Nine Data Dimensions and a Single N/A: The Limits of Badminton Analytics

A nine-tab spreadsheet sat on the screen at 9:40 on a Tuesday morning, and all nine tabs returned the same line: insufficient information to assess. Across the table, the client — a group preparing a dossier on a young player — drummed his fingers on the desk. He had paid for a report and received an empty grid. It took me four minutes to decide not to fill in a single cell.

Those four minutes were more uncomfortable than any meeting I have sat through. My job is to fill gaps, and a single N/A repeated nine times looks more like a confession than a professional conclusion. But the nine-dimension model I use for badminton — technical and tactical, player form and data, tournament system, world landscape, rules and institutions, coaching and support, risk surface, public narrative, industry transmission — did exactly what it was designed to do: it reported that the input material does not exist.

It began with an ordinary request. A young player, a handful of recent matches nobody broadcast in full, no point-by-point record, no tracking data. We had the tournament name, the results, and a few phone clips the player himself had sent. That is enough to write a news story. It is not enough to build a model.

In football this situation almost never arises. A second-tier match in Europe still has three independent event-data providers, expected goals, and running maps. Badminton is different. At Super 1000 level the data is reasonably complete: rally length, smash speed, winners, errors. Drop to Super 300 or Super 100 and the statistical sheet thins out until only the score remains. And the score is the worst possible analytical input, because it compresses everything into two numbers that explain nothing.

The nine-dimension grid was born after the summer of 2026, when I praised the pressing of a Shanghai side after a 4–0 win without checking the pressure index. Three days later they lost to the bottom club. My editor summoned me and said the line I still use as a checklist: you looked at the score and not at the structure. Since then every analysis of mine has to pass through nine gates, and any gate without data must be labelled as having no data.

Applied to badminton, that grid exposed what I consider the most valuable insight in this trade: most of what the market calls badminton data is really the tournament organiser's data, not the sport's data. People measure what is easy to measure, where it is easy to measure, in each event's own way, then stitch it into a time series that looks continuous.

Start at the lowest layer of the model: the tournament system. The BWF World Tour is tiered into Super 1000, 750, 500, 300 and 100, plus the World Tour Finals. Under the BWF points table, winning a Super 1000 event is worth 12,000 points; winning a Super 100 event is worth 5,500. The second figure is only 46 percent of the first. Yet reaching the semi-final of a Super 1000 is already 8,400 points — 2,900 more than a Super 100 title.

That produces a very concrete competitive logic. A player outside the top 20 has two routes: grind four or five Super 100 and Super 300 events across Asia in six weeks, or enter two Super 750s and risk a first-round exit. The first route is safer for points, but it teaches nothing about how to play the top ten. The ranking rises; the level stands still.

I tested that hypothesis against several seasons of schedule data and found a recurring pattern: the fastest climbers typically play one and a half to two times as many matches at the 100–300 level as at the 750–1000 level, while the group that holds a top-eight place shows the reverse ratio. That is a correlation, I know, and I will come back to it.

There is a second problem in the same layer: calendar density. A player eligible for the World Tour Finals must fulfil a minimum number of events in the system, plus national-team obligations. Add it up and some seasons contain eighteen to twenty genuine competition weeks, before counting travel between three continents. Nobody publishes that figure as a workload metric. It exists, everyone knows it, and it appears in no official statistical table.

The second layer: technical data. There is no optical whole-court tracking as in football, so there is no equivalent of expected goals. What exists is rally length, smash speed, winners and unforced errors. All four depend on how each organiser records them. Radar placement directly affects measured smash speed; for the same stroke, the gap between two events can exceed ten percent. Comparing two players' smash speeds across two events means comparing two different measurements and giving them one name.

Because there is no PPDA, I had to build substitute indices. The two I use most: the share of points won inside the first four shots, and the average number of shots needed to end a rally when the opponent serves short. The first measures the ability to seize initiative straight from the serve; the second measures how far a player is dragged into the opponent's rhythm. They are the closest translation of pressure into badminton language.

They still cannot measure what I need most. In eighteen years of watching, I have learned that one data layer never appears on a spreadsheet: breathing rhythm, footwork, competitive state. In Bangkok in 2026, with the stands nearly empty because of the pandemic, I sat in an office reviewing more than a hundred old matches and realised that footwork announces the outcome of a rally three or four shots early. A player starts taking shorter, more lateral steps — a sign that fitness or nerve is leaking, and it appears before the score turns.

