Expected Points in Badminton: Rereading Lee Zii Jia's Paris 2026 Bronze

The third game of the Paris 2026 Olympic men's singles bronze-medal match ended 21-11. Lee Zii Jia beat Lakshya Sen, having lost the opening game 13-21. Stop there, and the only image anyone keeps is a Malaysian player exploding after falling behind. What I wrote into my notebook was not the scoreline. What I wrote was four columns of numbers, and the fourth column forced me to rewatch the footage a third time.

For more than twenty years, sitting in Penang and covering badminton for the Malaysian market, I have grown used to separating a match from the story the media tells about it. A scoreline lies in a very polite way: it is right about the result and wrong about the cause. A player who wins three straight points with three different winners gets called inspired. The same three points, if they come from three unforced errors by the opponent, get called a collapse. The scoreboard does not distinguish between the two.

Context: from football xG to badminton expected points

I came to badminton from the betting desk, and I came to the betting desk from data. In 2026, at 47, I built a simple xG model from StatsBomb data for the Malaysia Super League. Faisal Halim was then a winger at Pahang with an xG/90 of 0.41, well above the league average, while bookmakers still priced him at 11.0 to score. I put 500 ringgit on him to score against Selangor and he scored twice. I won 2,200 ringgit. But what I kept was not the money; it was a line I wrote in my notebook: I do not trust the story, I trust the number that tells the story.

A year later, at the 2026 World Cup, I calculated PPDA across 120 European and Asian qualifying matches. Russia posted a PPDA of 8.1, meaning they allowed opponents more passes in their own half than any other side in the top twenty. The media called them lucky hosts. A PPDA of 8.1 is a confession by an entire tactical system: they willingly conceded territory, packed bodies in front of the box and waited. Russia beat Saudi Arabia 5-0 in the opener and advanced with six points. Since then I have written numbers first, conclusions second, and I never settle a judgment on a single metric.

Badminton has no xG. Nobody publishes shuttle-contact data by court zone at a resolution fine enough to run a model. So I built my own measures, three of them.

The first is expected points (xP): for each rally situation I estimate the probability of winning that rally based on where both players stand, shuttle speed and who holds the initiative. A smash winner in a two-on-one position carries high xP but says little. A point won after seventeen shots from a defensive position is worth recording.

The second is a pressure index, borrowed from how PPDA is calculated: the number of touches the opponent is allowed before I am forced to intervene, divided by successful interventions. The lower the number, the earlier a player breaks the opponent's rhythm.

The third, the column I added last, is landing depth: the average distance from where the shuttle lands to the opponent's back boundary line. This column measures something television cameras almost never show.

Rereading the bronze-medal match

In game one, Lee Zii Jia lost 13-21. Watching the footage back, he did not smash worse. He smashed less. Lakshya Sen pushed the shuttle deep into both back corners, forcing Lee to retreat, then dropped it to the net on the next beat. Average rally length in game one sat around twelve shots. Not long enough for Lee to exploit his physical edge, not short enough for him to finish early. A dead zone.

In games two and three, the numbers moved in exactly one column. Lee Zii Jia's landing depth increased, meaning his pushes travelled closer to the back boundary. Lakshya Sen lost his position at the net, and once he lost his position at the net he had to lift the shuttle high. Only then did Lee have something to smash. Lee's win rate in rallies over fifteen shots rose markedly across the last two games, not because he was fitter, but because the number of long rallies went up, and he was already the man who wins in that zone.

Based on my experience watching matches, this is the hardest kind of shift to spot on television. It produces no highlight. It only produces rallies in which the opponent has to run half a step more.

The contrarian angle: do not buy the character story

Most of the Malaysian post-match writing that day pinned the bronze on one word: character. Lee Zii Jia left the national association, managed his own career, was once banned from competition, then came back and won a medal. The story is too neat to ignore. My problem with it lies elsewhere: it cannot be verified, and therefore cannot be used to forecast.

At Euro 2026, my model predicted Germany would win and Italy took the title. I had ignored the psychological factor in high-pressure knockout matches. After the tournament I encoded 120 knockout matches from 2026 to 2026 and added a new variable: the average distance between the three lines when a team trails. That variable measures behaviour, not composure. The lesson: something unmeasurable can still be true, but it is not allowed inside a forecasting model.

With Lee Zii Jia in Paris, the shift in landing depth and rally management is behaviour. It is measurable, repeatable and verifiable through footage and federation statistics. Character is the conclusion drawn from that behaviour, not its cause. Blending the two is the fastest way to devalue a model.

In Vietnam, where Nguyen Thuy Linh and Le Duc Phat are the leading names in women's and men's singles, the problem is harder still. Their international match count each year is small, the sample is thin, and any conclusion drawn from it must carry an error warning. A small sample still beats no sample, provided the writer is willing to say plainly what is being measured.

Signals for the next cycle

During the current transfer window, the market is pricing free-agent players on reputation and on recent win counts. Both measures are noisy, because they lump together a player who wins through structure and a player who wins because opponents keep missing.

In 2026, a Thai broker asked me to value a young midfielder playing in J-League 2. I used xG, PPDA and distance covered to recommend a fee of 80 million baht, 30 percent below the selling club's original asking price. The deal closed exactly on that number. The principle is simple: when valuing a player, look at the win rate in rallies over fifteen shots and the average landing depth across the final two games of each match. Those two numbers tell you whether he wins through structure or through luck.

The smash winner lies, but expected points never do.