Tennis and the Zero-Denominator Trap: When an Empty Data Sheet Still Generates a Story

In January 2026, on Court 7 at Melbourne Park, I sat taking notes on an Australian Open qualifying match with a statistics sheet open on my laptop. After two sets I had first-serve percentage, second-serve points won, and break-point conversion. Total sample: 118 points. Statistically, that is not a sample capable of saying anything certain about a player's ability. Yet that evening I counted four online analysis pieces asserting the player had "found the formula" for the new season — four pieces, from one match, from a denominator any working analyst knows is insufficient. That was the moment I realised the biggest problem in tennis analysis is not a shortage of data. It is that an empty data sheet can still produce a fluent story. Tennis has the densest event calendar and the most fragmented data structure of any widely followed individual sport. A professional plays 20 to 25 tournaments a year across at least three surfaces — hard, clay and grass — with entirely different bounce, speed and friction characteristics. A sample that is "large enough" on hard court can become meaningless the moment a player enters a grass season lasting under five weeks. Grass is the clearest example. Only about three weeks separate Roland Garros from Wimbledon, and most players contest at most two events in that window. With two tournaments to judge a player's grass adaptation, you are talking about a few hundred points — enough to describe what happened, not enough to predict what will. The rankings structure compounds the problem. The rolling 52-week points window allows a player to climb sharply on one favourable week, then lose exactly those points a year later. The ranking therefore reflects accumulated results, not the quality of the process behind them — a distinction very few reports bother to make. Then there are the advanced metrics. In football, xG has been standardised and validated across hundreds of thousands of shots. In tennis, metrics such as shot quality or serve impact are still being built, and they depend directly on the quality of the point-by-point foundation. An ATP 250 event in its outer rounds may not have enough cameras to record ball position. The analyst then enters data by hand, and every model downstream inherits that error. Even the seemingly objective elements are shaped by collection conditions. The ATP serve clock was designed to standardise match rhythm, but it only functions where officials and measurement systems are fully present. On outer courts, with fewer cameras and thinner staffing, parts of the rhythm data disappear before any model touches it. When I worked in fact-checking early in my career, the rule was: every number must be traceable to its source, and every tactical claim must be cross-checked against at least two quantitative indicators. It sounds simple, but it eliminates most of what currently passes for analysis on social media. Take break-point conversion. A player creates 8 chances and converts 4 — 50%, which sounds impressive. But with a sample of 8, the 95% confidence interval spans roughly 16% to 84%. The 50% figure carries almost no information. Another player converting 2 of 8 — 25% — may be entirely equivalent in true ability. Yet post-match reports still write that player X is "ice-cold on break points" or player Y "lacks nerve", as though a stable trait had just been measured. The same happens with first-serve percentage. Match-to-match variation for a single tour-level player is often larger than the average gap between players. In other words, most of the difference we see in one match is noise, not signal. But noise cannot be told as a story while signal can — so people tell the part that is easy to tell. Then there is the word "form". In tennis, form is usually inferred from a recent results sequence: semi-final, quarter-final, final. But results sequences depend heavily on the draw. A semi-finalist may have needed only three wins against opponents outside the top 50; a fourth-round loser may have eliminated two seeds. Without opponent-quality adjustment, a form table is just a fixture list rearranged by feeling. In football I once built a pressing metric to compare teams on a single scale. The tennis equivalent is an index adjusted for opponent and surface. Most media reports do not do this, partly because of cost, partly because a simple story sells better than a complex truth. There is a principle I learned over years: error does not vanish because you fail to measure it. It merely migrates into the conclusion. A model without a confidence interval still makes predictions — it simply does not know how much to trust itself. In 2026 I learned that even a 95% probability still has a 5% that laughs. The first data rebellion was never about overthrowing anyone — only about proving that a number deserves to be heard. Now the contrarian part, and this is the point I want to keep longest. The data explosion in tennis has not made conclusions more accurate. It has made them more certain. Those are two entirely different things, and the sports analytics industry is systematically confusing them. When an analyst gains more data, the natural reflex is to draw stronger conclusions — more confident, more decisive, with less room for doubt. But more data should only narrow a conclusion if that data genuinely reduces uncertainty. For a 118-point qualifying match, three extra metrics do not enlarge the denominator. They merely make the article look more professional while the conclusion still stands on the same foundation of sand. And there is a deeper layer few in the industry want to discuss. Most point-by-point data in tennis does not flow into analytical articles. It flows into betting companies, where the same table is used to price risk. That is the darkest side effect of digitising sport: analysts write with data that the betting market has already used to finish pricing. Data does not lie; it is the reader of data who makes excuses. So when a model says a player has a 72% chance of winning, the right question is not whether that number is true or false, but how many points it was computed from, on which surface, and over how many weeks. If the answer is "none", then it is not a prediction — it is a sentence typed in the font of a prediction. The greatest concern is not that we lack tennis data. It is that we have learned to write as though we do not. An empty spreadsheet stops no one from publishing a nine-dimension report. It only stops the writer from saying the hardest sentence in the trade: there is not enough information to conclude. The question for next season is not who will win. It is: how many analyses will be written from insufficient denominators, and how many readers will notice before the next match begins?