The Empty Spreadsheet in Badminton: What Remains When Every Metric Disappears

2:14 AM, Surabaya

The ceiling fan was spinning at its third speed, the one I have learned is the noisiest but also the one that helps me focus best. My third monitor had just returned a blank sheet. Not blank because of a connection error, not blank because I mistyped a query. Blank because there was genuinely nothing inside it.

I have been in this trade long enough to know that an empty dataset is not a technological failure. It is a fact about the sport I cover every day. And that fact is far more uncomfortable than a defeat for the team I love.

That night I wrote nothing. I turned off three monitors, left one glowing, and reread an old note from June. In that note was a sentence I had written to myself: most of the badminton data the audience sees is recorded selectively, and that selection produces a systematically distorted picture, not a randomly distorted one. The sentence was true in June. It is still true now. I had simply never written it out in full.

When every tournament stops, that is when I hear my own pulse.

The breathing rhythm of a badminton season

Professional badminton does not run on a European football calendar, where you can plan a year in advance and know exactly which week holds which fixture. It runs on a chain. The Badminton World Federation (BWF) tiers its events: the World Tour Finals at the summit, then Super 1000, Super 750, Super 500, Super 300, and below that the Super 100 group alongside International Challenge and International Series events. Each tier invests differently in record-keeping.

What almost nobody tells the audience is this: the lower the tier, the less data exists. A Super 1000 final may generate dozens of automatically captured metrics, an instant replay system, and on-screen statistical panels after every game. A qualifying-round match at an International Challenge sometimes leaves behind exactly one thing: a final score, pasted onto a website with an interface from a previous decade.

Between those two worlds sits a gap I call the data trough.

That trough is not evenly distributed across time. It clusters in specific windows of the year, when the major Asian and European events pause and the calendar lands in the transition between Olympic qualification and continental championships. In those weeks the number of matches barely drops. The volume of data produced collapses to almost nothing.

That is when my work changes nature. From analysis into archaeology.

An ecosystem with four layers, three of which leak

Over the years I have mapped the badminton data supply chain into four layers. I use it as a chart to know where I am standing whenever I write.

Layer one is official data. This is what the BWF records inside a sanctioned event: game scores, match duration, occasionally the fastest smash speed and unforced error counts. This layer is trustworthy on facts but thin on tactics. It tells me who won, not why.

Layer two is commercial data. Sports data companies sell live information to bookmakers and media platforms. This layer is thicker and faster, but it is designed for a different purpose: pricing probabilities over short windows. It does not care about explaining a shot. It cares about updating a line.

Layer three is hand-notated data. This is where I live. I time rallies, count net approaches, log the distribution of points by rally length. It is slow, it is tiring, and it is the only thing I can actually verify.

Layer four is inferred data. I rewatch video and label what I could not count live: a player's court position on receive, directional tendencies, rhythm changes after taking a lead.

The problem is that the first three layers leak into one another, and layer two is routinely relabelled as layer one. A reader sees a number, believes it is official. Rarely is it.

The hole is not in the source code. It is in the eyes of the person reading the source code.

Three kinds of gaps, only one of them harmless

When I work with badminton data, I sort gaps into three categories. I borrowed the taxonomy from medical statistics, where researchers have wrestled with missing patient data far longer than I have wrestled with match data in Nanjing or Jakarta.

The first is structural missingness. An entire tournament has no recording system. This is the least harmful kind analytically, because I know exactly what I do not know. I cannot accidentally reason wrong from an empty set. Total emptiness is transparent.

The second is random missingness. A camera on court two fails mid-session, a notator is absent for personal reasons, a session is moved after a technical fault. This is also relatively harmless, because the absence does not correlate with the outcomes.

The third is selection-driven missingness. This is the dangerous kind, and in my experience covering live events it accounts for most of what I call badminton's data illusion. People notate carefully the matches featuring famous players, at televised tournaments, on evenings with crowds. Everything else vanishes.

The consequence is not the missing numbers. The consequence is that the remaining dataset does not represent the sport. It represents the sport as covered by media.

