Injury Mislabeling: The Classification Gap Professional Tennis Has Known About Since 2026

On 9 June 2026, on Court Philippe-Chatrier, in the men's singles semifinal at Roland Garros, Carlos Alcaraz took the first set 6-3, lost the second 5-7, and in the second game of the third set called for the medical team. A medical timeout was recorded. The match ended 6-3, 5-7, 6-1, 6-1 in favour of Novak Djokovic.

That evening I sat in Paris with two screens. One showed the live feed. The other showed the tracking sheet I build for every big match: minutes played, first-step accelerations, changes of direction, average rally length in service games, and the real rest intervals between games.

By the end of the second set, Alcaraz's change-of-direction column had dropped roughly a quarter below his own average from the previous round. The first-step acceleration column, the initial push to reach the ball, fell earlier than the scoreboard column did. Alcaraz still won the second set. The body had filed its report before the scoreboard filed its own.

The label written on the medical card that night was cramp. In a strictly medical sense, that label was probably correct. And that is exactly where I want to begin.

A correct label can still be a useless label. Cramp describes a symptom. It does not describe a cause, does not describe onset timing, does not describe the load base that produced it. Everyone who left that stadium carried away a word. Nobody left with a question.

I have followed professional tennis since 2026, which makes thirteen years. Seven of those years I have earned a living by reading medical cards, withdrawal records and post-match statements, then cross-checking them against publicly available movement data. That work taught me something uncomfortable: most of the injury debate in this sport is a debate about vocabulary, not about body structures. And the vocabulary is being applied wrongly.

Tennis has had a classification standard since 2026. The failure sits at the operational layer.

Many fans assume tennis lacks an injury classification system. The sport has one, and it is fairly rigorous. In April 2026, a group of international sports medicine specialists published a document in the British Journal of Sports Medicine titled "Consensus statement on epidemiological studies of medical conditions in tennis." The author group included Babette Pluim, Colin Fuller, Mark Batt and colleagues from the International Tennis Federation, the ATP and the WTA.

That document set out very specific requirements. Every case must be recorded by anatomical structure or by diagnosis, never by symptom. Acute onset and gradual onset must be clearly distinguished. Tennis must use a broader concept than "injury" — the concept of a "medical condition" — because the share of gradual-onset conditions in this sport is markedly higher than in collision sports. And every statistic must carry a standard denominator, expressed in playing hours or match exposures.

In other words, tennis already knew that a mislabelled record produces false conclusions downstream. It wrote down how to label correctly. Then it walked onto court and did something else.

Now look at the daily operational layer.

When a player calls for the trainer mid-match, they receive up to three minutes for a treatable medical condition. The physio's sheet records the treatment given. The public record notes that the match was interrupted; it does not record a diagnosis. When a player withdraws, the draw sheet shows two familiar abbreviations. There is no diagnosis column in the public file.

The official media channels of the majors typically publish one short line: player X withdrew with an issue in a certain region of the body. A region. Not a structure. Not a diagnosis. No onset timing. No denominator.

Then comes the press conference. There, language compresses into safe phrases: a little tight, a bit sore, we'll see, nothing serious, I'm fine. These phrases are not wrong. They are simply unusable for any calculation.

So we hold two datasets. One sits on the scientific shelf: slow, aggregated, rigorously classified, published months later. The other sits in the daily news feed: fast, granular down to the individual game, and almost entirely unclassified. In the news cycle the fast set always wins. In research the slow set always wins. Nobody has built the bridge.

In March 2026, the International Olympic Committee published its updated methods for recording and reporting epidemiological data on injury and illness in sport, also in the British Journal of Sports Medicine. That update broadened the framework, adding fields for mental health, for non-injury illness, and for gradual-onset conditions. The standard kept rising. The operational layer stayed still.

The gap sits in classification, not in the player's body. That is the sentence I have repeated most often over seven years, and the sentence most often misunderstood. I am not saying injuries are imaginary. I am saying the way we name an event determines whether we can prevent it.

My register: 1,412 entries and five types of mislabeling.

Since 2026 I have kept a private register I call the football-and-tennis medical file. Each entry carries five fields: the player, the public label that appeared in media, the date the event was announced, the playing load over the preceding twenty-one days calculated from public data, and whether that label changed within fourteen days.

As of the current Grand Slam cycle the register holds 1,412 entries. Among them I have classified five recurring types of mislabeling frequent enough to count as systemic rather than incidental.

The first type is the structure-shifted label. The label is attached to a body region rather than to a specific structure. The clearest case is Andy Murray. Throughout 2026, media described his condition with a single word: hip. In January 2026, Murray underwent hip arthroscopy. In January 2026, he underwent hip resurfacing — an intervention aimed at the joint surface itself rather than the surrounding soft tissue. Two procedures, two different structures, two different rehabilitation protocols, two different risk profiles. Both carried the same four-letter label.

