The Four-Shot Rhythm and the Hidden Metric Shaping Elite Tennis

Four shots, and a stand that never got to its feet

Fourth set, seventh game, three-all. The server tosses high, the ball lands in the diagonal box, a two-handed return drives deep into the left corner, an inside-out forehand follows, then a soft blocking volley. Four shots. The point ends before the stand can rise. I was sitting in the press area, pen still in hand, and only one thought was in my head: the point that just happened barely exists in any statistic the broadcast will put on screen after the match.

That was the moment I realised I was watching a different sport from the one most data panels describe. First-serve percentage, first-serve points won, break points converted — all correct, all useful, and all silent about the thing that decides the shape of a match: how many shots a point is built from, and who controls that rhythm.

Context: When the stat sheet misses half the match

For years I built prediction models on the milestones everyone can see. First serves in, aces, unforced errors, break-point conversion. Those numbers are seductive because they are easy to collect, easy to compare, and easy to fit into a thirty-second broadcast graphic.

The problem lies elsewhere. An elite tennis match is not decided by what appears on the scoreboard after each game. It is decided by the structure of each individual point: who initiates the second and third shots, who forces the opponent to hit one more ball, and who accepts that the point will end inside the first four shots.

The Australian analyst Craig O'Shannessy formalised this framework while working with Novak Djokovic's team from 2026. He called it the "first four shots" — serve, return, the server's plus-one, and the returner's plus-one. The core conclusion is simple and deeply uncomfortable for anyone who loves long rallies: the majority of points at Grand Slam level end inside the first four shots.

Which means the bulk of the match — everything after the fourth shot — accounts for only a small share of total points. And that small share is the part most replayed, most discussed, and most remembered. That is the first paradox anyone working with tennis data must accept: the most beautiful thing is not the most decisive thing.

The evidence chain: Four shots, the plus-one forehand, and the price of the second-serve return

Every tennis point has a short causal chain. The server controls shots one and three. The returner controls shots two and four. If the point ends inside four shots, the outcome is effectively settled by each side's first two strikes. If the point stretches beyond four shots, the emphasis shifts to fitness, movement, and the ability to absorb pressure late in the rally.

This is where the first hidden number appears. Based on my experience tracking matches in Sydney and Melbourne across several seasons, I log point rhythm game by game, not just match by match. What I found was not a simple trend but a two-tier structure. Games in which the server wins inside four shots have a markedly higher hold probability. Games in which the server is dragged into a fifth shot or beyond have a markedly higher break probability — regardless of how good that player's first-serve percentage looks on the night.

Put differently, first-serve percentage does not tell you who is controlling the match. It only tells you who is serving well in that moment. Those are two entirely different things.

The second hidden number is the plus-one forehand. In elite men's tennis, the forehand that follows the serve is the single highest-value shot in the entire shot system, and its value does not come from ball speed. It comes from position. A player who unleashes a plus-one forehand from the middle of the court, after a serve that pushed the opponent out of position, manufactures a point with a very high win probability without needing a single long rally. That is why top players spend hundreds of hours a year rehearsing that second shot, while crowds remember only the long rallies in the fifth set.

The third hidden number, and the most misunderstood of the three, is the second-serve return. The returner's win rate on second-serve points is one of the strongest predictors of match outcome at Grand Slam level, yet it almost never appears on a broadcast stat sheet. The reason is practical: it is hard to measure. It depends on where the returner stands, which zone the return lands in, and whether the server is forced to hit the third shot from an unfavourable position.

I remember a match in Melbourne. The player who lost had a first-serve percentage roughly seven points higher than his opponent, a higher first-serve points-won rate, and nearly double the aces. Reading the post-match sheet, he won. But he lost. The reason was that he won only about a third of his second-serve points, while his opponent won more than half of his. The opponent's second-serve return pushed him into the third shot from deep behind the baseline, and from there every pretty number above became meaningless.

That was when I began to understand the real meaning of the line I use often in my analysis: numbers never lie, but they can stay silent. The post-match sheet is not wrong. It is simply silent about precisely the thing that decided the match.

Assemble those pieces and a complete evidence chain appears.

Point rhythm is a structural metric, not an outcome metric. It describes how a player builds a point, not what happens to that point. That makes it more stable and far more predictive than outcome metrics such as first-serve points won.

At the next layer, point rhythm shifts with the surface, and shifts predictably. On fast hard courts, point rhythm is visibly shorter than on clay. That sounds obvious, but the consequence is not obvious at all. A player can post excellent numbers on hard court and terrible numbers on clay without changing a single technical element, purely because the structure of points changes and exposes weaknesses that were previously concealed.