Suppose an unheralded player is competing with a rare style built on net control. He wins repeatedly at Super 100 events with that approach. But because none of those events are fully notated, the finding sits outside the field of view of anyone who only reads data. Three months later, when he breaks into the main draw of a major, people write that he came from nowhere. He did not. He was always there. The data arrived late.

Fourteen hours and a lesson in counting by hand

I started counting by hand at seventeen, in a room in Surabaya, during a World Cup in Russia. I sat for fourteen consecutive hours dissecting a single Portugal match, logging every touch. The result cost me sleep: the best player on the pitch had only eighteen touches but generated more expected goal value than the rest of a major national team combined.

I wrote three thousand words. An Indonesian forum reposted it. Twelve thousand reads overnight.

But the real lesson was not in the number. It was that I knew exactly where I had miscounted. I remember missing one touch in the seventy-third minute because I left my chair to make tea. I remember agonising for two minutes over a disputed moment where I was not sure who touched the ball last. That uncertainty, on that day, did not appear in the article.

That was the first time I understood something I still teach myself before every piece: a homemade dataset is only as honest as the footnotes attached to it. If I log three thousand touches without noting that at least seven of them were uncertain, I have sold the reader a precision I do not own.

I walk into the church of data not to pray, but to listen to the noise of the truth.

Since then, every analysis I write carries a small section at the end, in small type, stating the sample size, the estimated error, and where in the recording process I lacked confidence. Readers almost never read it. But I have to write it, because otherwise I am no longer a data notator. I am just a storyteller with charts.

When a net cord destroys an entire model

In badminton, the equivalent of a football hitting the post is a shuttle clipping the net tape and dropping on the opponent's side. It happens constantly. It decides points. And it is nearly invisible in every statistical model I have ever seen.

A net cord is not destiny — it is only an extremely small deviation between expectation and probability.

I once built a small dataset of just a few hundred such moments from matches I rewatched. The original purpose was to demonstrate that luck distributes evenly over time, that it cancels out in the long run. I was wrong in an interesting way.

What I found: net cords were not randomly distributed by scoreline. They clustered more heavily at important points. That sounds like a major finding, and I nearly wrote it up as a full article. Then I stopped.

There are at least three mechanisms that could produce the same pattern, and I could not separate them with my data.

Mechanism one: psychological pressure changes a player's contact point, sending the shuttle closer to the tape. Mechanism two: at important points, a notator like me pays more attention, so I notice and log more net cords — recording bias, not an on-court phenomenon. Mechanism three: at important points, both players play safer, hitting more toward the middle, which geometrically sends shuttle paths nearer the net.

Three mechanisms. One pattern. No way to separate them unless I log net cords at unimportant points with identical attention.

I killed the piece. It remains the best decision I have made as an analyst.

The blind spot nobody wants to name

Here I have to say something most sportswriters avoid, because it does not generate traffic.

Live data supplied to betting companies is the darkest side effect of sport's digitalisation. Not because betting itself is a sin, but because its incentive structure is entirely different from that of sports analysis.

An analyst wants to understand why a player wins. A bookmaker wants to know the probability a player wins over the next two hours, and needs to know it fast enough to move a line. Those two goals produce two different products from the same match. The second layer of data I described above — commercial data — is funded by the second goal. Yet it flows backwards into the first layer in public perception, and eventually into the articles readers believe are tactical analysis.

The result is an ecosystem where fast, second-accurate data is also the shortest-horizon and least explanatory data. And the slow, labour-intensive, highly explanatory data — like my hand notation — is produced less and less, because nobody pays for it.

The more precise the number, the wider the distance between the person and the match.

Instant review and the illusion of transparency

Badminton has an instant review system allowing players to challenge line calls. It sounds exactly like football's story about referee-assistance technology.

And it fails in exactly the same way.

Replay technology does not eliminate controversy. It moves controversy from the court into the review room and into the grey areas of the rulebook. Previously people argued whether the umpire saw it. Now they argue whether the camera angle is sufficient to conclude, whether a millimetre of feather sits inside or outside the line, whether the system is correctly calibrated for the arena's altitude.

Technically, the system is more accurate than the human eye. Socially, it creates a new kind of uncertainty — one that looks more scientific but is no easier to accept. Fans do not react to precision. They react to a sense of injustice. And a sense of injustice does not shrink when you hand someone a three-dimensional rendering.

