When Table Tennis Data Returns Zero: The Analyst Who Must Refuse to Conclude

Late at night in Guangzhou, I reopened a table tennis analysis file after long hours of work. The score column was blank. The list of core facts was empty. No player name. No tournament name. No timestamp. No source. Only one label survived: a single line reading "table tennis." Everything else had evaporated from the system.

In my trade, that is the most dangerous moment. Missing data is something I am used to. The danger lies in the fact that the report structure is still intact: a nine-layer analytical framework, nine empty boxes, and an invisible pressure urging someone to fill them. I know that feeling far too well. With one stroke of the pen I could write a sentence that sounds entirely reasonable: "The world number one has just endured a suffocating points-defence stretch." It sounds smooth. It sounds credible. But it has no root.

I have long reminded myself of one thing: numbers do not lie, but the people who read numbers do. That night, it stopped being a professional motto. It became a warning.

Table tennis has become a sport of numbers

It is no accident that table tennis became a domain of statistics. A match lasts less than an hour but contains hundreds of rallies, each only a few seconds long. The density of data points here is among the highest in all adversarial sports. The International Table Tennis Federation and the professional tour series have standardised ranking points, entry allocation and the cyclical points-defence system. A player is not only competing against the opponent across the table; the player is competing against a clock of ranking points running backwards.

I began tracking table tennis data in the early 2000s, when statistics were still printed on paper. Later, as sports data platforms exploded, I learned one thing: table tennis does not lack numbers. Table tennis lacks people who read numbers correctly. Names that Vietnamese fans know well — Ma Long, Fan Zhendong, Wang Chuqin or Sun Yingsha — shaped the world rankings for years, and each of those names is a chain of data points rather than a simple legend.

Modern metrics allow a match to be broken into layers: efficiency in the first three shots, meaning the serve, the receive and the third-ball attack; the rate of points won on own serve; receive quality against different spin types; average rally length; win rate at deciding points. It all sounds scientific. But each of those metrics only means something when attached to a specific player, a specific tournament, a specific opponent.

I have seen the opposite happen. In 2026, I used an expected-goals model to argue that defending World Cup champions Germany risked elimination in the group stage, after calculations showed their expected goals conceded far exceeded the expected goals they created. The article was mocked. When Germany lost to South Korea, I received thousands of apologies. But the biggest lesson was not that I was right. The lesson was that if the input data had been empty, I would have had nothing to be right about.

Nine empty boxes and one minimum condition

A serious table tennis analysis cannot begin with inspiration. It needs a chain of evidence. A deep analytical framework has nine layers, and every layer has a minimum activation condition.

The first layer is technique, tactics and equipment. To discuss a playing style, I need to know whether the player is a defensive counter-attacker or an imposing attacker, whether they use pimpled rubber or smooth sponge, and whether they have changed equipment recently. Without a player name, this layer cannot be reached.

The second layer is player data and head-to-head records. World ranking, age, form cycle, points-defence pressure, head-to-head results over the past two years, win rate against foreign players, ability to handle deciding points. Every figure needs a name to attach to.

The third layer is the event system and points rules. Which tier a tournament belongs to, what role it plays in the Olympic cycle, how it affects the selection picture. I have seen draw branches read completely wrongly simply because the analyst did not understand same-association separation rules.

The fourth layer is the competitive landscape, particularly the balance between Chinese table tennis and the rest of the world. Seats in the world top ten, titles at major events, the depth of the under-21 pipeline. Without a named association, this layer is an empty frame.

The fifth layer is rules and governance: competition-rule reform, selection regulations, disciplinary rulings. Who benefits, who loses, what the historical precedent is. This is the most sensitive layer, because it touches power.

The sixth layer is coaching staff and the youth development system: the head coach's capability, the stability of the team, the age structure of the main squad, the conversion efficiency from junior ranks to the senior team.

The seventh layer is the risk surface: injuries, fixture congestion, technical changes still inside their adaptation period, the risk of being decoded by opponents.

The eighth layer is the public narrative and market expectations. How long a media label such as "the race for the majors" or "the emerging golden generation" can survive, and how wide the gap is between expectation and underlying reality.

The ninth layer is industry transmission: equipment, coaching, events, commerce, policy.

Those nine layers exist to force the analyst to state the origin of every claim, not to lengthen the report. When the list of core facts is empty, all nine layers collapse in silence. The only thing still standing is a domain label — and a domain label is not yet analysis.

The paradox: the biggest risk is fabricated data, not missing data

Outsiders often assume the risk in an analysis lies in a shortage of numbers. I think differently. The biggest risk lies in there being so many numbers that nobody can check them all, and in the fact that the report structure creates pressure to fill the gaps with something that sounds plausible.

I have said that a league table is only a summary, while raw data is the testimony. But even testimony can be distorted if the person taking it is not honest. With an analysis whose input is empty, the danger is not a wrong conclusion; it is a conclusion that sounds right. A fictional player, a fictional tournament, a fictional timestamp — all of them can be written smoothly, and all of them are fabrication.

That is why I set a hard rule: when information is insufficient, state clearly that information is insufficient. Do not speculate to fill a box. Every inference must carry a confidence label, separating what has been cross-checked from what is only a guess.

Correlation is not causation — a line every data analyst knows by heart, yet one that is easily forgotten when a beautiful conclusion is needed. A small sample is not enough to draw conclusions about a model. A domain label is not enough to build a story. And a nine-layer report framework, however neat, cannot replace having at least one real data point to start from.

I have gone against the crowd many times. In 2026, I analysed data from 240 matches in a Chinese second-tier league and argued that a team with no stars but the league's best expected-attack metric would win promotion with a very high probability, while the editorial board called it reckless. At the end of the season, that team won the title by five points. I retell this story not to boast that I am good at going against the crowd. I retell it to stress that going against the crowd only has value when a chain of data points stands behind it. Going against the crowd without evidence is merely attention-seeking.

What I have drawn from many data cycles is this: short-term form is often inflated after a few rounds and a few handsome wins. People call it an early-season explosion. But without a sufficiently long model, and without context on crowds, schedules and fixture density, any judgement about form is just a feeling packaged in terminology.

The limits of data are something I must always state

I write less than before. Slower than before. But every article has a section on the limits of data. That is not formal humility. It is part of the method.

There are things data cannot see: the psychology of a player before a home final, the pressure of an Olympic spot slowly closing, the unfamiliar feel after changing rubber. Those factors appear in no statistic, and an honest analyst must acknowledge that they exist.

When the stands are empty, I once said, I see the truest team, because the numbers are no longer distorted by the crowd. In table tennis the same holds. An internal match with no spectators, a training session with no cameras, sometimes reveals a player's true nature more clearly than a packed final. But to read those moments, we still need at least one anchor: who it is, where they are playing, and when.

What must happen next

The empty-data incident that night was not a failure of the model. It was a failure of a process — a data-extraction step had broken somewhere between the original article and the final analytical table. The domain label was still recorded correctly, while every content field was empty. That detail says a great deal: the fault lies in content extraction, not in classification.

The next steps are clear. Halt any conclusion flowing downstream from an empty analysis. Re-run the extraction process on the original text. Check whether the article body actually reached the processing stage. Record the publisher, the author, the publication date, and distinguish reporting from opinion from aggregation. Above all, make refusing to conclude a default rule rather than an exception.

I still believe raw data is the testimony of a match. But an empty testimony accuses no one; it simply waits for someone to pin a charge on it. What I leave for myself, and for anyone holding a spreadsheet: when the analytical frame is already built and the data has not arrived, do you choose to wait in silence for the data, or do you choose to write that frame a story that sounds plausible?