When the Stat Sheet Goes Blank: Verification in Professional Tennis

At minute 47 of the fourth set, the screen in front of me went blank. The first-serve percentage column dropped to zero. Points won on second serve dropped to zero. The point-by-point timeline, which had run without interruption for three and a half sets, froze at minute 39. Out on court, the match went on. The stands kept roaring. The ball kept bouncing. Only the data was silent.

In the press workspace there are two kinds of reaction. The first opens a draft that is already three-quarters written, pastes in a few impressionistic lines — "his serve looked really solid today" — and hits send. The second stays put and checks where the feed broke. I belong to the second kind, which is why I spent an extra forty minutes on a match a colleague had filed long before.

The story worth telling is not those forty minutes.

Context: tennis data is not a source, it is a chain

At Grand Slam level, no single machine owns the whole truth of a match.

At the head of the chain sit the ball-tracking camera systems — Hawk-Eye and its equivalents — mounted around the court, typically around ten cameras on a fully equipped show court. Those cameras produce two different products: in/out decisions for electronic line calling, and ball-coordinate data for analysis.

In the middle sits the aggregator. Since 2026, the ATP and the WTA have operated a joint venture called Tennis Data Innovations to manage the official data stream for both tours. That data flows onward to broadcast scoreboards, to betting operators, to player profiles on tournament websites.

At the tail end sit independent datasets. People like Jeff Sackmann, with his open ATP data repository, have spent years reconstructing match history from public results and statistics, letting anyone cross-check backwards.

These three layers do not always agree. When they disagree, the writer has to choose whom to trust. Numbers whisper. The person willing to listen hears an entire match — but only after learning where the whisper comes from.

The problem multiplies further down the pyramid. A Challenger court does not have enough cameras. A qualifying match may have nothing but a person with a notepad. The data still exists there, but it is thinner — and that thinness never shows up on the chart.

The core: a chain of evidence

I started with the number most likely to be wrong: first-serve percentage.

It sounds harmless. It is the number of first serves that land in the box divided by total first serves. But it depends on a convention: how do you count a serve that clips the net and lands in? In many systems that is a "let", replayed, not counted. In other data streams it gets logged as a missed first serve. Same match, same player, and first-serve percentage can differ by two or three percentage points purely because of the convention.

Across a match with 120 first serves, three percentage points is about four serves. Not much. But enough to reorder a player inside a ten-name comparison table.

Then come the rally metrics: points won on second serve, net approaches, win rate in rallies of four shots or more — what the datasets call rally length. All technically valid, and all dependent on whether the system captured the end of the point. If the camera loses the ball at the sixth shot, that rally vanishes from the sample.

And here is where I want to slow down most: break points.

A five-set match might contain only six or seven break opportunities. Six observations. With six observations you cannot measure "nerve at the decisive moment." You can measure a slightly biased coin sequence. But in print, those six observations routinely become a stable, repeatable, forecastable psychological quality.

A concrete case. In the 2026 Roland Garros men's singles final, Carlos Alcaraz beat Jannik Sinner in five sets over 5 hours 29 minutes. The Spaniard saved three championship points while Sinner served at 5-4 in the fourth set.

What the data says: Alcaraz won three consecutive points in that situation.

What the data does not say: that he possesses a special capacity at decisive points.

To claim the second, I would need a far larger sample — hundreds of break points across multiple seasons, benchmarked against that same player's ordinary point-win rate. When you run that comparison across years of ATP data, the gap between "break-point win rate" and "ordinary point win rate" tends to shrink substantially, and for most players it sits inside the range of random variation.

That does not make the final any less great. It just makes the story less certain.

My verification routine for each match has four steps: reconcile the official scoreboard against my own log; check how many rallies carry coordinates; identify the points where sources disagree; and write down what cannot be verified. The last step is the one most often skipped. Nobody wants to publish a blank cell.

When the line is called by a machine

If I had to pick the single biggest change in professional tennis over the past few seasons, I would pick the departure of line judges.

The US Open used automated in/out calling from 2026. The Australian Open followed. In October 2026, Wimbledon announced it would drop line judges from the 2026 season. Roland Garros moved to electronic calling in 2026 as well. All four Grand Slams now call the lines by machine.

On error rates, this is progress. No more officials surrounded at the back of the court, no more three-minute arguments between games.

But I keep an old reservation, and I will be explicit that it is a reservation, not a conclusion: when every ball is measured to the millimetre, the cost of a risky shot goes up. Players learn that a forehand a few centimetres inside the sideline buys only marginally more than a safe ball up the middle — while the error margin is far larger. If the calling system gets more precise, the reward for hitting the line does not rise, but the risk stays the same. That is a skewed equation.

What I lack is evidence strong enough to turn the reservation into a finding. To prove it, I would need to compare the rate of line-clipping winners before and after the switch, within the same group of players, on the same surface, controlling for ball and weather conditions. No such dataset exists publicly. So it stays a hypothesis.

The blind spot sits off court

There is a trap I once fell into, and it bears directly on how we read tennis data.

In 2026, I was running a results model for European competitions. Home advantage in the model was priced at roughly 0.45 goals per match. When competitions returned to empty stadiums, that figure fell to about 0.08 across nine rounds. The model was not technically wrong. It was wrong because it lacked a variable that had never previously existed: the crowd. Home advantage is not only geography, until it disappears.

Tennis has analogous variables under different names. Surface. Altitude above sea level. Ball type. Roof conditions. None of these appear in a player profile, but all of them live inside the results.

A player who wins 70% of matches on hard courts does not carry that number onto clay. The ball bounces higher, preparation time lengthens, and long rallies become the default rather than the exception. Pool every surface into one win rate and you are averaging two different sports.

One more variable analysts forget: the calendar. The ATP and WTA points structures turn title defence into a timing problem. A Grand Slam title is worth 2,000 points. Runner-up, 1,300. Semifinal, 800. Quarterfinal, 400. A Masters 1000 title is worth 1,000, the runner-up 600.

Which means a defending Grand Slam champion who loses in the first round the following season drops 2,000 points — roughly two ranking places gone after a single match. No technical metric on the match stat sheet displays that pressure. It sits off court. Misjudge one variable and you lose your bearings for an entire year.

The counter-reading

There is a paradox in how we read tennis through numbers.

The more detailed the stat sheet, the easier it is to believe it explains the match. But every metric added increases the number of ways to misread it. A player who serves better in the fifth set is not necessarily serving better than he did in the first — perhaps the opponent has faded physically, perhaps the balls are heavier in the humidity, perhaps he simply changed his serve direction after being read in the third.

Correlation is not causation. A high rate of points won on second serve correlates with winning — but it correlates because both depend on the quality of the first serve. Serve well on the first and you rarely face a second; when you do face one, it is usually from a position of comfort.

This is why I never put a metric into a piece without stating which system produced it, on which court, and by whom. Before you trust a number, ask where it was born.

In the specific case of that press room, the answer was: it was born on a server that stopped responding for forty minutes. And the blank stat sheet is not a rare event. It is the default state of tennis data across most of the world's tournaments — the only difference is that there, nobody sees the white space.

What to watch next

The current data points to a clear trend: the biggest events are consolidating data into fewer providers, while independent open datasets grow more complete through community effort.

Next season I will watch two things. First, whether any provider publishes a detailed methodology for positional data — ball coordinates and player positions. Second, whether anyone attempts a cross-check between two independent sources at Challenger level.

If neither happens, we will keep living with stat sheets that are clean, complete, and unverified. A season missing details is like a match missing stoppage time.