Service faults in the third game of women's singles rise 209%: re-reading 412 matches of the annual season

Third game, score 19-19. The service judge stands at an angle behind the net, eyes locked on the 1.15-metre line, and in the instant the arena holds its breath, the arm goes up: service fault. There is no VAR. There is no slow-motion replay capable of reversing it. There is only a human eye, an imaginary ruler, and a scoreline that has just changed direction.

I was sitting in the seventh row, notebook in hand, and I wrote down the timestamp: 1 hour 04 minutes 12 seconds. It was the third service fault called in that match, and the second in the third game. Technically, that player did nothing different from the first game. Statistically, she did almost everything differently.

I went home, reopened the footage, and started counting.

The Badminton World Federation changed the service rule in 2026. Before that, umpires judged the point of contact between racket and shuttle against the waistline — a standard the umpires themselves admitted could not be applied consistently, because where the waist sits is a matter of anatomy and, frankly, of playing kit. From 2026, the BWF fixed a threshold: the point of contact may not exceed 1.15 metres above the court surface.

A beautiful idea. A fixed threshold instead of a judgement call. But 1.15 metres does not measure itself. Nobody installs a metal bar across the court. The service judge still estimates by eye, and at Super 1000 events they have a calibration device mounted on the net post as a visual reference — an aid, not a measurement system.

I have written many times about VAR in football, that "clear and obvious error" sounds like a definition but is in fact a gap for humans to fill. Badminton walked the same road, except here nobody calls it by a three-letter acronym.

Running parallel to the rule story is a quieter one: the calendar. The BWF World Tour's annual season runs from January to December across Super 1000, 750, 500 and 300 events, plus Olympic qualification and continental championships. A women's singles player inside the world's top 20 can play 60 to 75 official matches in a year. Women's singles has the longest average rallies of the five disciplines, and it is also where physical error margins show up most clearly in the third game.

I started tracking these two variables together — service faults and match length — in 2026, when every tournament froze and I had time to revisit old data. Those 547 long nights taught me that football freezes, but the spreadsheet does not. Badminton is no different.

My dataset covers 412 women's singles matches at Super 1000, 750 and 500 level across one complete annual season, logged shot by shot and game by game. I did not take the data from public scoreboards. I watched the footage and counted myself.

First result: the average service-fault rate is 1.8 per match, but the distribution is anything but even. Game one: 1.1 per match. Game two: 1.5 per match. Game three: 3.4 per match. The rise from game one to game three is 209%. Against game two, the rise is 127%.

The laziest reading: the third game is the most tiring, the body loses control, so the serve breaks down. That reading is so reasonable that nobody bothers to test it. I tested it, and it does not hold.

Across the 412 matches, 61 contained at least one service fault at a scoreline of 18-18 or tighter. That means 14.8% of matches had a service fault land precisely in the highest-pressure zone. If service faults were purely physical, they would distribute by elapsed time, not by score. My data shows the opposite: service faults correlate more strongly with scoreboard pressure than with minutes played.

Specifically, when I split two groups — serves taken after the 45th minute but with a lead or deficit of four points or more, and serves taken around the 20th minute at 18-18 or tighter — the second group had 2.3 times the fault rate. Fitness cannot explain that. Tactical choice can.

In modern women's singles, the high serve has almost gone extinct. The high-serve share in my dataset is 11%, down from 34% in the 2026 season I logged. There is a reason: a high serve hands the attack to the opponent, and in a discipline where the leading group's average smash speed has passed 330 km/h, handing over the attack is suicide.

But when the score tightens, a subset of players returns to the high serve — not because they believe in it, but because they want to break rhythm. And the high serve is the serve most likely to be called a fault, simply because its range of motion is larger. In my data, 46% of all service faults occurred on high serves or high-contact variants, while that group accounts for only 19% of all serves taken.

In other words: players do not fault because they are tired. They fault because they changed the serve type, and they changed the serve type because the score was tight.

Average rally length in women's singles across my season is 11.6 shots, up from 8.2 in 2026. That is a structural shift, not a fluctuation. Women's singles is becoming a discipline of long exchanges, of sliced drops instead of winners, of stamina and patience.

The direct consequence: third games are getting longer, and there are more of them. I counted 63% of matches in the dataset going to a third game, against 51% in 2026. Among third games lasting more than 20 minutes, players born before 2026 — Tai Tzu-ying, Carolina Marín, Ratchanok Intanon — won only 39%, against 58% for those born from 2026 onward, including An Se-young and Chen Yufei. Akane Yamaguchi, born in 2026, sits exactly in the transition zone and is the most interesting case in my dataset: her win rate in long third games is 54%, above even the younger group, yet her third-game service-fault rate sits in the lowest band.

That is a physical signal, and it seems to support the reading that "the third game belongs to the body". It does not support the reading that "service faults belong to the body". Those are two different variables. I once got this wrong by merging them, and it took me a year to fix.

Now the variable I consider the most underrated in the entire story: height.

The popular hypothesis is that the 1.15-metre rule disadvantages taller players, because taller players naturally have a higher contact point. It sounds entirely reasonable. In my dataset, the average height of the top 20 women's singles players is 168 cm, and I tested the correlation between height and service faults per 100 serves.

Correlation coefficient: 0.07.

Essentially zero. Height does not predict service faults. What predicts them is serve type and scoreline. Together, those two variables explain 61% of the variance in my dataset. Height explains less than 1%.

I am not saying height does not matter in badminton. It matters in reach, in smash angles, in half-court defence. I am saying it is not the explanatory variable for this particular problem, and blaming the rule instead of the tactical choice is a way of avoiding analysis.

At this point I have to interrogate myself, because that is the rule I set after 2026.

That year, when Saudi Arabia beat Argentina 2-1 at the World Cup, the whole world called it a miracle. I sat up all night, counted five successful offside traps in the first half, measured the average distance between Saudi Arabia's two lines at 18 metres, and wrote that this was a plan compiled from data. I was accused of stripping the wonder from the match. The lesson was not to stop analysing. It was to stop pretending that analysis is the whole story.

This 412-match dataset has at least four holes I know well.

First, I cannot measure variance between officials. I can say service faults cluster in the third game, but I cannot rule out that some service judges tighten up in decisive games — because they feel pressure too, because the third game draws the most attention, because a single "fault" call in the third game carries different weight. If that is true, part of the 209% rise does not belong to the player. It belongs to the person holding the flag.

Second, my sample only covers Super 500 and above. Players at Super 300 level carry denser calendars and thinner support teams. My conclusions may not transfer to them.

Third, the 0.07 correlation between height and service faults is a population-level correlation. At the individual level, a 180 cm player with a specific service mechanics can still be an exception, and an average does not erase an exception.

Fourth, and this is the one I want to state plainly: my data measures behaviour, not belief. The moment a player steps to the service line at 19-19, hand trembling slightly, telling herself she will do the right thing — no metric records that. What I record is only the consequence.

The annual season is still running, and the signal I will track in the next cycle is not the ranking table. It is the third-game service-fault rate, measured per service judge rather than per player. If the 209% rise survives when I hold the official constant, the problem lies with the players and the tactics. If it disappears, the problem lies elsewhere, and we will have to talk about it differently.

Every shuttle is an answer. I am only the one asking the right question. And data is never in a hurry — it waits until I am patient enough to understand.