The Empty File: When Professional Tennis Misreads Its Own Body

Set four, game five, the scoreboard showed a first-serve speed of 178 km/h — roughly 23 km/h below this player's career average. No doctor stepped onto the court. No injury file was opened. In the tournament's medical room, a status line sat untouched in its blank field: "Fit to play." The numbers had spoken, but no one read them. Three days later, the player withdrew with a diagnosis that anyone tracking serve data had already seen coming.

I sat with that match's dataset for a long time. Not to find out who was wrong. But to answer a narrower, more uncomfortable question: what happens when an injury file is allowed to remain empty, and why does an entire system stay silent in the face of that emptiness?

That is the story I want to tell today. Not about a missed serve. But about a blank space in a file — something more dangerous than any bad data I have encountered in thirteen years of work.

Context: a sport that measures everything except the thing that matters most

Professional tennis is one of the most densely measured sports on the planet. Every top-tier ATP and WTA event records serve speed, first-serve percentage, points won on second serve, net approaches, distance covered, sprint counts, heart rate, even racket-head angle at contact. Hawk-Eye logs every bounce to within millimetres. We know a player ran 0.3 seconds faster over ten metres than last season. We know he won 68% of points when serving wide.

But when that player walks into the medical room and says "it hurts a little," the system suddenly goes blind. His injury file at many tournaments remains an empty field. No data records when the pain began, how intense it was, how it progressed set by set, and most importantly: how long ago it was flagged before it erupted.

This problem belongs to no single tournament. It is systemic. Players compete as individuals, sign contracts with their own teams, and their medical data — where it exists — falls under their own control or their team's. Organisers have no obligation to publish. The ATP and WTA collect certain metrics for research and media, but most clinical detail stays in private drawers. Meanwhile, a Premier League footballer walks into training with a sensor on his ankle, and that data flows straight to the club's analytics room.

I spent three years at a sports-data company in Paris, specialising in football. When I moved to covering tennis, my first impression was not the poverty of technology. My first impression was the poverty of record-keeping discipline. An 18-year-old footballer at the Paris FC academy can have a thick file logging every hamstring pain since age 15. A 22-year-old player ranked in the world's top 30 can walk into a Grand Slam with a medical file of a few lines provided by himself and his team.

That is the central paradox of this sport: we measure serve speed to the millimetre per second, yet we do not measure a player's pain to the day.

Core insight: the gap is not in the body, but in how we measure it

I found the gap not in the player's body but in how we measure it. That sounds like a slogan. But it is the result of a long process, and I want to retell that process with concrete numbers, because without numbers this sentence is only a feeling.

In 2026, as a third-year sports-analysis student, I interned at the Paris FC youth academy. My task was to review the U19 team's medical files. I found midfielder Lucas Moreau, 18, had suffered three hamstring pain episodes in fourteen matches, yet the coaching staff kept starting him. I charted injury frequency against training load and showed that if he kept playing at that load, his muscle-tear risk reached 87%. The coach reluctantly gave the boy one week off. Lucas avoided serious injury and scored twice in his next three matches.

What I learned was not "I predicted correctly." What I learned was a structural error: the medical file recorded three pain events, but no one accumulated them into a trend. The data existed, but it existed in scattered form — three points across a season, with no line connecting them. And with no connecting line, no one sees the slope.

In tennis, that slope is even harder to see, for two reasons.

First, the tennis calendar shatters data continuity. A player can compete in Doha on hard court, three weeks later move to clay in Monte Carlo, then two weeks later to grass in Halle. Each surface has a different load profile on the hamstring, lower back and ankle. A model built on hard-court data gives wrong results on clay, where spin and sliding change the injury mechanism entirely.

Second, tennis has no "substitution" mechanism to shed load. In football, a slightly sore player can be pulled at minute 60, and second-half load redistributed. In tennis, if a player enters a fifth set, he must finish it. No release valve. No substitution. Only a body and a clock.

Combine the two, and you get a system where injury data is fragmented by surface and by event, and where no brake exists at match level. We measure serve speed in the fifth set, but we do not measure hamstring load tolerance in the fifth set.

