Empty Cells and the Discipline of a Badminton Analyst

My right knee aches whenever Guangzhou turns cold, and on the night of January 9, 2026, it ached louder than usual. I had been at the screen since eleven, three files open side by side: a BWF World Tour results sheet, a workload tracker for eight players I was keeping an eye on, and a third sheet. The third sheet was completely empty. No tournament name. No players. No scores. Not one row to start from.

In fourteen years of working as an analyst, I have learned that an empty sheet is not a technical glitch. It is a result. And that result, read correctly, says a great deal about where badminton data is generated and who gets left at the edge of it.

The knee pain taught me to count, and I have never stopped counting.

My work runs on a nine-layer process. The first layer is technical and tactical: playing style, how a player sets up a point, physical cost, smash speed, rally length, net-point win rate. The second is form and player data: recent results, result quality, schedule density, ranking points, head-to-head. The third is the tournament system, graded on the Super 1000, 750, 500, 300 and 100 scale. The next four cover the world landscape, rules and institutions, the coaching team and support system, and the risk surface. The last two are narrative, expectations, and the transmission chain into the wider industry.

Every layer has a valve. If the first layer returns nothing, the other seven return nothing with it. Not because they have nothing to say, but because they have nothing to hold on to. A form analysis that does not know which tournament a player just contested, at what density, against whom, is just prose wearing a data jersey.

On January 9, that is exactly what happened. The source I received had no tournament name, no players, no technical metrics, no form data. The third sheet was empty, and it emptied the other seven layers with it.

There are two ways to handle an empty sheet. The first is to fill it with guesswork: build a plausible-sounding story, drop in some terminology, add a few strong opinions so the piece carries weight. The second is to record the empty cell exactly as it is, with the reason, and accept a shorter, slower piece that gets shared less. I chose the second, not out of any lofty professional ethic, but because I once tried the first and the market taught me an expensive lesson.

In 2026, fresh out of competitive sport after a knee injury and starting to collaborate with a data-analysis blog in Guangzhou, I used expected goals to dissect Eran Zahavi's form in the Guangzhou R&F shirt. He scored 27 goals in the Chinese top flight, but his season-long expected-goals figure was only 21.5. A gap of 5.5 goals sits outside the sustainable range. I published a forecast that he would settle around 20 goals the following season, and was laughed at. In 2026, he scored exactly 20.

The lesson was not that I got it right. The lesson was that the chain of evidence has to close before I open my mouth. If Zahavi's expected-goals figure that year had been 26.8 instead of 21.5, I would have had nothing to write. Staying quiet when there is nothing to write is part of the job, not a failure of it.

In badminton, the chain of evidence is far thinner than in football, and that is the first thing I have to count. A Super 1000 match can be logged with average rally length, unforced-error rate, net-point win rate, top smash speed, and distance covered per game. A Super 100 match, or a qualifying-round match, sometimes leaves only a score and a duration. Taking data from the lower tier and comparing it with the top tier means mixing two different frames of reference and calling it analysis.

So in every pre-match note I name exactly three metrics. Three numbers, no more. Each one needs a verification checkpoint; if I cannot check my own figure again the following weekend, that figure does not belong in the piece.

Which three depends on the pairing. In a match where one side controls the net, the deciding metrics are net-point win rate and average rally length: whoever stretches the rallies is winning the war of stamina. In a match where one side leans on attack, the deciding metrics are the unforced-error rate in the final two games and smash speed across the first fifteen points.

The boundary of a number matters as much as the number itself. On the night of June 27, 2026, before South Korea met Germany in Kazan, I went back through Germany's pressing data and saw their back line leaving space behind it in every group-stage match. I wrote a note predicting South Korea would win 2-0, while the bookmakers had them at 10.0. Kim Young-gwon and Son Heung-min scored. The piece reached more than two hundred thousand reads.

On the night South Korea beat Germany, I looked at the screen and saw every probability lying, including the ones I had just used to be right.

In May 2026, when the Bundesliga returned during the pandemic, I tracked 81 matches without crowds and recorded that home teams won only 28 percent, against 44 percent before the shutdown. Home advantage all but evaporated, and my betting model fell apart. A programmer colleague pushed me to publish immediately. I waited two more rounds. In June, my prediction streak returned 32 percent profit.

On December 9, 2026, the World Cup quarter-final between Brazil and Croatia. Brazil generated 2.3 expected goals against Croatia's 1.2 and led in extra time. I put my faith in the model and predicted Brazil would advance. Goalkeeper Dominik Livakovic made eight saves, two of them in the shootout. Brazil went home, and I lost a large sum. Expected goals do not measure resilience.

Since then, every piece I write on a knockout match carries a line on goalkeeper save quality, and a line spelling out the risks the model has not priced in.

There is a trap here that data people fall into, and I have fallen into it. It is turning silence into a moral posture.

The empty sheet on January 9 could be evidence that the source was not good enough. It could also be evidence that I was lazy. Those two possibilities look identical on a screen, and there is only one way to tell them apart: how many calls I made, how many databases I opened, how long I waited before concluding the sheet was empty. Having no data and not having gone looking for data are fundamentally different states, yet they produce the same result on a spreadsheet.

At the same time, some things never fit inside a cell. When the stands are empty, I understand that data also needs noise to exist. A packed arena in Jakarta or Birmingham changes how a player picks the landing point in the closing points, changes even the breathing between two rallies. No column in my spreadsheet records that.

My job is not to strip the noise out. My job is to sort it: which noise is interference to discard, and which noise is an environmental variable to keep.

The third sheet was still empty when I shut the machine down at nearly three in the morning. I left it that way, saved it with the date, and wrote one note: three data layers need filling before anything can be said about any player in this cycle.

The transfer market is just a spreadsheet that learned to wear a shirt.

The next cycle will answer a single question: whether the badminton data chain keeps extending down to the Super 300 tier and the qualifying rounds. If it does, analysts will have more ground to stand on. If it does not, we will keep getting better at describing the top eight players and keep knowing nothing about the rest of the badminton world.