The Empty Scoresheet: The Limits of Badminton Data

The Empty Scoresheet and the Limits of Badminton Data

An A4 sheet of paper sits on a coach's desk in Hanoi in the fading light. It carries exactly four digits: 21-18, 21-19. No smash speed, no count of rallies past twenty strokes, no winning rate on decisive rallies, no breakdown between unforced errors and errors forced under pressure. Thirty minutes earlier the match ended, and with the final applause almost all of its data ended too.

I keep hundreds of such sheets in a drawer. Some are handwritten in Vietnamese, some scrawled in English in an arena in Jakarta, some hold nothing but a date and two players' names. For forty-three years people have called these match records. I call them evidence of a collective memory loss.

Why I stay behind after every match

I entered the profession from a broadcast studio, not a machine room. In 2026 I hosted many major events, including the Table Tennis World Cup and the Sudirman Cup. Badminton reached me through a microphone, when I had to say something meaningful during the sixty-second interval between games. That was the first lesson: without data, a commentator is left with adjectives.

Years later I moved into sports betting analysis, and my job became a constant commute between two worlds. On one side is football with its dense warehouse: expected goals, PPDA measuring how many opponent passes are allowed before each defensive action, heat maps, transfer valuations. On the other side is badminton, a sport that broadcasts thousands of hours a year but leaves behind very little that can be counted.

In 2026, when the entire tournament calendar stopped, I sat at home rewatching hundreds of matches and learned Python just to run a few correlation models. That work taught me two things. First, baseline data — the weak signals nobody prints on a scoreboard — is where matches are actually decided. Second, badminton has a data gap far larger than fans imagine.

A sport without an equivalent yardstick

Compare two situations. After ninety minutes of football, a viewer can look up a team's expected goals, shots inside the box, duels won, distance covered, and even the passing sequence that led to a goal. Those metrics come from commercial data collected at every match in the top leagues.

After twenty-one points of badminton, a viewer has the score. If the match belongs to the top tier of the BWF World Tour, there are a few basic numbers on the online results page: points won on serve, longest rally, highest smash speed of the match. Outside that tier, the numbers are close to zero.

Since 2026, electronic line-calling has been used at the top events, mainly to decide whether a shuttle landed in or out. That system generates an extremely valuable body of positional data: shuttle trajectory, landing point, the distance a player must cover between two strokes. But that data sits with organisers and technology partners, almost never opened to the public. The result is a sport measured very carefully at the operational level and almost not at all at the public level.

This creates a paradox familiar to anyone in analysis. A scoreboard tells us who won. It does not tell us why. And in most cases, why is the only question with predictive value.

The data is not wrong; I simply forgot to ask where it stands.

First evidence chain: rally length is the cheapest baseline data

If I had to choose one substitute for badminton's missing metric, I would choose rally length.

Rally length is the number of times the shuttle crosses the net from serve to the point's end. It has three advantages most other badminton metrics lack. It is free, because any viewer can count it. It is objective, because it does not depend on a subjective judgment of whether a stroke was good or bad. And it directly reflects the tactical intent of both sides.

A player who wants to extend rallies will lift the shuttle high, deep and close to the sideline, forcing the opponent to move along the length of the court repeatedly. A player who wants to finish early will try to press in the front half, accepting more risk on each stroke. When a player's rally distribution shifts right, toward shorter rallies, it signals an attempt to shorten the match. When it shifts left, they are stalling, dragging the opponent's stamina down.

Under the current rally-point system, every point is worth the same. That makes rally length a strategic variable rather than an aesthetic one. A player who wins twenty-one points through fifteen short rallies and six long ones is playing a completely different match from one who wins the same score through nine short and twelve long rallies. The scoreboard cannot tell them apart. Rally length can.

I usually record three figures from rally length for each player: median rally, the share of rallies past fifteen strokes, and the longest run of consecutive points. The median reveals the underlying style. The long-rally share reveals endurance. The consecutive-point run reveals the capacity for bursts — which in badminton does far more damage than a pretty smash.

The limits must be stated clearly. Rally length cannot distinguish a long rally because both players are playing solidly from a long rally because both are exhausted and merely pushing the shuttle over. To distinguish, you must rewatch the footage and cross-reference the moment in the match. This is why I never draw a conclusion from a single number. A measurement only means something when we know where it stands in the flow of the match.

