Small Samples and the Narrative Machine: Data Discipline in Tennis

On the third night of a tournament in Melbourne, I opened a raw data file and found it empty. The file had not failed to download, and the format was not broken. Every metric — first serve, return, decisive-point conversion — came back as zero, because the subject of analysis had never been identified. In eleven years running a data desk, I have learned that the most dangerous moment is not when the numbers look bad. It is when the file is empty and someone still wants a story on the page before deadline. When the whole world watches the goal, I watch the run off the ball — but when there is neither run nor ball, the only honest thing to say is that the desk has nothing to say yet. Tennis is the most data-dense individual sport on the calendar. Every point is a line, every game a sequence, every match a set of events that can be stripped down to each individual serve. That density is exactly why tennis breeds the worst narratives. Three good matches in an early round are enough for the media to crown an eighteen-year-old as the heir. A first-serve points-won rate that spikes for one week is enough for someone to write about a career turning point. I work for the Australian market, where the January summer turns Melbourne into the tennis capital of the southern hemisphere for two weeks, and where every editor wants a story before the player has even left the court. I long ago accepted that most of the beautiful numbers I see on a broadcast graphic are not data. They are decoration. My craft begins with verification discipline. Before I write a single line, I rebuild the longitudinal data chain behind a number, and I publish only when the source is traceable. I built a nine-layer process — technical analysis, form, tournament systems, tour positioning, medical, management, risk, media narrative and industry transmission. It sounds heavy, but the point of the process is not to make the report look impressive. It exists to force me, when faced with an unidentified player and a data set that is too short, to say plainly that no conclusion is yet possible. Take Emma Raducanu. In September 2026, an eighteen-year-old Briton entered the US Open through qualifying, won ten straight matches without dropping a set, and lifted the trophy. The world immediately had a perfect story: prodigy, overnight star, the future of women's tennis. I watched every data line she produced across those two weeks, and I remember that alongside the emotion I had to place one cold fact: ten matches are far too small a sample to define anything. After the title, her career was torn apart by injuries, coaching changes and expectations no one could control. That does not make the title any less great. But it turns every heir narrative into a rounding error rather than a statement. At my Melbourne desk, we separate process metrics from outcome metrics and noise metrics. The process group — first-serve points won, return points won — tends to hold steady across a full season cycle. The outcome group — titles, wins, ranking points — is what the audience sees. The noise group is the spikes that appear for one week and vanish, usually the product of a soft draw or an opponent in decline. Tennis narratives are built on the noise group, because it is the loudest and the easiest to sell. The real job of a data journalist is to translate backwards from noise to process, and to tell readers when the evidence chain is not yet long enough to do so. I learned this from a different divergence. In 2026, at a major tournament, I spent most of my time measuring a football team's pressing frequency — the PPDA — and concluded that the side reached the final through system, not inspiration. Weeks later I received written confirmation. But that very moment taught me the opposite lesson: one correct metric does not mean every metric around it is correct. I began testing myself with a single question on every piece I wrote: if a counter-evidence metric existed, where would it be. When I cannot find it, that is not proof I am right. It is a sign I am reading the data through my own bias. In tennis this test matters more than anywhere, because every season brings a fresh spin of small samples. A young player wins an ATP 250 and is called a phenomenon. An older player wins a five-setter and is called reborn. Both statements can be true, but neither should be published before I can see the data chain across a season, across three seasons. A metric is an X-ray machine, not a scoreboard. It is there to see through the tactical surface an opponent is hiding, not to colour in a narrative that was written in advance. The most counter-intuitive thing I have drawn from that nine-layer process is this: an empty result is also a valid result. When the input has no title, no source, no player identity, every model must return a null value — and that is honest behaviour, not failure. Sport has a dangerous habit of treating no data yet as invent some data. I call it the narrative trap: when a reporter has nothing to write, they write about the crowd's feelings, about atmosphere, about a player who transformed without a single process metric behind it. Numbers never lie, but it took me many years to know when they tell half a truth — and when they say nothing at all. The silence of data is not a gap to fill. It is a refusal to side with the easy story. The next round of the season will bring new small samples, and again someone will want me to write about them before they mature. My job is to keep the file open, log the publication date, trace the source, and wait for the chain to grow long enough. The one recommendation I send to the newsroom: do not publish a fresh narrative without at least one process metric beside it, and be open when the evidence chain is still too short. If my story tomorrow reads like a dry data table, at least it is honest. If it reads like a fairy tale, I have failed — because I sold you a gamble on a denominator that was never large enough.