To a reader, those two events look identical. To a data analyst, they belong to entirely different categories. When a label cannot distinguish between two different things, every dataset built from that label is meaningless. And note this: the 2026 arthroscopy was an intervention on the labrum; the 2026 resurfacing was an intervention on the joint surface. A fan following Murray across those two years could not have known that the severity of the problem had changed. The label concealed the turning point.

The second type is the time-shifted label. The label appears on the day the player breaks down, while the event has been accumulating for months. Dominic Thiem is the case I use most often when teaching interns. In June 2026, in Mallorca, Thiem left the court with a problem in his right wrist. He underwent surgery in August of the same year. The public label appeared in June.

In my tracking sheet, Thiem's rolling three-month load index peaked for his entire career in the second quarter of 2026. This is inference from publicly available match data, not a medical record, and I say so explicitly every time I present it. But the meaning is clear: the label was applied in June, the phenomenon began in March. An injury is a story — but that story begins long before the player collapses.

Thiem retired in October 2026 in Vienna, at home, losing in the first round. Between those two dates lie three and a half years in which he never returned to his previous level. A label that arrives twelve weeks late may not have saved the wrist. But it could have changed how the coaching team allocated the schedule during the decisive window.

The third type is the category-mismatched label. This is the Alcaraz case from the opening. Cramp belongs to the symptom category, not the diagnosis category. Under the 2026 framework, such a case must be recorded with the temperature index, the point in the match, the number of games played, the recovery time since the most recent long match, and pre-match hydration status.

The scoreboard retained exactly one word. From 1-1 in the third set, Alcaraz won two of the remaining twelve games. That is the only number the official system kept alongside the label. It says something happened. It does not say what.

In tennis, the category error surfaces somewhere else too: the boundary between injury and illness. Any condition that is socially unclear tends to be pushed into the second category, because the second category attracts less scrutiny. The result is a data category full of cases that are never analysed, only hidden.

The fourth type is the ramp-mismatched label. The label describes the injury event correctly but leaves blank the most important phase: the return to competition. Alexander Zverev is the cleanest example. On 3 June 2026, in the Roland Garros semifinal against Rafael Nadal, Zverev rolled his ankle and left the court in a wheelchair, having lost the first set in a tie-break and standing at 6-6 in the second. The public diagnosis that followed was multiple torn lateral ligaments. That label was accurate.

What went unlabelled was the seven months out and the eighteen months after. Zverev returned to competition in January 2026. Only in June 2026 did he reach a Grand Slam final again, at Roland Garros itself, losing to Alcaraz in five sets. Between those two points lies a stretch the public data records very faintly: matches played, hours played, the degree of mechanical change in his service motion, the number of withdrawals at smaller events.

The ligament injury was recorded. The mechanical rebuild was not. And in my experience, the recurrence rate lives in the second half.

The fifth type is the chain label. This is the most dangerous type, and the one that draws the most pushback when I raise it. A chain label occurs when several independent events across several years are collapsed into a single word, usually a word describing character rather than a body: injury-prone, fragile, made of glass.

Bianca Andreescu is the case I have followed longest. In my register, the entries relating to her span multiple years, with different mechanisms, different body regions and different competitive contexts. In 2026 she won Indian Wells, Toronto and the US Open in a single season. Collapsing the entire subsequent period into one label is the most serious classification error a person can commit, because it converts a discrete series of events into a permanent attribute. The 2026 document warns explicitly against pooling. Nobody reads that part.

Juan Martín del Potro illustrates the southern side of the same problem. For years the public label attached to him was knee. The knee is where he broke down, and where he underwent multiple surgeries before retiring in Buenos Aires in February 2026. But his chain of events began with the left wrist in 2026, then the right wrist, and only then the knee. The body is a chain. The label sits on the last link, while nobody measures the first.

The load index is being read wrongly, and that corrupts both diagnosis and commentary.

There is a problem deeper than labelling. It concerns the metrics we use to describe load.

In tennis, the two most displayed metrics are total distance covered and sprint count. They are packaged as effort indicators. But high distance covered is also a consequence of being pushed wide in rallies you do not control. A player repeatedly stretched accumulates distance very quickly, and most of that distance earns no advantage. In other words, ineffective running also produces pretty numbers.

I build my tracking sheet on the opposite principle. I do not use total distance as a load metric. I split distance into three buckets: movement to reach the ball before the opponent, defensive movement, and recovery movement back to position. The second bucket is the most physically costly and the least mentioned in broadcast coverage.

Read that way, the load picture changes completely. A player winning quickly in three sets may cover less total distance than a player losing in five, yet their defensive-movement share of total distance may be far higher. And that share, in my reading of the data, is the variable linked to cumulative injury.