Second evidence chain: the smash-speed trap

No metric is loved by badminton media more than smash speed.

Whenever a major event takes place, a number appears in reports: the fastest stroke of the tournament, usually around four hundred to four hundred and twenty kilometres per hour in singles and higher in doubles. Under laboratory conditions, some records have been published above four hundred and ninety kilometres per hour. Those numbers are impressive, and entirely real.

The problem lies elsewhere: the link between peak smash speed and match outcome is very weak.

There are three reasons. First, the measured speed is the speed off the racket; the shuttle then decelerates rapidly through the air. A four-hundred-kilometre smash may reach the opponent slower than a three-hundred-and-seventy-kilometre smash aimed at an awkward position. Second, smash speed counts only technically successful strokes; it does not count misses, and at the elite level, missing is a far larger source of lost points than being beaten. Third, the fastest smash of a match usually appears at a few emotional peaks, not on the decisive rallies.

A number removed from its context is only a lie that has been polished.

If I wanted to measure a player's attacking strength with data that genuinely predicts, I would choose three other things. One is the share of points won on rallies requiring a second smash — the ability to finish after the first stroke has been blocked. Two is the share of points won after the opponent has pushed the shuttle deep to the sideline — the ability to attack from a pushed-back position. Three is the share of unforced errors in the third game, when stamina has run out. None of these three are published, which is precisely why we talk so much about smash speed: it is the only thing available.

Reading the breathing of four kinds of champion

At Paris 2026 I watched four finals and took notes on the same template. The results showed four very different ways of winning, even though every scoreline was identical in one respect: the winner took it in two games.

Viktor Axelsen of Denmark defended his men's singles gold with a style reducible to one phrase: controlling height. He almost never allowed his opponent a rally at mid-height — the band where an attacking player can swing freely. Every time the opponent lifted, the shuttle went deep or was forced down, pushing the next stroke into a difficult zone. His median rally in those matches sat low relative to the men's singles average, not because he plays fast, but because he ends rallies before they can lengthen.

Kunlavut Vitidsarn of Thailand reached the final by the opposite route. He accepted long rallies, defended from the back half, and waited for errors. This style produces few highlight reels, but it drains opponents efficiently. Its weakness: against an opponent capable of finishing from mid-height, the defender never gets the chance to extend the rally. The Paris men's singles final illustrated that limit clearly.

An Se-young of South Korea won women's singles gold with what I call court coverage. She moves in a nearly constant pattern: short steps, low centre of gravity, always standing where the opponent's next stroke becomes predictable. Notably, she does not win by accelerating. She wins by reducing her opponent's options to very few. When a player has two options instead of four, their error rate rises without them usually noticing.

In men's doubles, Lee Yang and Wang Chi-lin of Chinese Taipei defended their gold. Their style rests on a simple principle: always create rallies in which both partners can participate. In doubles, the most important baseline metric is the share of rallies in which one partner is forced to stand still. This pair kept that share very low, which is why their stroke quality held through the match.

Four kinds of champion, four uses of rally length. Every scoreline looked like a straight-games win. Read only the score, and you would conclude they were all the same.

The contrarian angle: correlation is not causation

There is one mistake I made and took years to correct.

At the 2026 World Cup I analysed the entire group stage with expected goals and concluded Croatia would lose the final because their figure was lower. I ignored two things: the alternation of pressure over time, and set-piece situations outside the model. Croatia reached the final. Since then, every analysis I write includes a mandatory section stating which metric is being overlooked and under what conditions it will diverge from reality.

In badminton the same trap appears in another form. People observe that champion players often have a high long-rally win rate, then conclude that winning requires good defence. But in most cases causation runs the other way: players can extend rallies because they are already ahead, forcing the opponent to take risks. The long-rally sequence is a result of advantage, not its cause.

Another example is sample-size error. Three matches say nothing about a player. Thirty matches are still not enough if all thirty took place under the same conditions, the same shuttle type, the same arena. Badminton is unusually sensitive to physical conditions: humidity directly affects shuttle trajectory, indoor draughts affect the accuracy of high strokes, and shuttle types differ between events. Data collected in two different arenas cannot be placed side by side without adjustment.

