When the Coding Sheet Comes Back Empty: Where Elite Table Tennis Analysis Breaks

On a Friday night I reopened the coding sheet from a WTT Champions quarterfinal. Eleven columns: server, spin type, placement, tempo, steps taken, recovery time, receiver, stroke choice, rally outcome, direction changes, notes. Four thousand rows. Not a single cell filled. My colleague had attached one line: "I didn't get to the coding, watch the video first."

I sat looking at that sheet for fifteen minutes. My head already held a complete story: the winner controlled the serve, the loser died on the over-the-table backhand, game four was the turning point. Smooth. Plausible. And entirely unsupported.

Thirty years in this trade, and I still tell students never to do exactly what I was about to do: reconstruct a table tennis match from memories of beautiful rallies. Every revolution begins with a forgotten number left on a desk. This time the forgotten number was zero.

Where the analysis chain breaks

At national-team level, a modern tactical decision is not made by one person. It travels a chain: camera footage, a coder labelling every rally, a database, a statistical model, a coaching meeting, an on-table adjustment. Every link has an input and an output. Every link can fail silently.

The failure everyone guards against is the loud one: a model predicts wrong, a coach misreads a trend, a player cannot execute. The more dangerous failure is the quiet one, where the input is empty and nobody marks it empty. A blank spreadsheet still looks like a spreadsheet. It has column headers, formatting, a properly named file. Only the content is missing.

Table tennis is unusually exposed to this error because its event density is among the highest of any racket sport. A player may play three or four matches in a week at a WTT event. Each match runs four to seven games. Each game averages thirty to forty rallies. Each rally lasts under ten seconds. Multiplied out, one tournament produces thousands of raw data points. Nobody codes all of it by hand. Nobody watches all of it by eye.

That is why the analysis chain must be audited at the first link, not the last. If you only audit conclusions, you will always find a conclusion that looks reasonable. And a reasonable-looking conclusion is the cheapest commodity in this business.

Looking at what the camera skips

What viewers see on screen is not the real match. The real match happens in the gaps the camera leaves out.

Table tennis broadcasting follows the ball. That is the right visual choice and the wrong data choice. The ball is only the final output of a chain of decisions already completed. Roughly two to three tenths of a second before the paddle contacts the ball, everything has been settled: the centre of gravity has dropped, the foot has rotated, the shoulder has opened.

The first three seconds of a table tennis rally do not lie. The rest is just how we fool ourselves. Those three seconds contain the serve, the receive, and the first tempo exchange. Code those three seconds for every rally in a match and you have a model good enough to read most of the contest. Skip them and you are coding the part that was already decided.

Based on my experience tracking matches at WTT events across Asia, the metric I trust most appears on no scoreboard. It is steps taken per rally. No commentator mentions it. No broadcast graphic displays it. It says more than serve-win percentage.

Since 2026, the 40+ plastic ball has replaced celluloid. Spin dropped. Trajectories flattened. Rallies lengthened. When rallies lengthen, footwork becomes the decisive skill. The third-ball attack model has steadily lost potency. Points have migrated to the fifth, seventh and ninth ball. A player with only a good serve and a strong forehand loop is no longer enough. He needs the ability to switch between attack and defence inside a single rally, three times in that rally.

This is where I have been wrong before. In 2026 I predicted a young European player would reach the top ten within eighteen months on the strength of his backhand. He did not. Not because the backhand was poor. Because the footwork could not keep up with the backhand. A strong technique built on weak movement is only a strong technique under ideal conditions, and ideal conditions do not exist at that level. I rewrote that prediction in a short piece, not to perform humility, but to remember that my model was missing a variable.

Equipment rules and the redistribution of advantage

Table tennis is a sport where the governing body can shift the balance of power with a single document about materials. The 40+ ball was one such shift. The speed-glue ban, tightened from 2026, was another. The rule capping total rubber thickness, sponge included, at four millimetres, and the rule requiring one red and one black rubber face so opponents can read which side is producing which spin, were further shifts.

Seen through an analytical lens, every equipment rule change is a change of variables in the model, applied system-wide, simultaneously, to every player. The problem is that not everyone re-reads their historical data after the variable changes. A great many table tennis prediction models run on data blended from before and after these rule thresholds, unannotated and uncorrected.

In that case the model is not wrong because the algorithm is poor. It is wrong because the input is not homogeneous. This is another variant of the same disease: a data chain broken in the middle with nobody planting a flag.

Player data: numbers do not lie, but they do not say everything either

Modern table tennis world rankings operate on a rolling twelve-month window, taking a player's best results from the ITTF and WTT event system. This has a property few outsiders notice: it turns the ranking from a measure of form into a measure of scheduling endurance.

For a top-ten player, points are not only won, they must be defended. Last year's points expire on a fixed date. Miss the right event in the right week and those points evaporate. The pressure of defending points is often heavier than the pressure of earning them, because it carries none of the feeling of victory.

At the micro level, I split a player's profile into four layers. Layer one is win rate against opponents from other associations; for Chinese players this is the survival metric. Layer two is consistency at major events: a player who can win a WTT Contender but cannot reach a Grand Smash semifinal is two different people in terms of national-team value. Layer three is performance in deciding games, especially from eight points onward. Layer four is head-to-head record, split into three windows: all-time, last two years, and majors only.

Layer four is the most misread. A player may lead 6-2 all-time while losing 0-3 over the last two years. Media still call him his opponent's bogeyman, on the basis of a dead number.