In my sheet, baseline defenders typically show a defensive-movement share of total distance roughly fifteen to twenty percent higher than baseline attackers, depending on surface and opponent. That is a wide enough gap to change schedule allocation, if anyone chose to read it.

There is one more variable few people include: service count. Not service games, but actual serves struck in a tournament week. A player going deep at an event can serve more than four hundred times in seven days. That number appears in no injury report, even though shoulder and wrist are among the most common injury sites in this sport.

The problem is not a shortage of data. Professional tennis has more data than any sport of comparable scale. The problem is that the data is packaged into broadcast product before it can ever be classified for medical purposes.

Data never lies; only the way we read it is wrong. But that sentence holds only while the raw data is intact. Once data has passed through three layers of broadcast packaging, what we are reading is no longer data. It is a product.

The women's tour: the longest-unfilled data layer.

There is a gap in this story I do not want to skip past.

Injury surveillance research in professional tennis has, for many years, sampled primarily from the ATP system. That means most of what we know about tennis injury has been built from men's data. When we apply that framework to the women's tour, we are not merely short of a sample. We are short of variables.

One major variable that has never appeared in the public register of any major tournament is the menstrual cycle. It is a load variable that affects bone density, plasma volume, core body temperature and recovery capacity. No data field records it. And because no field records it, it does not exist in any of the risk models we are currently building.

The second variable is the return pathway after maternity. This is a return-to-play route structurally different from injury rehabilitation, with milestones around muscle mass, pelvic control, load distribution and psychology. No public protocol exists for that pathway in professional tennis. Women have returned at the highest level for years, and every time they have done so with almost no dataset to reference.

The third variable is psychology. Only with the 2026 International Olympic Committee update did the classification framework gain an official field for mental health. Before that, withdrawals on those grounds were recorded with empty words, and empty words generate no calculations.

In my register, the label injury-prone is applied to female players noticeably more often than to male players with the same number of medical events in the same period. I do not yet have enough data to call this a firm conclusion. But it is a signal I will keep tracking, and I am naming it because staying silent about such a signal is its own kind of error.

A lesson from an interrupted season.

In 2026, when I was twenty-three and had just started as an analysis assistant at a sports data company in Paris, competition stopped because of the pandemic. The whole industry pivoted to vague tactical analysis to keep slots in broadcast. I proposed a different direction: build a model of injury-recurrence risk after an interruption, based on data from previous seasons that had been broken off, such as the Ligue 1 season disrupted by the 2026 strike.

I collected 1,200 medical records from five clubs. The result showed muscle-tear rates rising roughly twenty-three percent in the first four weeks after football resumed. The model was later used as a reference tool by several lower-division clubs.

What I want to describe here is not the result. It is how the result was used. We had exactly one indicator: which week since the return date. The first four weeks were the red zone. Then amber. Then normal.

Looking back, I see the limitation clearly. That model measured time, not players. Two men in their third week back could be at completely different risk levels, because one had played two matches and the other had played five. Our label was right by the calendar and wrong by the biology.

That was the first time I understood that a technically correct model can still drive wrong decisions when the unit of classification does not match the real unit of the problem.

A wrong label is not always a mistake. Sometimes it is a choice.

This is the hardest part of the piece, and the part I want to write slowly.

The most comfortable hypothesis about mislabeling is incompetence. That hypothesis is wrong, or at least incomplete. I have spent years reading medical records, and the level of expertise inside professional tennis medical teams is far higher than fans imagine. The problem is not competence. The problem is incentive.

Start with the player. A public diagnosis is a contract risk. It affects sponsorship negotiations, wild cards, protected rankings, and table position. A player in the middle of a negotiation has an obvious incentive to describe everything with a vague word. That does not make them dishonest. It makes them choose the largest unit of language that still works.

Next, the coaching team. A coach who says his player has a cumulative tendon problem has just admitted a scheduling error. A coach who says his player suffered an unlucky accident admits nothing. Same event, two labels, two liability outcomes.

Next, media. An accurate diagnostic label generates no headline. A pooled label generates a headline, and often generates an entire character profile. The phrase injury-prone is far easier to write than a torn lateral ankle ligament with bone marrow oedema. This is where I have to look at myself, because I earn a living writing about injuries.

Then the tournament and federation level. A full diagnostic label creates accountability. A complete register of competitive load creates questions about the calendar. No organisation voluntarily opens that door.

I am not concluding that all these people are coordinating a cover-up. I am concluding that all of them benefit from vague labelling, and none of them pays a price. That is a sufficient condition for a systemic error to survive across generations.

A three-minute medical timeout is a label, and it is also a tactical weapon.

There is another dimension I rarely see discussed seriously.

