate what the eye has already seen. The numbers that genuinely change how we understand a match, the ones that force us to rewatch a clip in slow motion, are few. And of those few, most depend on season context; without that context, the number itself says nothing at all.

Take the question of ranking points defence. Every professional player lives inside a rolling 52-week system: points earned at the same event last year expire and must be re-earned this year. A player can hold his form, hold his technique, hold his fitness, and still drop dozens of places simply because this year's calendar offers him no opportunity to defend. Fans reading the news will see the headline "player X declines", when the truth is "player X had no event to play". This is precisely why I refuse to analyse a player without a specific temporal context. Without dates, without a week of the year, without a phase of the season, every judgement about form is an illusion.

The same happens with the concept of draw luck. A player who reaches the semi-final of a Masters 1000 without facing anyone inside the top 20 may be celebrated by the media as a "breakthrough". But if you place his section of the draw on the scales beside the champion's section, the difference in average opponent quality can amount to two or three ranking places. The result is real, but its meaning is entirely different from that of a player who had to eliminate three top seeds to reach the same round. Without draw data, these two stories look identical in print. That is why I always retain one line in every report: favourable draw or brutal draw, not pure achievement.

I have been criticised for this approach. An editor once told me that readers do not want to read about what we do not know; they want to read about what we do know. He was right commercially and wrong professionally. An analysis without an evidentiary foundation is not an analysis; it is a piece of emotional description wearing a professional label. In fifteen years of following matches, I have seen too many predictions delivered with absolute confidence and then collapse within a fortnight. Those predictions were not wrong because the writer was poor; they were wrong because they were built on a foundation the writer never re-checked.

What is notable is that this error does not discriminate by level. Leading experts commit it too. In 2026, at the round of 16 of the World Cup in Russia, I was a 23-year-old intern and predicted Spain would beat Russia simply because they held 71.4 percent possession and completed 1,029 passes. The possession figure did not tell me that they created only 0.9 xG across 120 minutes. That was the first time I realised a popular metric can be simultaneously arithmetically correct and semantically wrong. The lesson did not make me abandon data; it made me start every article with the hardest question: under what conditions is this number being measured, and do those conditions still hold right now.

In tennis that question is even harsher because of the extreme seasonality. Surfaces change week by week. Ball bounce changes with altitude and humidity. A player can win on hard court in Australia in January and lose in the first round in France in May, and there is no contradiction in that. But if you read only annual aggregate statistics, those two results dissolve into a meaningless average line. Old data is not wrong; I simply used to place it on the operating table in the wrong season.

When that empty data package appeared on my screen, my first reaction was to check the connection. My second was to check the source code of the extraction tool. My third, and correct, reaction was to write a report stating clearly: no information is recoverable, the process failed at the input layer, and any further analysis would be fabrication. I sent that report to the person in charge and received a short reply: "Can you not infer from what is already known?" The answer was no. A sports data analyst is not permitted to infer from nothing merely to fill a page.

This leads to a paradox that I think the sports analytics industry has not confronted seriously enough. We have built an industry on the assumption that data is always available, always complete, always in the right format. But data is produced by people, connections, servers, and systems that charge money to restrict access. A paywalled page, a JavaScript-rendered page that returns no text, a video without captions, a deleted page: all of these create gaps in the chain of evidence. And when such a gap appears, the industry's default reaction is to fill it with intuition rather than to acknowledge it.

I think this is where my working method differs from most colleagues. When there is no data, I do not write a longer article to compensate. I write a shorter one that states clearly there is no data. That does not endear me to editors, but it preserves the reader's trust. And in an industry where thousands of predictions are issued every season and then forgotten in silence, preserving trust is the only asset with long-term value.

The data gap also points to a larger problem in how we evaluate players. Players competing at major events have far more public data than players competing on the Challenger or ITF circuits. That means our analytical models are trained predominantly on the elite cohort, and when applied to lower tiers we are using a toolkit designed for an entirely different context. A player ranked 180th in the world may possess technique comparable to a player ranked 40th, but we lack sufficient sample to prove it, so we default to treating him as worse. This is a form of systemic bias, not a conclusion about ability. And it can only be recognised once we admit that our data has structural gaps.

There is one small detail from the empty-package incident that I keep as a reminder. When I re-checked the output file, I found a line of annotation generated automatically by the system: "Time sensitivity: not assessed. Source quality: not assessed." Those two fields were blank not because the system was lazy, but because it was honest. It did not know, so it said it did not know. We humans rarely manage that. We tend to fill blanks with a plausible guess, then forget it was a guess, then cite it as fact, and finally build an entire season narrative on a foundation of guesses whose origin has been forgotten.

That is why I believe the most important discipline in sports analytics is not modelling skill but the ability to say "I do not know" at the right moment. It does not matter how complex a model an analyst can build if he cannot distinguish between data and absence. And in tennis, where every passing week brings a new ranking, every tournament a new sample, every injury a new variable, absence occurs more often than we care to admit.

I do not trust a number, but I trust the story it tells after I have interrogated it at least three times. The first time I ask what it measures. The second I ask under what conditions it was measured. The third I ask how it would change if those conditions vanished. Only if all three answers are consistent do I begin writing. If the second or third has no answer, I stop and tell the editor that we do not yet have an article. That is not timidity; it is the only way an analysis survives a fortnight.

Looking back on that incident, I see it reflects something the sports analytics industry often overlooks: the value of a report lies not in its length but in the solidity of each claim inside it. A ten-page report with ten verifiable data lines can be worth more than a hundred pages with a thousand numbers of unknown provenance. And an empty report that states clearly why it is empty is worth more than either, because it saves an organisation a bad decision.

There is one thing I learned from this incident that I have never written down: the silence of data can be a stronger signal than any number. When a system returns nothing, that is information. When a tournament publishes no statistics, that is information. When a player does not compete for six weeks without an injury statement, that is information. People tend to read only what is written, whereas most of the truth in sport lies in what is left blank.

In the current season, with the majors approaching and the pressure of points defence beginning to form, I will keep this method unchanged. Every article will begin with a question about provenance. Every number will carry a line of conditions. And whenever the data is insufficient, I will write exactly what I wrote in the report of 12 July: insufficient information to conclude, and here is why that matters.

Measurement error is the least likeable friend I have, but the only one who never lies to me in a meeting room. And sometimes that friend arrives in the form of a completely empty spreadsheet, reminding me that my job is not to fill gaps with sound but to point out precisely where the gaps lie. Old data is not wrong; I simply used to place it on the operating table in the wrong season. This season, I will check the calendar before I operate.