Less discussed is the cyclical behaviour of point rhythm within a single match. It is not fixed. In the opening games, rhythm tends to be shorter because both players are still finding the ball. In the decisive games of a set, rhythm tends to be longer because both sides accept less risk, return more safely, and wait for the error. This is where data analysis struggles most, because it demands splitting the data by match phase rather than aggregating across the whole match.

And at the deepest layer there is a genuine tension between two schools. The four-shot school argues for ending points as early as possible, attacking with the serve and the plus-one forehand. The defensive school argues for stretching points, turning the match into a contest of fitness and patience, and letting the opponent miss. Both are correct — but correct for different groups of players, at different career stages, on different surfaces.

My mistake was applying a single model to all of them. I once burned my own model on Croatia. That was the day I learned to listen to the data. The lesson transfers directly to tennis: a model built on one cohort of players collapses the moment it meets a cohort with a completely different point structure.

Take a historical example. The 2026 Wimbledon final between Novak Djokovic and Roger Federer is one of the most analysed matches in tennis history, and also one of the clearest demonstrations of the limits of any data model. Federer won 218 points across the match; Djokovic won 204. Federer won more games. Federer had the clearer periods of control in the second set. Yet Djokovic won the match, after saving two championship points in the sixteenth game of the fifth set.

Look only at total points and you conclude the result was statistical noise. Look at the structure of points in the decisive games and you see the opposite. At the most important moments, Djokovic controlled shots two and four — the return and the plus-one. Federer played better for most of the match, but in the decisive moments he hit more shots and let his opponent control more of them.

This is precisely where the hidden number earns its keep. In decisive games, point rhythm stops being descriptive and becomes predictive. And in most of the models I have built, it was ignored entirely.

If you want to verify this yourself, the method is not complicated. Pick a match, log the number of shots in every point, and log the score at the moment that point was played. Then split the data into two groups: points played when the score was level or within one game, and points played once the set was settled. The pattern shows up immediately. In the first group, point rhythm correlates strongly with outcome. In the second, the correlation all but vanishes. That is why whole-match stat sheets are often useless for explaining close matches.

I have spent years telling colleagues in the data room that we need to stop measuring matches with what is easy to measure and start measuring with what matters. Point rhythm is the perfect example. It is not so hard that it cannot be collected — it needs one person logging it, or a good enough video-recognition system. But it sits in none of the standard data packages broadcasters buy, because it does not produce a pretty on-screen graphic.

The final piece of the evidence chain is this. Point rhythm is not a single metric; it is a system. It links the serve to court position, court position to shot selection, shot selection to psychological pressure at the big points, and psychological pressure to the physical capacity to sustain that structure across three or five sets. When one link weakens, the whole system weakens, and the post-match stat sheet will not tell you which link broke.

The counterintuitive angle: Correlation is not causation

Here I have to say something I do not enjoy hearing.

Short point rhythm correlates with a high win rate. That does not mean playing short points produces victories. This is the most common reasoning error in data-driven sports analysis, and it has damaged more of my models than any technical flaw.

The problem lies in the direction of causation. Players strong enough to end points early tend to be the best players. Short point rhythm is a consequence of quality, not its cause. If a lower-ranked player deliberately speeds up to imitate the point structure of a top player, the usual result is the opposite — more unforced errors, fewer winners, and a match that ends faster in the wrong direction.

There is also a denominator problem. Average point rhythm varies enormously with the opponent. A player can post very short rhythm against a defensive specialist and very long rhythm against an attacking one. Folding an entire season into a single average produces a meaningless index. I call this the average of things that cannot be averaged, and it appears in almost every tennis model I have seen.

What the data cannot say

There is one thing point rhythm will never measure, and I need to state it before closing.

The metric cannot capture the silence of a stadium before a serve on break point. It cannot capture an eighteen-year-old playing his first Grand Slam fourth round, feeling the ball disappear from his hand for three seconds. Those things live in no dataset, and they still decide matches.

Every shot leaves a footprint. The best are not those who run the most, but those who leave footprints in the right places. But a footprint does not tell you how the person felt when they put their foot down.

Signals for the next round

Over the coming rounds I will track a single signal: second-serve points won when the score is level. If that rate drops for a given player across two consecutive matches, it is an earlier warning than any other metric that his match structure is wobbling — before the scoreboard shows it.

Error log

In the match I analysed last week, I predicted a player would win with seventy percent probability based on short point rhythm. He lost in four sets. My model was right about the metric and wrong about the human. I record that here for one reason only: an analyst who does not publish his mistakes is not an analyst, but a prediction salesman.