The three-minute medical timeout is the only event in a tennis match that both sides know precisely when it can occur, and both sides know what it can carry. Technically it exists for a treatable medical condition. Operationally it is a break that does not follow the rhythm of the match.

I have written before that excessively long review times are shredding match rhythm, and in tennis the equivalent phenomenon lives in the medical timeout. Two minutes of waiting is enough to cool a service sequence that was finding its groove. And the notable part is that both sides know it.

Combine those two facts and you get a very particular structure: a pause with tactical value, legalised by a medical label, where that medical label is not independently verified.

I have no evidence of any specific abuse, and I will not construct an indictment from what I do not have. What I have is a structural observation: when an action carries tactical benefit and must be accompanied by a medical label that can be written vaguely, the quality of that label will degrade systematically. Not because anyone is malicious. Because the system design contains no self-correction mechanism.

And this is where I have to audit myself. If this piece only catalogued other people's errors, it would become exactly the kind of article I oppose: a shallow indictment.

Paris FC taught me that bad data is more dangerous than no data.

In 2026, when I was twenty and interning at the Paris FC youth academy, I was assigned to review the medical records of the under-19 squad. I found an eighteen-year-old midfielder with three hamstring episodes in fourteen matches who was still being started continuously. I plotted injury frequency against training intensity and issued a warning that his risk of a muscle tear was very high. The coaching staff reluctantly gave him a week off. He avoided a serious injury and scored twice in his next three matches.

I have told that story many times at speaking engagements, and for a long time I told it as a victory. Recently I tell it differently. What I actually learned at Paris FC was not that I was good at predicting. It was that the club's records were terrible, and terrible records produce two possible outcomes: either nobody notices anything, or an intern notices something and is treated as a genius for two weeks.

When there is no data, people are cautious. When there is bad data, people are confident.

The same applies to my own register. I have had to delete forty-seven entries over the past two years, not because the events did not happen, but because the labels I had written myself did not meet the standard. One entry was recorded as a hamstring recurrence; on review it showed two different mechanisms at two different sites, and I had pooled them only because they sat in the same body region. That is precisely the error I had just criticised in others.

There is another temptation I have to actively block. When you build a risk model, you start seeing risk everywhere. A player who loses three straight matches starts to look like a physical collapse. Sometimes they are simply losing. Sometimes the opponent served better. Sometimes the surface at that event does not suit their game.

A player out of form may simply be out of form.

This is the boundary I must check every time I write. A risk model saves nobody; it only tells you where to look. If I turn it into a prophecy, it will do more harm than good.

The counter-example: when the label was right, the career lasted.

If my entire argument were criticism, it would have no empirical value. Fortunately there is a counter-example, and it is the largest one in this sport.

Rafael Nadal was diagnosed with Müller-Weiss syndrome in his foot in 2026, when he was nineteen. It is a structural condition, not an acute injury. It has no recovery date. It has a management protocol.

Over the following nineteen years, the public labels attached to Nadal kept changing: knee, abdomen, wrist, hip. Those labels were correct at the event level and wrong at the system level, because they obscured the most important fact: there was a fixed underlying condition, and every acute event above it had to be read in relation to it.

Nadal retired in November 2026 in Málaga, in Spain's Davis Cup colours. That career stretched nineteen years from the date of diagnosis, with twenty-two Grand Slam titles. I do not claim that number is the consequence of a single label. But I do claim it could not have existed if the underlying condition had not been named correctly in the very first year.

This is what I want readers to carry away. Correct labelling is not an administrative procedure. It is the condition that separates something manageable over twenty years from something that ends a career in two.

What would change if we relabelled from scratch.

I am not proposing that players' medical records be made public. That is private data, and that privacy matters more than my analytical needs.

What I propose is far smaller and can be implemented within a single season. Three additional data fields in the public record of every medical event: structure or structure group, estimated onset timing, and competitive load over the preceding twenty-one days. No detailed diagnosis required. No prognosis required. Just enough that the dataset does not break at the exact point that matters most. As a data professional, I judge the operational cost of those three fields to be far lower than the preventative value they generate.

But I also have to say something else honestly. Even with those three fields, we will still mislabel. Because the human body does not deliver data in categories, and because there is always a lag between the moment a body begins to change and the moment a person recognises the change.

My job is not to erase the lag. My job is to measure it.

Based on my experience following professional-level matches for thirteen years, what keeps me in front of two screens every night is not the feeling that I am about to predict something. It is the feeling that the numbers are still saying something nobody has agreed to write down.

I do not believe in luck; I believe in numbers that have been verified. And the verified numbers are telling me that next season will repeat this season's sequence precisely, unless classification changes before treatment does.

If next season you see a player leave the court with a medical card, try one small thing. Before trusting the label, ask yourself: when did that player's body start sending signals, and who in the room saw them before they were given a name?