From Wu Lei's xG to Player Valuation: How Data Is Rewriting the Rules of Sport

In 2026, at a stadium in Shanghai, Wu Lei took seven shots with a combined xG of just 1.2. In other words, with chances of that quality, an average striker would score about one goal. But what caught the attention of Japanese analysts was not the goal. It was the number 8.4 km — the distance Wu Lei covered off the ball in that match, double the AFC Champions League average. Three months later, Urawa Red Diamonds were eliminated in the quarter-finals on penalties. The PPDA figure of 8.7 that I published, a measure of pressing intensity, began to appear in the tactical meetings of Japanese coaches.

That was the moment I understood something: the scoreboard is for the audience, but data is for the people who make decisions. Yet it would take several more years, and a few mistakes, before I understood that faith in data must itself be tested like data.

Context: why results are not enough to tell a story

Table tennis and football, however different in pace, space, and number of participants, share one foundational problem: the final result does not fully reflect the process that produced it. A player who wins 4-0 may have won because the opponent made unforced errors at decisive moments, not because of a superior tactical system. A team that wins 3-0 may have won through two corners and one mistake by the opposing goalkeeper.

For years, Asian sports media built its stories on results. Win, and you are praised; lose, and you are criticized. That approach is easy to grasp, but it ignores one of the most important variables: sample size. A single match is a very small sample. A single tournament is also a small sample. To assess an athlete's true ability, one needs hundreds, even thousands of recorded and classified rallies.

That is why I abandoned emotional writing. Every judgment now had to be tied to at least three quantitative indicators. The writing became longer and slower, but the professionals began to trust it. In 2026, ahead of the World Cup in Russia, I published a series predicting that France would win. The basis was not "mental strength" or tradition, but a model combining xG, a high defensive line indicator of PPDA 9.2, and the average distance covered by midfielders. I was mocked for ignoring Brazil and Germany. When France beat Croatia 4-2 in the final, the mockery vanished. A sporting director at Beijing Guoan called to invite me as a data consultant for the winter transfer window, starting with tracking overlooked young French players.

Core: how data prices value

Intuition is a lazy variable; data is a judge that never sleeps.

In the transfer window, noise always exceeds signal. Dozens of rumors appear every day, and most of them have no basis. The analyst's role is not to spread rumors, but to rank them by the quality of their evidence.

I divide sources into three tiers. Tier one is official information: signed contracts, release clauses, club announcements. Tier two is semi-official information: statements by agents, wage-bill movements, medical-test schedules. Tier three is unsourced rumor, usually born on social media and amplified by the sharing loop. Only tiers one and two belong in a valuation model.

Ranking rumors by tier does more than filter noise; it measures how the market shifts. When a tier-three rumor is confirmed by a tier-two source, the expected value of the deal changes instantly. In the transfer window, the timing of information is sometimes more important than the information itself.

In table tennis, the data system is less developed than in football, but the principle is identical. Instead of xG, one uses the point-win rate by serve type. Instead of PPDA, one uses the win rate in short rallies (under three exchanges) and in long rallies (over five exchanges). The gap between these two indicators reveals playing style: a player strong in short rallies is a proactive attacker; a player strong in long rallies is a counter-attacker who lives on patience and reading spin.

In a match I watched live, a young player reached a 68% win rate in short rallies but only 39% in long rallies. The scoreboard showed he won 4-2. But the model showed that if the opponent could drag the match into long rallies, his win probability fell below 50%. That is the kind of information a scout needs, and the kind the scoreboard never provides.

In the transfer market, the principle matters even more. A player's value lies not in the number of goals, but in repeatability. A striker who scores 15 goals from a total xG of 8 is a lucky striker, not a good one. A striker who scores 12 goals from a total xG of 14 is an unlucky striker who possesses strong positioning skill. In the transfer window, smart clubs buy the second kind, not the first, because the market always overprices the lucky one.

The same holds for table tennis: a player who wins many titles thanks to a favorable draw is worth less than a player who loses in the semi-finals but faces stronger opponents. The Elo rating, adjusted for opponent quality, reflects this better than a plain ranking. And in a system with small samples, the margin of error is always larger than people assume.

This leads to a rarely discussed issue: career length. In esports, a player's career is far shorter than a footballer's, yet the youth-development and post-retirement support systems are almost nonexistent. Data can measure reaction speed and decision time, but it cannot measure what happens to a twenty-year-old when a career ends at twenty-five. A valuation model that counts only present value is a model that ignores long-term cost.

Contrarian angle: correlation is not causation

But this is where I must argue against myself.

In 2026, I pointed out that Urawa lost because of a high PPDA. But if Urawa had won, would I have concluded the opposite? That is the trap of retrospective analysis: we choose the indicator after we already know the result. An indicator that can explain every result is a useless indicator.

Data does not speak for itself. People choose which indicator to plot, which time frame to slice, which source to trust. Each choice is an editorial decision, and each editorial decision can be abused. I trust data, but I do not trust the honesty of anyone standing between data and the reader — including myself.

In table tennis, the problem is larger. The data system is poorly standardized, samples are small, and many tournaments do not publish detailed figures. A model built on thirty matches has a far wider margin of error than one built on three hundred. Presenting a number without presenting its margin of error is an act of intellectual dishonesty, even if the number is correct.

And here is the crux: correlation is not causation. A team that wins after changing tactics does not mean the tactics produced the win. They may have won because the opponent was tired, because of a referee's decision, or simply because of luck. By the same logic, a player who wins after switching racket rubber does not mean the new rubber is the cause. It may be pure coincidence within a short sequence of rallies.

Takeaway: a signal for the next cycle

So, instead of asking "who is better," ask "what is the win probability in each scenario." Instead of asking "how much is this player worth," ask "what is this valuation model's margin of error, and how many matches is it built on."

In the coming transfer window, the signal worth tracking is not the loudest deals, but the deals with the smartest contract structures. Genuinely valuable contracts usually sit with small clubs, where there is no money to buy attention, but enough patience to buy data.

A final question for the reader: if you could see only one indicator to judge an athlete, which would you choose — and are you sure you did not choose it merely because it confirms what you already believed?