When the Data Table Is Empty: The Line Between Analysis and Guesswork in Tennis

On July 15, 2026, in my apartment in Queens, New York, I reopened the statistical sheet from the Wimbledon final played the day before. Roger Federer was serving in the fifth set at 8-7, 40-15 — two championship points within reach. Novak Djokovic saved both, dragged the match into a tiebreak at 12-12 and won it 7-3. Then the summary page produced one line that made me stop: Federer had won more total points than Djokovic across the match.

That night I read about forty articles. Most opened with that same line and went straight to a verdict. One accurate data point, used as a ruling. I published nothing for the next two days, because I could not answer my own question: how much of the result does winning more points actually explain? After pulling apart game by game, I estimated that share at roughly 35 to 40 percent. The rest sat in the weighting of critical points, in return positioning, and in something the stat sheet has no column for.

The tennis data pipeline

Tennis has the densest data infrastructure of any individual combat sport. That is both an advantage and a trap.

From 2026, the US Open used Hawk-Eye to support line calls. By 2026 that tournament moved to fully automated line calling; the Australian Open followed from 2026; the ATP announced broad adoption from the 2026 season. The 25-second serve clock appeared on the ATP Tour in 2026 and spread to the Grand Slams the same year. Off-court coaching was trialled by the WTA from 2026, extended by the ATP from 2026, and gradually formalised across the Grand Slams during 2026–2026, with differing limits at each event.

The second layer is match data: first-serve percentage, points won on first serve, points won on second serve, return points won, break-point conversion, winner-to-unforced-error ratio. The third layer is the 52-week rolling ranking system, where every week that passes erases a week from the year before.

All three layers can fail. A qualifying match at an ITF event has no Hawk-Eye. A mid-match retirement cuts the sample nearly in half. A roof closes for rain and ball conditions change from the seventh game onward. In those situations the pipeline returns an empty cell. How a writer handles that empty cell decides whether he is doing analysis or doing prose.

The chain of evidence when data runs short

After one bad call, I set myself a rule: every judgment must rest on at least two independent layers of evidence, and when the second layer is missing, I must write insufficient information to assess rather than fill the gap with adjectives.

Total points is the most abused metric in tennis. It is easy to collect, easy to read, and almost always right at match level while wrong at game level. In the 2026 final, Djokovic won three of four tiebreaks. A tiebreak is where every point carries nearly absolute weight, while a point in the third game of the first set carries a fraction of that. A flat sum cannot describe a structure with tiered weighting. When I isolated second-serve points won inside break-point games, the gap between the two players opened far wider than the total-points line suggested.

Break-point conversion is the noisiest metric on the sheet. A player can post roughly 20 percent in one match and 50 percent the next with identical return quality, simply because the sample is tiny — usually six to twelve chances per match. In 2026 I published a break-point ranking for an ATP 250 event based on nine chances, and had to correct it three weeks later. Since then every table of mine carries the sample size beside the number, and every conclusion carries a confidence band.

Ranking-point structure matters more than ranking position. A player ranked fifth may sit only a few hundred points above the twelfth, yet most of those points are concentrated in two weeks of a single major. The defence window is what the ranking table does not show. From March to June I track three columns: points to defend over the next eight weeks, matches played in the last 21 days, and consecutive three-set matches. When two of the three deteriorate together, my estimate of that player leaving the top 10 within six weeks rises to roughly 60 to 65 percent. That is a number that moves with the data, not a fixed forecast.

Surface conditions are the most underrated variable. On the same hard-court surface, the ball at Indian Wells and Miami returns a different bounce profile. On the same clay, Monte Carlo and Rome produce different slide. A player serving well in dry, cool conditions can lose 5 to 8 percentage points of first-serve points won once humidity rises. If I do not log the weather, I am comparing two things that cannot be compared and calling the output form.