I do not say this to deny progress. Modern medical teams of top players work very seriously. They have force plates, sensors, recovery-tracking software. But most of that data is for one person: the player himself. It does not enter a shared database. It is not cross-referenced. It is not verified by anyone outside.

And when data exists in only one room, it is in its most ignorable state.

When the file is empty, the system defaults to good news

This is the mechanism I want to expose, and it has a name: optimistic default.

In most sports record-keeping systems, an empty field does not mean "unknown." It means "no problem." If a player's medical file records no injury, scheduling algorithms read it as "eligible to play." If a load-tracking dashboard has no red dot, the coach reads it as "can push harder." Emptiness is not neutral. Emptiness is a false positive signal.

I have seen this repeat. I have seen a young player enter a fourth consecutive tournament week because nothing on paper said he shouldn't. I have seen a dashboard show perfect green while a player's serve speed had dropped 9% over three weeks. That green was not data. It was the absence of data, coloured in.

In 2026, when the season stalled because of the pandemic, I proposed building a "post-interruption injury-recurrence risk" model, based on data from previously interrupted seasons. I collected 1,200 medical records from five clubs. The result showed muscle-tear rates rose 23% in the first four weeks after football returned. My boss approved it, and the model became a diagnostic tool for lower-division clubs.

But there is a detail I never told publicly. Of those 1,200 records, more than a quarter had at least one blank field where the onset of injury should have been recorded. When I fed those records into the model, they did not error. They ran smoothly. The model read the blank field and assumed "no prior injury event" — meaning lower risk than reality. Had I not manually checked each record, my model could have produced dangerously wrong results: it would have told clubs some players were safer than they truly were.

That was when I understood the fate of this profession. Not finding injuries. But tracing the places where injuries were never recorded.

I once told a young colleague: the most dangerous part of a dataset is not the cells with numbers. It is the cells without them. Because an empty cell does not shout. It stays silent. And that silence, if you do not question it yourself, flows into your decision like a fact.

Contrarian angle: demanding complete data is also a trap

Here I must argue against myself, because that is the discipline I set for this work.

After years of hitting data gaps, my natural reflex is to wait. Wait for enough data. Wait for a complete file. Wait for every cell to be filled. But when I look back at the major decisions of my analytical career, I notice a different pattern: in many cases, waiting for perfect data did more harm than acting on imperfect data.

In sports injury, time is a live variable. An overloaded hamstring will not wait for you to finish your model. A risk model saves no one; it only tells you where to look. And the most important thing it tells you, sometimes, is simply: look where there is nothing to look at.

This runs against most people's instinct in the industry. When a file is empty, the reflex is to pass it through. When a file is empty, there is even a subtle social pressure: if you question a player too much while he is winning, you are seen as pessimistic, a spoilsport, someone who does not believe in fighting spirit. I have been called that myself when I questioned a young player on a winning streak.

I still remember one time. After a five-set match, a player left the court looking completely normal. The medical report said: no intervention. But in the data, I noticed a small detail: his second-serve success dropped from 62% to 41% across the last two sets, while his movements to his right to cover the backhand spiked. No cell recorded "back pain." But the shape of the data was speaking a different language from the report.

I noted it. I did not publish. I did not want to be a stalker of another person's body. But the next week, the player withdrew from a tournament with lower-back spasms.

This is not evidence that I am clever. It is evidence that match data, read correctly, is a medical file writing itself every minute. The problem is that almost no one is tasked with reading it that way.

The most dangerous system error: an empty file can flow straight into the final product

In working with automated sports-data pipelines, I met a type of error so quiet it makes you shudder. I call it "empty-file propagation."

The mechanism is simple. A system ingests data. The input is empty — no title, no source, no information points, no entity resolved. The system should stop, raise an error, ask a question. But without a validation gate, it does not stop. It moves on. The empty file is processed as a valid entry. It flows downstream, through further analytical steps, and at the end of the chain it becomes an item in the user-facing product — an item that looks normal, labelled, named, but empty or misleading in content.

This is the highest risk in this whole story, and I want to be clear: the greatest danger of data is not bad data. Bad data can be detected. The danger is empty data allowed to pass through as though it were good data.

An empty file entering a final product makes no sound. It raises no alarm. It simply leaves the product missing a truth that should have been there. And in the tennis environment, where decisions to play, rest, operate or delay surgery all rest on such files, a missing truth can be the decisive truth.