Here I want to say something blunt about my own profession. When review systems give us a binary answer — in or out, right or wrong — we tend to believe every other question has a binary answer too. We begin to think about tactics in terms of right and wrong rather than probability. We begin hunting for the decisive shot, the decisive error, the decisive moment.

Badminton does not work that way. Badminton works by accumulating deviations so small that nobody records them.

The counterintuitive angle: correlation is not causation, and it is worse than that

Everyone knows this line. In badminton analysis, it is not strong enough.

The deeper problem is that causality can run against intuition, and our data usually cannot tell the difference.

Take a hypothetical example, constructed from logic rather than any specific match. Suppose I collect data from a hundred matches and find that winners have a higher average rally length than losers. The intuitive conclusion: better fitness produces victory, so long rallies signal a well-trained player.

But the mechanism may run the other way entirely. A player leading may deliberately extend rallies to hold a safe rhythm and force the opponent to take risks. A player trailing may end rallies earlier with high-risk shots. In that case, extending rallies is a consequence of leading, not a cause of it.

If I write the first version, I have sold bad coaching advice to everyone who reads it.

For years I have applied a personal rule: use causal verbs only when I have a mechanism. Without a mechanism, I write in the language of correlation and say so explicitly. That is why my pieces contain more sentences like 'the data suggests' than 'the data proves'. Not because I lack conviction. Because I once got it wrong, publicly, and had to rewatch seven matches over sixty hours to understand that I had asked the wrong question.

That happened at a European championship. I predicted a team would win because they had the tournament's highest total expected goals. They went out in the quarter-finals. The eventual champion was a side with a strange defensive structure that appeared in no attacking table. Later I discovered that their centre-back pairing had allowed opponents just twenty-three touches inside the penalty area across four hundred and fifty minutes. Twenty-three touches. I had not looked at that data because I did not think it was data.

Data does not lie. But I asked the wrong question.

Rereading the past with eyes that already know

There is a trap I fall into more often than any other, and it is dangerous precisely because it does not produce obvious errors. It produces an article that is factually accurate and inferentially false.

Once the result is known, every data point becomes obvious. Causal chains rearrange themselves in my head. The winner seems to have played with more confidence from the start. The loser seems to have shown fatigue from the second game. Those things may be true. But I cannot know they are true, because I only saw them after learning who won.

The only method I have found against this is writing down my pre-match expectation, in words, with a timestamp, and putting it away. Later, when I write, I compare. If my pre-match expectation matches the result, I am allowed to use data to explain mechanism. If it does not, I must write that I was wrong and explain why.

I still keep that habit. It makes me look less clever to some colleagues. I accept that.

Why I still notate by hand

People ask why I do not use automated data, why I still time rallies when everything is available. The short answer: automated data is designed by someone who wants to sell data, whereas the data I need is designed by someone who wants to understand a match.

The longer answer: when I count myself, I am forced to decide what is worth counting. Every such decision is a hypothesis. If I count net approaches, I am assuming the net matters. If I log point distribution by rally length, I am assuming tempo matters. Hand notation forces me to look directly at my assumptions. Automated data hides them.

It is also why I built a personal database during the period when global sport froze and there were no matches left to analyse. I rewatched hundreds of matches from prior years and created more than two thousand four hundred fixed situations. I found that short corners in one national league had risen by over two hundred percent compared with two seasons earlier, while the scoring efficiency from them had fallen by a third.

An interesting finding. And almost nobody cared.

I wrote five thousand words about it. Readership was low. But I learned something more important than the finding itself: before every piece, I must ask who will care about this. If the answer is 'only me', it is a note, not an article.

The data trough and the trap of filling silence

Back to that night in Surabaya. When the screen returned a blank sheet, my first reflex was to go find other data to fill it. Another source, another forum, another old video. I realised I had an unhealthy reflex: I was afraid of silence.

For a writer who works with data, silence is the most uncomfortable state. Nothing to count means nothing to tell. And my profession pays me to tell.

But there is a kind of honesty that exists only in silence. The gap between tournaments is when I can see things speed does not permit: signs of exhaustion in a player after a dense run of events, a change in how he walks onto court, the quietness of a coaching team recalculating everything.

