The Empty Spreadsheet in Vietnamese Badminton: Why I Refused to Write On

At 2:40 a.m. on 12 March, I reopened my tracking file for the Vietnam International Challenge. The sheet had forty rows, one for each player I intended to include in the weekly report. The "average rally length" column was blank in thirty-eight rows. The "net-point win rate" column was blank in all forty. The "top smash speed" column held exactly two figures, and I had keyed both of them by hand from a 720p video that kept dropping frames.

Close to three in the morning, a young coach called. He asked what I thought of a player who had just come through qualifying in men's singles. I said: "I have nothing to say yet." The line went quiet for a few seconds. Then he asked again: "You're the data guy, aren't you?"

Yes. And that is exactly why I did not answer. In this trade, the cheapest answer is always the one given while the data is still empty. People call it intuition. I call it an unpaid debt.

Vietnamese badminton carries an obvious paradox. On court, we have landmarks worth remembering. Nguyen Tien Minh once reached the world's top five in men's singles, a position no one else in Southeast Asia has matched in the same discipline since. Nguyen Thuy Linh spent years as the anchor of women's singles, at one point breaking into the world's top twenty. Le Duc Phat, Nguyen Hai Dang and Vu Thi Trang are the names that have kept Vietnam's entries alive across the BWF World Tour system.

But step off the court and into the record-keeping, and the gaps appear fast. The BWF World Tour is tiered with precision: Super 1000, Super 750, Super 500, Super 300, Super 100. The Vietnam Open sits at Super 100. Each tier carries its own standards for ranking points, national entries and hosting conditions. Those figures are published openly, and I can verify them.

What is not public sits inside the match. Football has Wyscout, StatsBomb and Opta — a data ecosystem dense enough that someone sitting in Hanoi can still build a model. Badminton is different. Hawk-Eye appears at major events, but granular shot-by-shot data is barely accessible. At domestic and lower-tier tournaments, most figures have to be entered by hand. A six-hour session can generate several thousand rallies. Coding all of them by hand requires a team working in parallel, and no tournament in Vietnam pays for that.

That is why my columns are empty.

In 2026 I learned the first lesson about this. I was sixteen, writing a World Cup analysis blog. After Germany lost 0-2 to South Korea in the group stage, I had built a whole piece on possession figures, arguing that the team controlling the ball wins. The blog collected more than two hundred mocking comments. I spent the next three weeks rewatching every German match, counting each pass into the final twenty-five metres, and discovered that the metric I had used was only a shell. South Korea's PPDA in that match was just 6.8 — they pressed aggressively and with structure, never sitting deep.

The Russia World Cup shock taught me: flawed data is more dangerous than intuition. The flaw was not a missing number. The flaw was having a number that measured the wrong thing.

When I moved into badminton, I carried one rule across: no metric enters an article until I can trace its origin. Every number has a genealogy; I need to know its ancestors. A "400 km/h smash" figure in the media may come from a racket-mounted sensor, from a court radar, or from a measurement taken at the instant the racket meets the shuttle — three different ancestors, three different results, and no legitimate cross-comparison. I once saw two articles in the same week report smash speeds 40 km/h apart for the same rally. Both cited "statistics".

The first tier of the index system I built for badminton is rally structure. This replaces what football calls possession. In badminton, rally length does not tell you who is winning, but it tells you who is setting the rules. I split rallies into four bands: under 6 seconds, 6-12 seconds, 12-20 seconds, and over 20 seconds. A player whose distribution leans toward the short band usually chooses early termination — smashes, drop shots, attacking the flanks. A player whose distribution spreads across the two long bands usually plays control, moves the opponent, and accepts living inside long rallies.

What interests me most is not the distribution itself but the win rate within each band. I call it the strength curve. A player might win 70% of short rallies but only 35% of long ones — meaning the opponent only has to extend the rally to find the key. Conversely, some players win evenly across every band, and that is the mark of a complete structure. In the data I collected from Nguyen Thuy Linh's international matches, her curve spread fairly evenly, but the strongest node sat in the 12-20 second band — the phase after the opening exchange, where tactical trading begins.