I think of a principle I carried from Paris FC. Back then, my coach, a very principled Frenchman, said something I have kept: bad data is dangerous, but it is still data — you know it is there. What is more dangerous is what is not there, because you do not even know you have to look for it. He was not talking about football. He was talking about how an organisation deceives itself.

A club with poor data but awareness of its poverty is safer than a club with "clean" data that does not know the cleanliness comes from missing measurement. By the same logic, a player with a complete but bad injury file is safer than a player with an empty file defaulted to healthy.

Paris FC taught me that bad data is more dangerous than no data. But thirteen years later, I must correct that sentence for accuracy: data that looks clean but is in fact empty is the most dangerous of all.

Why tennis files are more prone to emptiness than other sports

To understand why tennis is especially prone to this error, I need to analyse the sport's structure, not just its data.

Tennis is an individual sport organised as an ecosystem of independent entities. A player is a business. Her team — coach, fitness trainer, physio, personal doctor — is paid by and accountable to the player. The tournament organiser is responsible for staging, not for the player's long-term health. The ATP and WTA coordinate, but access to medical data is limited by privacy and by contractual relationships with each player.

This structure has a direct consequence: no one holds data across a player's career. There is no central repository. Each team holds one piece. When a player changes coach, the transferred file can be lost. When a player changes doctor, treatment history can scatter. When a player moves from one management agency to another, data can stay behind.

In football, the club is the enduring entity; players pass through it. Medical files belong to the club. In tennis, the player is the enduring entity; teams pass through him. And when teams pass through, the memory of his body can pass through with them.

This is the fundamental reason tennis injury files tend toward emptiness. Not for lack of technology. Not for lack of resources. But for lack of an entity holding continuous memory.

I have seen the consequences of this in many cases. A player returns from injury, and his metrics in the first two months are not compared with his own metrics from two years earlier. A player switches surfaces, and the new surface's injury risk is not accounted for because the old data sits in a different form. A young player enters the first tournament swing of his career with a packed schedule, and no system compares his load with the load a same-age player once bore.

Injury is a story — but that story begins long before the player collapses. The problem is that in tennis, the storyteller changes too often, and the listener was never present from the start.

What Grand Slams actually measure, and what they ignore

When a Grand Slam takes place, there is a vast machine of measurement and media. Every point is logged. Every metric is pushed to the scoreboard. Every match is analysed by dozens of experts.

But within that machine, there is a structural gap. The scoreboard does not show a player's recovery state in the two weeks before the event. It does not show hours of sleep. It does not show simmering tendon inflammation. It does not show that the player cut heavy training volume by 15% in the final week before entering.

These things exist. The player's team knows them. But they do not enter public space, and therefore do not enter any public analysis. Fans see a player walk on court and assume he is healthy. Journalists see a player lose and assume he has lost form. Both are reading a book with half its pages torn out.

I am not proposing that players' medical data be made public. That is privacy, and it matters. But there is a space between two extremes: total disclosure and no disclosure at all. A space that lets the public know a recovery process is under way, without needing clinical detail. A standard that lets analysts know "no data" does not equal "no problem."

I think of this whenever I read a line like: "Player X enters the tournament in good physical condition." No one outside X's team can confirm that. The phrase is not data. It is a belief repeated until it becomes convention.

And when a belief is repeated long enough, it becomes fake data. That is the most dangerous fake data of all, because it comes from good intentions.

How to re-read an empty file: three questions I always ask

After many years, I formed a small procedure for facing an empty file. It is not complex, but it works, and I want to share it with you as a tool, not advice.

Question one: is this empty cell "no event" or "not measured"? This distinction is everything. In some cases, an empty cell truly means the player never had a problem in that location. In many others, it only means no one ever measured. These two situations lead to completely different risk conclusions, but on screen they look identical.

Question two: does the match data agree with the report? When a player is recorded as having no medical intervention, but his movement metrics shift markedly between sets, one of two things is wrong: the data, or the report. In my work, I always suspect the report first, because the report is written by humans under pressure, while data is written by a system under repetition.