No metric records the fact that a person walked into an arena half a second slower than three months ago. But it exists. And in the long run it predicts more than peak smash speed.

I am not saying intuition replaces data. I am saying there is a layer of truth my method cannot reach, and I must accept that rather than invent a metric to cover it.

What I do when there is nothing to analyse

I have a procedure, and it is not pretty.

Step one: I write down in words what I genuinely know about the upcoming match or tournament. No inference, only facts. This list is usually embarrassingly short.

Step two: I list what I believe but cannot verify. This is the long list, and it is the most important one.

Step three: I write out my expectation with a confidence rating from one to five. I put it away.

Step four: I find what I can measure myself in real time without anyone supplying it. In badminton that list always starts with rally length, point distribution, net approaches, and time between points.

Step five: I ask who will care. If nobody, I write for myself.

This procedure gives me something data never does: an explicit boundary. And when I know where my boundary is, I can write with real confidence instead of fake confidence.

The error lies with the reader

There is a way of seeing that I try to hold in every piece, and it explains why I rarely argue about specific numbers.

When a match is controversial, the majority reflex is to correct the number. People assume that if the number is right, the controversy disappears. In my experience the opposite holds. The same dataset, two readers, two opposing conclusions, and neither is lying. The error is not in the table. It is in how the reader assigns meaning to the table.

I once handed the same dataset to two coaches. One saw stability. The other saw stagnation. Both read the numbers correctly. Both drew different actions for the following week.

This changed how I write. I argue less about which number is right, and spend more time on the conditions under which a number was produced, by whom, and in whose service. That is the part that determines meaning.

The view against consensus

The analysis community holds a quiet belief: more data is better. I no longer believe that.

Over the past seven years, the number of metrics available for a badminton match has risen markedly. The quality of conclusions, in my assessment, has not risen correspondingly. The reason is simple: new metrics are added without mechanisms. People measure more. They understand no more.

The paradox is that readers feel safer seeing more numbers. Volume creates an illusion of depth. A piece with seven tables looks more credible than a piece with one table and three paragraphs explaining mechanism. But the second is usually more correct.

I used to think this was a writer's problem. Now I think it is a structural one. In an environment where readership is measured, the piece that looks complicated beats the piece that is simple and correct, at least in the short run.

I have no solution. I have one personal choice: write fewer metrics, explain more mechanisms, and accept being shared less.

Signals I will track in the next cycle

When the data trough closes and tournaments return at density, there are three signals I will track, and I am writing them here to bind myself.

First, I want to see whether lower-tier events remain data-empty. If they do, I will notate one tournament myself, just one, in enough detail to serve as a reference document. One fully recorded event is worth more than ten half-recorded ones.

Second, I want to track the movement of young players from qualifying into main draws, and measure how many of those jumps are actually recorded by anyone. My prediction: almost none. If that holds, it confirms my argument about data selectivity.

Third, I will check whether coverage of regional tournaments starts drawing on a single commercial source. If everything uses the same source, diversity of interpretation disappears without anyone noticing.

These are not predictions about results. They are predictions about the sport's information infrastructure. For someone in my line of work, that is the more interesting thing.

What a blank sheet taught me

That night, after switching off three monitors, I sat for another thirty minutes in the dark watching the fan turn.

I realised my job for years has not been to find the truth about matches. My job has been to build a system that tells me when I do not know. Those two sound similar but differ at the most important point: the first can lead to arrogance, the second cannot.

An empty dataset taught me more than a full one. A full one gives me answers. An empty one forces me back to the question.

In badminton, people say the best player is the one who makes the fewest errors. I no longer think so. The best player is the one who knows precisely which errors he will make, and does not pretend otherwise.

That is all I have after a night without data. Not a discovery. A limit.

And a limit, I think, is the only thing worth writing in a sports analysis.

If you read only one paragraph

Whenever a dataset looks perfect, ask who recorded it, who paid for it, and who disappeared from it. Those three questions explain most badminton controversies better than dissecting the numbers again.

For a writer like me, real discipline lies in daring to leave part of the sheet blank, instead of filling it with speculation dressed up as data.

When the crowd counts the points, I count the gaps between them.