A Blank Cell in Table Tennis Data Is More Dangerous Than a Wrong Number

A Blank Cell in Table Tennis Data Is More Dangerous Than a Wrong Number

The spreadsheet sat there, three empty columns, and nothing raised an alarm.

I was reopening a batch of rally-by-rally data from a WTT event, the part I was coding for the weekend analysis. The three most familiar fields had vanished: serve landing point, rally length, third-ball outcome. The dashboard above stayed green. No red flag, no warning line, no exclamation mark. That green is what kept me at the desk until almost dawn.

Forty-four years of watching table tennis, with my early years spent as a fact-checker at a sports magazine, taught me something cheap that few people accept: a wrong number denounces itself, while an empty cell stays perfectly silent. A wrong number looks odd. People frown, make phone calls, cross-check. An empty cell looks tidy. It sits quietly in the table, waiting for some line of code to read it as zero.

I call it the green error — the error that does not make a sound.

Most table tennis data infrastructure is built to handle abundance, not absence. That sentence sounds dry. It is the centre of everything below.

Table tennis entered the digital era nearly a decade later than football and basketball. The reason is scale. The table is 2.74 metres long, the ball is 40 millimetres across, and a world-class loop travels above 100 kilometres per hour. No camera system reads that trajectory without multiple synchronised angles and a good interpolation model.

Three milestones mark that race. In October 2026, the ITTF raised the ball diameter from 38 to 40 millimetres; spin dropped, speed dropped, rallies grew longer. In July 2026, the plastic ball replaced celluloid, and bounce and spin shifted once more. In 2026, WTT was created and turned rally data into part of the broadcast product itself.

Together those three milestones produce a consequence few people state: the more data is generated, the more places data can disappear without anyone noticing.

At Paris 2026, Fan Zhendong took men's singles gold, Ma Long closed his Olympic career with a sixth gold medal, and Chen Meng defended the women's singles title against Sun Yingsha. Every match like that generates thousands of data points: loop speed, revolutions of spin, landing points, footwork rhythm. Television viewers see graphics glide across the screen in a fraction of a second. Nobody sees how many empty cells behind those graphics were filled by interpolation.

One data pipeline serves four different customers, and they want four different things. Broadcasters want clean graphics, top speed, rally counts. Federations want the integrity of the ranking. Coaches want micro-patterns inside each service sequence. Betting markets want live data refreshed every second.

The ranking is the clearest example of dependence on completeness. The ITTF ranking is calculated from a player's best eight results in a rolling twelve-month window. Old points expire, new points replace them. This mechanism does not require numbers to be correct — it requires numbers to be present. A result missing from the record does not produce a wrong ranking. It produces a ranking that looks normal, computed from seven results instead of eight. Nobody detects it. There is nothing to detect.

That is the trap, and it is not exclusive to table tennis.

In Vietnam the gap is more visible. National teams still regularly reach SEA Games medal matches, but the number of matches captured by a proper system can be counted on one hand. A domestic tournament can produce hundreds of fine rallies, and most of them are not recorded well enough to answer the simplest question: where did the player lose the point, before the ball left the racket.

The transfer cycle now running pushes the problem further. Transfer rumour is a blank cell dressed in data clothing: someone heard someone say something, and that claim enters another person's spreadsheet as a confirmed event. Noise drowns signal, and the reader has no way to separate a verified entry from an open one.

To make this easier to picture, I split the empty cell into three layers, from the bottom up: mechanism, match, market.

Layer one: the empty cell read as zero

A rally lasts 1.8 seconds. The tracking system loses the ball for roughly 0.4 seconds while it passes behind the player's back. The software records the landing point as unknown. At the aggregation step, the averaging command reads the empty field as zero — not because anyone wrote it wrong, but because most computation libraries treat an empty cell as a skippable null, and skipping in a sum is identical to adding zero.

The result: that player's average rally length falls from 5.3 to 4.6. The dashboard prints an entirely reasonable line: a trend toward finishing points earlier. The coach reads it, believes it, and adjusts training around it.

I tell this example not to frighten anyone. I tell it because I made exactly this mistake, back when I first wrote code to count how often a player stepped into the middle of the table. Afterwards I set a hard rule in all my scripts: every empty cell must carry a reason code — tracking lost, no rally, sensor fault, undetermined. If an empty cell has no reason code, the program halts and produces no output.

Shenzhen taught me that haste in reform only produces a well-watered graveyard.

at full speed.

In sports data, that graveyard is a set of handsome metrics built on empty cells nobody bothered to label. A thirty-column report in which eleven columns have unclear provenance looks far more professional than a six-column report with complete notes. That fake professionalism is the hardest thing to remove, because nobody wants to be the person proposing that columns be cut.

A language problem sits right beneath this layer. An empty cell can carry four completely different meanings: no rally occurred, a rally occurred but the system lost tracking, data existed but the operator forgot to enter it, or the rally was never inside the measurement scope at all. Four situations, four different treatments. If all of them get the same label, the information was destroyed at the moment of entry, and no algorithm can bring it back.

This is why I talk about labelling before I talk about models. Models are the glamorous part. Labelling is the deciding part.