The role variable is the most expensive lesson. In 2026 I built a model for a player moving from clay to hard court and concluded the win rate would hold. It dropped about 7 percentage points over eight weeks. The data was not wrong; it measured exactly what it was designed to measure. The error was assuming the playing role was constant, when a new coach had pushed that player roughly half a metre behind the baseline and asked for deeper returns. That half metre erased most of the edge the model had calculated.

So every analysis of mine contains a section called the role variable. Tactical system first, numbers second. If a player is asked to attack the net more, his baseline metrics will fall, and that is design rather than decline.

Rules and on-court transparency. There is another gap data cannot fill: the mechanism of explanation on site. When a player calls a medical timeout at a turning point, the crowd and the television audience receive a brief notification, not an explanation. When the 25-second serve clock is exceeded without a clear penalty, the official carries no obligation to state a reason. Transparency only has value when someone on court is accountable for stating the reason, and the current mechanism does not deliver that. This is the part of the data I call negative data — it exists as silence, and silence does not enter a spreadsheet.

Injury risk and schedule density. I place injury risk in the hardest-to-quantify bucket, because it depends on three variables at once: entry density, surface type, and the number of matches lasting beyond two hours. A player contesting four three-set matches in ten days in humid conditions carries a markedly higher probability of soft-tissue trouble than one playing two. But that probability cannot say who will break down, and I will not write otherwise.

Short samples and the illusion of dominance. On the WTA side the sample problem is even more visible. A player winning four Roland Garros titles across roughly five years renders any model built on a ten-match sample meaningless, because the model cannot separate a durable skill from a long lucky streak. I once fitted a clay regression on 14 matches and got a confidence interval so wide it was useless.

Media narrative and the expectation gap. The hype cycle in tennis runs on a very short clock: five straight wins and a player is a title contender; two losses after that and it is a crisis. I measure the expectation gap across three columns: tournament results, ranking trajectory, and commercial value. Most praise and most criticism diverge on the second and third columns, while only the first has data dense enough to support a claim.

Industry transmission: prize money, sponsorship, equipment. The downstream flow is fairly legible: major prize pools determine academy investment, academy investment determines the depth of the next cohort, and that depth feeds back into sponsorship value. A small change in qualifying-round prize structure may shift hundreds of small decisions inside coaching teams within eighteen months. That is a mechanism argument, not a measurement. I rate it medium confidence and would need two more data layers before turning it into a conclusion.

The contrarian angle: an empty cell is a data point

An empty spreadsheet does not say the match had nothing worth reporting. It says the pipeline broke somewhere: entity recognition, extraction, or simply the tournament sitting outside coverage. An empty court does not make the result wrong, it only strips away our illusions. In nearly every case where I received an empty dataset, the culprit was collection rather than the court.

Sports media applies an opposite force: the gap must be filled before deadline. When data does not arrive, adjectives do. Character, class, decline — three words standing in for three missing columns. I understand the pressure, and I still choose differently.

One concrete example. A player's first-serve percentage rose from about 58 to 67 percent after a coaching change, and four articles that same day credited the new coach. My dataset showed that over the same window the player moved from clay to hard court, played fewer matches, and faced opponents ranked roughly 15 places lower on average. Four variables moved together. Attributing the entire difference to one of the four is a defenceless reading, and I do not write it.

Fans look with their eyes; I look with a probability distribution. The truth sits deep under the numbers, where headlines never reach. If the table is empty, I will say it is empty, list the missing fields, and state the verification threshold required to fill them. That is an answer, not an evasion.

What to track in the next round

Over the next six weeks I will track three signals. The group of players entering their largest points-defence window — that group usually surfaces before the ranking moves. Matches with a gap above 12 percentage points in second-serve points won that nonetheless went the other way; that is where my model is weakest. And the count of matches lasting beyond two hours thirty in high humidity, an early indicator of injury risk in the following round.

I do not write about tennis; I only transcribe scripture from data. When the data has not arrived, the most honest transcription is to record the gap itself — with a timestamp, a source, and a threshold. The next round will show who filled the gap correctly, and who merely drew an outline around it.