Question three: if this empty cell were filled, would my conclusion reverse? If the answer is yes, then I am never allowed to conclude based on that empty cell. I must state clearly that there is a blank, and that the blank may contain the opposite of what I am seeing.

These three questions require no high technology. They require a habit: when looking at a dataset, do not look at what is present, but at what should be present.

I believe in verified numbers. But I also believe part of verification is determining which number you are missing. A risk model saves no one; it only tells you where to look. And sometimes, the place to look is exactly where you were told there is nothing to see.

When football was paralysed, I began mapping risk from what no one bothered to look at

I want to return to 2026, because that was when I shifted from a young analyst to someone who questions his own tools.

When the pandemic stalled the season, the whole industry focused on vague tactical analysis. People wrote about unverifiable scenarios. I did not want to do that. I wanted to do something verifiable.

I collected data from previously interrupted seasons, including historical strike-hit Ligue 1 campaigns. I sought to measure muscle-tear rates in the first four weeks after football returned. The figure rose 23%. But more important than the number was its structure: most injuries occurred in players with at least one blank field in their recovery file. Not players with bad files. Players with incomplete files.

This was the moment that reshaped how I see this work. I realised my job is not to predict injury. My job is to find the places where injury can pass through without being recorded.

When football was paralysed, I began mapping risk from what no one bothered to look at. And when I moved to tennis, I carried that principle with me, but had to adjust it. Football has clubs. Tennis has individuals. In football, I could ask a club to fill the blank cells. In tennis, I often have no one to ask. I can only observe, cross-reference, and ask questions in public space.

That is why I write. Not to reveal secrets. But to increase the number of people who know that an empty file is not a clean file.

The bigger question: when does silence become a statement

In tennis, when a player says nothing about his injury, that is usually understood as neutral silence. It is not. Silence is a statement. It states: there is nothing wrong worth mentioning.

I have said that a risk model saves no one; it only tells you where to look. The same is true of data systems. They save no one. They only decide who gets seen.

And in tennis, one of the least seen things is a body quietly declining. It does not appear on the scoreboard. It does not appear in the rankings. It appears only in small metrics — a falling serve speed, a shifting movement trend, a second-serve point-win rate eroding over three months. These metrics, if anyone read them, would tell the story in advance. But almost no one is tasked with reading them that way.

I do not believe in luck; I believe in verified numbers. But when a number does not exist, I must believe in the question that produced it — or should have produced it.

For a player, this is a matter of professional survival. An unrecorded hamstring can become a lost season. An untracked ankle can become a two-year surgery. Career length in tennis today has been extended by sports science. But that benefit is distributed unevenly, because it requires something not everyone has: someone responsible for continuous record-keeping.

What I want to see change

I am not demanding a revolution. I am demanding a small, enforceable standard.

First, a clear definition of the state "no injury data." When a player enters a tournament, if the system has no reliable recovery data, that state must be recorded as "unverified," not as "healthy." This linguistic difference seems small. It is not. It decides how algorithms and humans make decisions.

Second, a set of match metrics tracked specifically for injury purposes, not for media. Not average serve speed, but its variability within a match. Not distance covered, but movement trend by set. These can indicate overload before any pain is reported.

Third, a simple rule: any medical file with a blank field at a key position must be treated as a file with risk, not as a clean file. Emptiness must trigger caution, not reassurance.

These three things need no new technology. They need a change in how we think about the absence of data. Absence is not zero. Absence is an unanswered question.

Open ending: the next empty cell

I return to the match at the start of this piece, the one I sat with for a long time over its dataset.

That player returned to competition. He won some matches, lost some. On forums, people wrote about him "recovering his form." His file remained almost empty. I do not know whether the pain in his hamstring is still there. I only know one thing from the data: his first-serve speed over the following three months remained below his old level, and it did not recover in a straight line.

I do not write this to predict anything. I write it to say that I am watching. And I am watching not what was recorded, but what has not been recorded.

In my profession, every empty cell may be an injury waiting. Every empty file may be a season fading. And every time we look at a clean dataset and breathe a sigh of relief, we may be misreading the body of the sport itself.

The question I leave behind is not which player will be injured next. The question I leave behind is: which empty cell, in the dataset you are looking at, is waiting to be questioned?