I once thought data was truth, until the 2026 World Cup taught me to be afraid.

There is another layer rarely discussed. Most detailed badminton data today — especially landing-point and player-position data — is collected mainly for two purposes: officiating and supply to betting companies. This is the darkest side effect of sports digitisation. A sport that does not invest in opening data to fans, journalists and youth academies nevertheless invests very well in selling data to parties that profit from uncertainty of outcome. The consequence is that the people who follow the sport most seriously are given the least information.

The sports media industry is repeating an old mistake at another level. Streaming platforms outbid each other for rights on the assumption that viewers will pay to watch more. But when rights prices hit a ceiling while viewing time per person stops growing, the gap is covered by selling detailed data to third parties. Badminton, with its dense calendar and loyal but modest audience, is more exposed to that model than most.

Institutions: three changes reshaping the sport

Three rule and organisational changes directly affect the data we can observe.

First is the fixed-height service rule, with contact at one metre fifteen, applied from 2026. It removes part of the advantage of tall servers and makes the service situation less uncertain. The data consequence: serve-win rates at elite men's level have become a more stable metric than before, and when a metric stabilises it loses predictive value. What is stable no longer distinguishes one player from another.

Second is the proposal to move to a five-game, eleven-point format, which was put to a vote and failed to reach a majority. This debate is unfinished, and it matters to analysts for a very specific reason. A shorter format increases randomness of outcome, and when randomness rises, the value of every predictive model falls. If the format changes, all the historical data we are accumulating becomes hard to compare with future data — a structural break that analysts routinely underestimate.

Third is calendar density. The BWF World Tour comprises dozens of events each year, tiered from the top thousand-level events down through seven-hundred-and-fifty, five-hundred, three-hundred and one-hundred levels. Top-ranked players carry mandatory participation obligations. The consequence is that overload becomes a permanent variable, and any form analysis must begin by asking how many matches a player has played in the past three weeks.

I have often seen a player judged to be declining purely because their results worsened during a scheduling block. Their rally pattern did not change. Their error rate did not change. Only the rest minutes between matches changed. A good model must distinguish the two, and to do so it needs data that tournaments currently do not publish.

Only when the arena falls silent do I hear the baseline data whisper.

What I missed

In 2026 I spent two full weeks on a feature about a football team using a high defensive line to set offside traps. I pursued the topic so hard that I ignored everything else happening, and when that team was eliminated I realised I had missed important squad changes on the other side. Curiosity without limits becomes a systematic form of blindness.

Since then I set myself a rule: at most three hours per day on one subject, the rest for parallel tournaments. The rule slows my output but reduces what I overlook. And in every piece I keep a small section stating what I did not see, could not measure, or lacked data to conclude.

For Vietnamese badminton, that section is currently long. From Nguyen Tien Minh — once among the world's top five and a four-time Olympian — to Nguyen Thuy Linh, Le Duc Phat and the next generation, we have individuals capable of competing internationally. But we have almost no public data on them: no rally distribution, no error taxonomy, no detailed year-by-year head-to-head record. Any analysis of a Vietnamese player today rests on direct visual observation, and visual observation, however good, is a biased sample.

Signals for the next cycle

Four signals I will track in the coming cycle.

First, shifts in the rally distribution of the leading women's singles group. If attacking players begin winning more often against defensive opponents, it suggests shuttle quality and playing conditions are moving in favour of speed.

Second, the unforced-error rate in the third game of matches lasting over sixty minutes. This reflects the quality of the fitness base better than any distance-covered figure.

Third, how many matches a top-fifteen player must play across twenty-one consecutive days. When that number crosses the threshold, any form-based prediction becomes meaningless.

Fourth, whether organisers of top-tier events open landing-point data. If they do, it will be the first time the sport lets outsiders see the true structure of a match.

PPDA is only a stethoscope, but the one listening to the patient must be a monk who knows how to be silent. In badminton we do not even have the stethoscope yet. The first task is not predicting who wins the next tournament, but recording enough that next time the question can be answered with data instead of intuition.