When the stands are empty, I hear the sound of pressing footsteps most clearly under the dark of a pandemic night.

The third layer: individual players. Carolina Marín is the example I keep returning to when teaching the limits of probabilistic models. She won the world title three times, in 2026, 2026 and 2026, and Olympic gold at Rio 2026. Then came three anterior cruciate ligament injuries: 2026, 2026, and in the Paris 2026 Olympic semi-final, while leading and a few points from the final.

No model of ours forecast that sequence, because the variable behind it is not in the table. The model has age, matches played, distance covered, schedule. It does not have ligament durability, microscopic injury density, or how one particular body tolerates load across years. Forcing the data into a frame would produce a meaningless sentence like injuries are an occupational risk — true, and silent about this player.

Shi Yuqi winning the 2026 World Championships in Paris is the opposite failure. For several seasons his numbers were excellent while his results in major knockout matches did not match, and analysts read that as a psychological problem. Once he won, the same set of numbers was suddenly interpreted as big-match temperament. The numbers did not change; the story did.

That is why every report of mine carries two separate columns: hypothesis and evidence. When the data diverges from the hypothesis, I keep the divergence intact and name it a blank spot instead of rounding it into shape.

The fourth layer: coaching and support systems. This is where N/A appears most often, and where clients push hardest. There is no public database of how many full-time strength and conditioning coaches a programme employs, which federation keeps a full-time video analyst, who uses load-monitoring devices. Some differences I know for certain because I have stood in the gyms, but knowing for certain is not the same as being able to cite.

So I write it explicitly: this is direct observation, not verifiable data, and it should not enter a comparative model. A report may contain observation; a model should not.

The fifth layer: risk surface. In badminton, injury risk is higher than in most team sports, because players compete alone on court, accelerating and changing direction continuously with nobody to share the load. It is also the surface with the worst public data. Nobody publishes training volume, nobody publishes rest days between events, nobody publishes detailed injury history. Every risk model of ours therefore runs on half the information, and I say so to clients before presenting any probability.

The sixth layer: public narrative. This is the only layer where badminton generates data faster than anyone can read it. After the Paris 2026 Olympics, An Se-young publicly criticised how the national-team system operates, and that story travelled faster than any technical analysis of her. For a data monk, this is the moment to encode emotion as behavioural indicators rather than commentary: fewer events entered, longer gaps between appearances, denser public statements, changed national-team training schedules.

Those four indicators are measurable. Whether she is happy is not, and I do not put it in the report.

The summer of 2026 was the most expensive tuition I ever paid to learn that clean data cannot rescue a dirty hypothesis.

This is the hardest section for me to write, because it argues against my own trade.

Everything above rests on correlation. Players who grind small events climb faster — correlation. A higher share of points won in the first four shots among the top eight — correlation. But causation may run entirely the other way: players do not become strong by winning early; they win early because they are already stronger. And badminton has a rarely stated paradox: strong players often deliberately extend rallies, because long rallies drain opponents physically and open points in the second half. If so, a fast-win index may signal a lack of control rather than an advantage.

The second trap is more dangerous: filling blanks with estimated numbers and calling them data. When a cell says N/A, somebody always proposes interpolation, inference from a similar event, or an industry average. All of it produces a prettier and more wrong table. Systems do not collapse overnight; they crack from the moment I stopped questioning the foundation.

Moscow 2026 taught me the same lesson in another sport. I predicted Croatia would win on a clear expected-goals edge; the match ended level after 120 minutes and Croatia won on penalties. My model had no variable for fitness after minute 90 and none for home advantage. That Moscow night I did not watch a football match; I watched raw data laugh in the face of every probability.

The lesson was not to abandon data, but to place its limits inside the report, somewhere impossible to skip. A short section called data collection method — the one colleagues complained slowed us down — is in fact the only thing that makes the rest credible.

The next round of this story will not be decided by who has more numbers, but by who dares to publish how the numbers were obtained. When a tournament openly defines its metrics and measurement positions, the whole industry gains. When nobody does, every analytical table is just an opinion set in a nicer typeface.

I still keep that nine-tab sheet in a separate folder. Now and then I open it, look at the nine N/As, and remind myself: the hardest part of this trade is not finding the answer, but recognising that the raw material for an answer is not yet here.