An Empty Analysis File: The Discipline of a Badminton Data Writer

At 2 a.m. in Nagoya, I opened a file called Stage-2. It had nine sections, each with tables, bold headings, and a structure as polished as a report presented to a board. Every cell repeated nearly the same sentence: insufficient information. No tournament name, no athlete names, no scoreline, not a single line of data to hold on to. The striking part was elsewhere: the file's structural completeness was 100 percent, while its informational completeness was 0 percent. A complete and entirely hollow dossier. I sat looking at it for a while, because this is the hardest kind of data to handle in this trade: data about the absence of data.

In my daily work I run a two-stage pipeline. The first stage extracts the source: title, publishing body, date, named entities, and every verifiable information point. The second stage is the deep analysis: technique, form, tournament system, landscape, rules and governance, coaching staff, risk surface, media narrative, industry transmission. The second stage only carries value when the first stage has substance. When the first stage returns nothing, the second still runs, still produces all nine sections, still has empty space to fill, and the only thing it can do is repeat that it has nothing to say.

Badminton makes this problem more painful than football. With football I have xG, PPDA, pass maps, and per-possession data. With badminton, most events on the BWF World Tour do not publish fine-grained data: no xG equivalent, no automatic rally classification, no landing coordinates. If I want anything, I have to record it myself. Since 2026, after Japan lost to Belgium in the World Cup round of sixteen, I set one rule for myself: every statement must be tied to a number or a chart. When Japan push high, I do not see magic, I see the formula of collapse. That rule has saved me many times. Not this time, because the chart it should attach to is empty.

Before writing, I separate two things that are easily confused. Empty data is not a conclusion, it is a state of the pipeline. A match that took place is real even if nobody recorded it. An empty record only says that collection broke somewhere: the source was not specified, the date was skipped, entities were not extracted, or the original text never entered the machine correctly. In this trade that is an operational failure, not a failure of the sport.

The next step is to audit the pipeline before doubting reality. I walk upstream with four questions: which source published it, on what date, which names appear, and which information points can be independently verified. If all four questions go unanswered, the problem sits at the input, and every inference after that is decoration. This conclusion sounds simple, yet it saves me weeks of work because it ends an expensive loop: trying to analyse something that was never recorded.

The strongest temptation comes immediately after. When cells are empty, the writer's hand reaches automatically for words that cannot be verified: character, spirit, desire, fate. These words sound warm, read smoothly, and cannot be refuted by any data, because they are attached to no data. Every pressing sequence is a testimony, every number is a confession, but only when there is a sequence and a number. Otherwise the prose is just filling a gap with steam.

The rest of the work is rebuilding a minimum viable dataset. For a badminton match, five variables can always be taken from video even when the event publishes no statistics: the score of each game, the duration of each game, the distribution of rally lengths, the count of unforced errors, and the score at the two eleven-point intervals. People need faith to place a bet, I need data to be certain. These five variables do not give me the full picture, but they give me a picture that is not invented.

Rally length is the most underrated variable in that set. A player or a pair who sustains many long rallies is not necessarily the one controlling the match; often the one stretching the rally is the one being pulled around. That is why I never fully trust metrics such as distance covered or step count packaged as effort. Running a lot produces beautiful numbers, and beautiful numbers produce the wrong story. In badminton I would rather read the rally-length distribution of Japanese women's doubles pairs than read kilometres, because that distribution speaks about shuttle control, while distance only speaks about being dragged.

There is a second layer of data that is always verifiable, and I use it as an anchor. Akane Yamaguchi won the BWF World Championships in two consecutive years, 2026 and 2026 (source: BWF). Kento Momota claimed 11 titles in 2026 (source: BWF). Viktor Axelsen won the Tokyo 2026 Olympic men's singles (source: IOC). These three lines are correct, verifiable in thirty seconds, and explain nothing tactically. That is exactly my point: a true event is not automatically useful data.

At match level, what I watch most closely is the closing stretch. A three-game badminton match lasting 70 to 90 minutes is not far from the final twenty minutes of a football match with wide substitution rights: neither side is playing to open up the game any more, they are playing to wear down what is left of each other. When detailed fitness data is missing, I read indirect traces: rally length drops in the third game, unforced errors rise between points 14 and 18, and the quality of returns from the back court falls clearly. Those are things the eye can see if the viewer knows where to look.

In the risk table I build for every dossier, the scariest column is not the wrong one. A wrong conclusion can be fixed with new data. The scariest column is the fabricated one. A table of zeros produces two reactions: a disciplined writer leaves the zero, an undisciplined writer fills it with a story. Readers have no way to tell them apart, because the format of the fabricated and the format of the true look identical.

There is a near-default belief in the industry: more data means better writing. I think the real failure mode lies on the opposite side. The most powerful tool for making an analysis look serious is not statistics, it is format. Nine sections, tables, an index, jargon — all of it transmits a signal of rigour while the information gain can be zero. A nine-section report with every cell reading insufficient information is more honest than a nine-section report with cells filled by speculation, yet the two look so similar they are hard to separate.

Correlation is not causation is a line everyone knows by heart. The harder and less spoken line is: structure is not substance. A dossier with perfect structure and zero content is a warning about this trade itself, not about any tournament. On the market side, the same holds: in badminton, what gets hyped as a race — sponsorship deals, entry slots, media attention — usually belongs to the big brands, while real value sits in small federations, where a player with no analytics team still reaches the later rounds. Home court has never been an advantage, only noise encoded into points, and noise is also a form of data, except that it does not measure quality.

The signal I will track in the coming cycle is not the volume of data events publish, but its completeness rate: what share of matches carry a source record, a date, entity names, and verifiable information points. That is the least glamorous index and also the best predictor of analysis quality over the next six months. I do not remember the match, I remember the heat map of that match, and when the heat map is blank, the most honest thing an analyst can do is say it is blank. If a dossier with no data can still be written out in full, then what is being sold to readers, ultimately, information or merely the appearance of it?