The second tier is conversion efficiency. Badminton has no xG. There is no probability model for scoring from an individual stroke, because points come from opponent errors more often than from finishing shots. So I built a substitute: the defensive conversion index. The method is simple — count how many times a player is forced into a defensive position (shuttle lifted high, pushed back to the rear court, balance lost), then count how often, across those occasions, they win the point in that same rally or the next one.

This figure says more than any smash table. A player smashing at 380 km/h but converting only 18% of forced-defence situations is a player who looks good. A player smashing at 330 km/h but converting 31% is a player who is unpleasant to face. xG does not sign contracts, but it tells me where I am putting my signature. The defensive conversion index does exactly that for badminton: it separates the beautiful from the effective.

Based on my own experience watching matches at Super 100 and Vietnam International Challenge level, the gap between these two player types is routinely missed. Spectators remember the smash. They do not remember that after that smash, the opponent got up and reclaimed the rally over the next four exchanges.

The third tier is physical cost, and I learned it after a mistake. In 2026, when football paused for COVID-19, I built a Bayesian model on ten seasons of Bundesliga data and concluded RB Leipzig would win the title with 54% probability. Bayern Munich won eight straight. Leipzig took four points from their last five. The cause was not technical. My model had no variable for "stadium without fans". I later rewatched forty matches and found that Leipzig's young squad lost roughly 27% of its pressing intensity without home support. I had to publish a correction.

The season on paper only looks good while the model has not met reality. Since then, every model of mine reserves a column for variables that have no column: injuries, fixture congestion, intercontinental travel, ranking pressure. Match-fixing, injury, red cards – variables with no column. In badminton there are more of them: court surface, shuttle type, drift inside the arena, and above all the travel schedule. A player contesting three events in three countries across twenty days will show an entirely different strength curve than the same player at a single tournament.

On process, I work to fixed hours. Data collection at 9 a.m., drafting at 11, number-checking at 2 p.m., publishing at 5. Full source audit on Saturdays. I once missed a deadline by two hours after spotting a discrepancy of 0.02 in a statistics table. I trust data, but I trust process more.

For a new match I code three passes. The first pass is live, recording raw sequence. The second is a video review at 0.5 speed to count rallies and classify strokes. The third cross-checks against the organiser's official score sheet. If the first and second passes diverge by more than 5% in any column, I discard the entire dataset rather than average the values. The rule sounds extreme, but it follows directly from the Russia lesson: losing a row is better than keeping a wrong one.

There is a temptation every data person meets: filling the gaps. Those thirty-eight empty rows could be filled by rewatching video and clicking by hand. But doing that across thirty-eight players would give me thirty-eight datasets measured by eye, across thirty-eight matches, under thirty-eight lighting conditions, with no second coder. What results is a feeling typed into a spreadsheet cell and dressed in the costume of a statistical table.

More dangerous still is correlation read as causation. I have seen an argument circulate online: players who serve high win more often, therefore the high serve is a weapon. But the causal direction may run the other way — strong players serve high because they are confident enough to contest the next rally, not because the serve itself produces points. The Russia World Cup was not an anomaly; it was a reminder about small samples. One tournament, one match, one three-win streak — all far too small to conclude anything about a player.

At the current stage of the transfer market, the pressure is heavier. Training centres, national teams and sponsors all want a single metric on which to base a decision. Rumours about a player moving to another setup, about a wildcard, about a new sponsorship — all arrive as unsourced information. My handling is simple: I rank rumours by evidence level. An official federation document is level one. Confirmation from a coaching staff or an agent is level two. A single social media account is level three. Level three never enters my spreadsheet.

And there is one thing I remind myself of every week. I have issued public corrections before, and I will have to again. A model defended at any cost quickly becomes a belief, and beliefs do not come with confidence intervals.

The signal I am tracking for the next cycle is not smash speed. It is the strength curve by rally length among Vietnamese players at Super 100 and International Challenge events through the 2026 season. If a player is winning mainly in short rallies while opponents stretch the rallies, the curve will change shape — and that arrives earlier than any ranking table.

As for those thirty-eight empty rows, I am leaving them as they are. They remind me that the most correct answer is sometimes a question: am I short of data, or short of people to measure it?

Good analysis is asking the right question, not having a beautiful answer.