Layer two: what the empty cell hides

Table tennis is a sport of compressed geometry. Each side is only 1.525 metres wide, and elite players spend most of the match inside a twenty-centimetre band around the elbow — the junction between the forehand and backhand zones, where every decision slows by a hundredth of a second.

The most valuable data in table tennis is also the least recorded: how far the ball landed from the edge, and where the player's feet were about 0.3 seconds before contact.

Take a familiar attack pattern. A right-hander serves short to the opponent's backhand. The opponent pushes the ball to the middle. The server loops the third ball into the left corner. Three steps, one causal chain.

If I record only the final outcome — point won or lost — I have a correct but useless column. If I add the serve landing point, the pattern begins to appear. If I add the landing point of the return push, I can see who controls the chain. If I add foot position, I can see who is running and who is waiting.

Every layer left empty is a layer of explanation deleted. When the explanation disappears, people default to the most visible cause: form, mentality, age. Those three words are the shelter of every empty cell.

In doubles the problem multiplies. Doubles is a rotation system, and what decides the point is usually the space one player leaves for the other. If the data records only whoever struck the final ball, the entire coordination in front of it becomes a black box. The stat sheet will name one scorer, when in truth another player created that point with a movement nobody measured.

I once spent seventy-two hours rewatching a classic match in Madrid and counted 38 occasions when a midfielder moved into the central corridor. That number existed in no statistical table at the time. It existed only because someone sat down and counted. The same principle holds in table tennis as in football: what decides a match is usually not what gets recorded, but what sits between two recordings.

An assistant coach once asked me why I refuse to reach conclusions faster. I said that every fast conclusion is an unlabelled assumption, and unlabelled assumptions are cheaper than real data, but far more expensive when the bill arrives at the table. He did not like the answer. Two months later his team lost a group-stage match because a tactical adjustment was built on a metric computed wrongly from incomplete data.

Layer three: the empty cell enters the odds board

This is the part that bothers me most.

Live data supplied to betting companies is the darkest side effect of the digitisation of sport. I say this as a technical observation, not a moral lecture. A pricing model does not stop when data is missing. It redistributes weights. The empty cell is absorbed into another coefficient, and the output stays as smooth as ever.

The asymmetry lies here. For a coach, an empty cell costs knowledge. For an odds board, an empty cell costs trust. Both are hidden by the same feature: the empty cell does not appear in the final product. It appears only in the process that produced it, and nobody is allowed to watch that process.

Every tactical diagram is an organised lie in the face of the chaos of a match. I keep that sentence and extend it to data: every statistical table is an organised lie in the face of the chaos of measurement. The diagram is not wrong. It simplifies just enough to look beautiful, so people use it.

What I am about to say will irritate some colleagues: demand for more data in table tennis has nearly saturated, and the marginal benefit of the thousandth sensor is close to zero. What is missing lies at the perimeter of absence, not in the volume.

The analytics industry sells its clients a promise: more numbers mean less uncertainty. That promise fails because uncertainty does not vanish when it is measured more. It relocates, from not knowing to knowing without realising you do not know.

There is a phrase I learned from table tennis people in Southeast Asia, where budgets are smaller and the data far thinner than in the big centres: comfortable control. It means the ability to say I do not know yet without fearing a loss of prestige. In places where performance pressure is heaviest, that weapon is the first to be abandoned, because it produces no chart to present in a meeting.

When people replace the grass, they forget to replace what feeds the roots. In table tennis, the grass is sensors, high-speed cameras, real-time dashboards. The roots are the labelling process, the person accountable for explaining each empty cell, the culture that accepts a report with gaps. Replacing grass is fast. Growing roots is slow, and nobody takes photographs.

Another paradox: a model that cannot say I do not know is not a complete model, it is a confident one. Between the two, I always choose the first, even when it returns a less glamorous answer.

So looking ahead, what am I waiting for?

I am waiting for a new generation of dashboards where every metric must declare its coverage — the percentage of rallies that genuinely sit behind the number. For Vietnamese table tennis, where detailed match records remain scarce, this is a far cheaper opportunity than buying more cameras: record less, but label fully. A domestic tournament that captures four real fields, with a reason code for every empty cell, is worth more than a tournament capturing twenty fields of which seven are blank for reasons nobody understands.

For players such as Nguyen Anh Tu, Dinh Quang Linh or Mai Hoang My Trang, what they need from data is not a thicker report, but a report that answers exactly one question: where does the opponent score, and where do I lose points before the ball leaves my racket.

For me, next season will be the season I count empty cells instead of populated ones. If a system still cannot state where it does not know after twelve months, I will stop reading it, even if it is the only system I have.

Will a system that can say I do not know be discarded before it saves a single decision? I have no answer yet. But one thing I know for certain: the empty cell will still be there, silent, waiting for someone to read it as zero.

Data limitations: I have no access to the internal data-pipeline architecture of WTT or the ITTF, so the description of how an empty cell is read as zero is an inference model drawn from my own experience running statistical software, not confirmed technical documentation. The equipment milestones — the 40-millimetre ball from October 2026, the plastic ball from July 2026 — and the best-eight-results-in-a-rolling-twelve-month ranking mechanism are public information. The rally-length figures used as illustration are hypothetical examples to clarify the argument, not measurements from a specific match. Every conclusion here should be treated as a hypothesis requiring verification with fully labelled data.