Among the men's leading group, the structure has shifted markedly in two years. Fan Zhendong was the benchmark of stability for a decade. Wang Chuqin is a high-amplitude player, very high peaks and a fairly deep floor. Lin Shidong belongs to the generation born in 2026 and first reached world number one in early 2026 after winning Singapore Smash. The way he earns points differs sharply from his two predecessors: less reliance on raw power, more on tempo and early redirection.

Outside China, the counter-generation has taken shape. Sweden's Truls Moregard is the case my coding sheet annotates most, not because he is the strongest but because he is the least regular. His grip, his stance, his short-game handling all deviate from the standard template. Against players like that, head-to-head data has low predictive value, because playing them is itself an uncoded variable.

France's Felix Lebrun is the opposite case and, to my mind, tactically more interesting. Penhold. Elite modern table tennis is dominated almost entirely by the shakehand grip, so the existence of a penholder at the top is a technical paradox: it trades backhand defensive capability for speed close to the table. Lebrun turns that trade into an advantage by playing fast enough that opponents never get the match into the zone where the trade matters.

Brazil's Hugo Calderano is the third case, and the one showing that national-level data is no longer sufficient. In April 2026, at the World Cup in Macao, he won the title and became the first player from outside Asia and Europe to do so. A player from a country with no elite table tennis development system can still win a world title. Every model built on the "national depth" variable had to be rewritten after that result.

Event system and rules: the ranking as a governance instrument

The WTT event pyramid now has several tiers: Grand Smash at the top, then Champions, Star Contender, Contender and Feeder, plus the WTT Finals for the year's highest points group. This structure solved an old problem: the table tennis calendar used to be fragmented and players had to choose between competing systems. But it created a new one: the number of mandatory playing weeks rose, and so did the opportunity cost of rest.

In December 2026, world table tennis went through an event that I consider more important than any final that year. A group of Chinese Olympic and world champions, including Fan Zhendong, Chen Meng and later Ma Long, announced their withdrawal from the world rankings. The stated reason: rules obliging high-ranked players to compete in a set number of WTT events, with penalties attached to non-appearance.

The affair has two layers.

The public layer is the debate: who is right, whether stars have a right to rest after an Olympic cycle, whether WTT is commercialising the sport too fast.

The second layer is structural and far less discussed: when a ranking system can be used to compel a player to show up, it stops being purely a sporting measure and becomes a contract clause wearing a scoreboard's clothes. That is not wrong as event governance. But it changes the nature of the number every analyst uses as an input.

In other words, a variable in my model just changed meaning. The March 2026 ranking and the March 2026 ranking do not measure the same thing. Nobody notifies the analyst when that happens. This is precisely the silent failure I described at the start, but at system scale.

In early 2026, the governing body announced a review of that rule set. Whatever the outcome, one thing is clear: every data window crossing December 2026 carries systematic error. Anyone comparing long-term form without noting this is comparing two different things.

The development pipeline: depth versus individualisation

China's table tennis strength over three decades has never rested on one individual. It rests on a pipeline: the provincial system, national youth teams, internal selection trials, and an internal level of competition so dense that a player selected for the national team has usually faced harder matches than most international opponents face in an entire career.

That pipeline still runs. But the age structure of the leading group is shifting. Ma Long was born in 2026. Fan Zhendong in 2026. Wang Chuqin in 2026. Lin Shidong in 2026. The gap between the first cohort and the fourth is seventeen years. A healthy pipeline is not just one that produces many good players. It is one that produces good players at the right moment.

On the other side, rivals no longer try to imitate that pipeline. They take the opposite route: extreme individualisation. Moregard with an irregular game. Felix Lebrun with fast penhold. Calderano with a transnational training model, based in Europe and competing in Asia. These models are far cheaper and require no population depth.

The interesting part is that both models rest on the same thing. The Chinese pipeline uses data to find the best player inside an enormous pool. European teams use data to create a player who resembles nobody. Same tool, opposite purposes. Same trap: both can fail silently if the input is never audited.

The contrarian angle

There is an assumption almost the entire table tennis analysis industry runs on: that data is harder to obtain than conclusions. We invest in cameras, sensors, coding software, models. We treat data collection as the technical step and the conclusion as the intellectual one.

I think that ratio is inverting.

Data does not replace an old coach's intuition. It only hands him a more accurate map. But the map is only worth something if the blank areas on it are drawn correctly. The industry's problem today is not a shortage of maps. The problem is that blank areas get filled with guesswork, and nobody marks them as blank.

The biggest blind spot in elite sports analysis is not inside the model. It is that the model never says it has no data. A model always returns a number. A blank spreadsheet always returns a silence, and the human eye fills that silence with memory.

Memory of sport is a badly built instrument. We remember four beautiful rallies and forget forty ordinary ones, when those forty ordinary ones are the match. In table tennis the rally density is many times higher than in other racket sports. A five-game match can exceed one hundred and fifty rallies. Nobody remembers one hundred and fifty rallies. So the brain selects, and it selects by a spectator's criteria, not an analyst's.

There is a professional consequence I have to state plainly: most of the widely shared table tennis analysis is not built from data. It is built from a good memory and a good sentence. A skilled writer creates a stronger impression of precision than a data-rich writer who writes poorly. That is a failure of the information market, not of any individual.

Closing

The blank sheet from that Friday night eventually got filled. Two hours, rally by rally. And I found something the original story in my head did not contain: the winner did not control the serve. The winner won on the fifth ball, by redirecting roughly two tenths of a second earlier than his opponent.

Had I written that night, I would have written a wrong piece. And that wrong piece would have looked very reasonable, because it was written in the voice of a man who has watched table tennis for thirty years.

In elite table tennis, the question is no longer who has more data. The question is who dares to say that their data sheet is empty. Whoever does that may not be the best analyst. But they will be the only one not fooling themselves.