Badminton Still Has No xG Equivalent: The Data Gap After Paris 2026 and What It Costs

The arena fell silent in a way an eight-thousand-seat hall should never fall silent. In early August 2026, Carolina Marin was leading her Olympic semifinal in Paris when her right knee buckled during a movement she had made thousands of times. The umpire stopped the match. Medical staff walked on court. The electronic scoreboard stayed lit, still showing numbers, and nobody looked at it anymore.

I was sitting two metres from the screen with a pen in one hand and the other resting on a keyboard, and I realised I had nothing to write down. The match statistics were complete to an uncomfortable degree: smashes, net shots, unforced errors, outright winners, longest rally. Every box had been filled. The one box that actually decided a bronze medal for an entire badminton nation — the state of one person's ligaments at minute 34 — did not exist in any table.

Three days later, Gregoria Mariska Tunjung received an Olympic bronze she did not win with a single shot in a bronze-medal match. Marin withdrew. The result was established by a medical report, not by a stroke. Across the whole data ecosystem of this sport, no metric was ever designed to record that kind of outcome.

I tell this story to point at something structural, not at one athlete's misfortune. Badminton has a great many numbers and almost no model. Football crossed that bridge long ago: a shot from the edge of the box and a tap-in from five metres are converted into the same unit, expected goals. That shared language lets people compare a striker in the Eredivisie with one in the Premier League, argue about transfer values, and detect that a team on a winning run is doing it on luck rather than structure.

Badminton has no such unit. We count smashes. A smash at 11-4 in game one and a smash at 18-20 in the deciding game are counted identically. We count unforced errors. A shuttle into the net after a twenty-shot rally and one into the net on the second shot are counted identically. We count net shots without any way of knowing how much win probability that net shot created.

The 2026 move to rally-point scoring to 21 made this urgent. Under the old side-out system, a patient player could survive by lifting shuttle after shuttle. The new format shortened rallies, increased the number of decisive exchanges per match, and turned every rally into a directly scored unit. Once every rally is worth a point, the quality of each rally becomes the most important variable in the sport — and that is precisely the variable the current statistics do not measure.

The Instant Review System BWF introduced from the 2026 season technically owns more data than anyone needs: three-dimensional shuttle trajectory, landing timestamps, estimated error margins. All of it is used to answer one question — in or out — and then sealed. There is no open shot-level dataset an outside analyst can query. Football's private data vendors, by contrast, sell event-level packages to anyone who pays.

The first consequence sits in the most celebrated number in the sport: smash speed.

Every major tournament, broadcast graphics flash a three-digit figure in kilometres per hour. Viewers assume it describes the speed of the shuttle travelling toward the opponent. It does not. It is measured as the shuttle leaves the strings, within milliseconds. After that, a shuttlecock — roughly five grams, sixteen feathers arranged in a cone — decelerates faster than any other object in competitive sport.

The physics is simple: the feather skirt generates enormous drag, and drag rises with the square of velocity. At high speed the shuttle behaves like a tiny parachute. By the time it reaches the opposing racket, the remaining velocity is often a third, sometimes a quarter, of the broadcast figure. A smash recorded at 400 km/h may arrive at a speed a professional has ample time to read.

Smash speed therefore measures the racket, not the difficulty. It tells you a player has power; it does not tell you the shot had value. When I cross-referenced the fastest recorded smashes in my own logs against the rallies those smashes actually won, the correlation was weak — and for some players inverted. Their fastest smashes were blocked or counter-attacked, because a maximal swing leaves them planted and out of recovery position.

The second consequence is the black box called the unforced error.

Try to define it. A shuttle driven long because the player aimed for a tight angle — unforced error, or a reasonable tactical choice? A shuttle into the net while scrambling to return a steep attack — whose error? A shuttle landing in open court because the player was half a step late on shot twenty — technical fault or physical fatigue?

Official statistics classify by observable outcome, not by cause. Because each data provider applies its own threshold, two stat sheets for the same match can differ by fifteen percent on unforced errors. When I re-coded rallies manually using a different criterion — not whether the shuttle landed in, but whether the player was in position to play the shot — I had to revise several of my own earlier conclusions. Most "unforced" errors in long rallies were positional errors: the player arrived half a step late, had to hit from a poor posture, and the outcome was inevitable.

This changes how we judge players. A player branded error-prone may simply be operating at a higher movement intensity, attempting harder positions, and therefore contacting the shuttle in disadvantage more often. His error count is higher because he is trying more, not because he is less accurate. The stat sheet cannot distinguish the two, and neither can the reader of the stat sheet.

The third and most neglected consequence is drift.

Every professional badminton hall is its own aerodynamic system. Air conditioning, airflow direction, altitude, temperature, humidity — all bend shuttle trajectory in ways no metric records. Istora Senayan in Jakarta is famous for draft that turns the push to the back boundary into a private art form. Kuala Lumpur behaves differently. European halls in winter behave differently again, to the point that organisers must select a different shuttle speed grade from BWF's scale.

I once spent an entire tournament estimating drift by court zone from the trajectory of high lifts, then comparing it against out-of-bounds error rates. The gap between the two ends of the same court was large enough to explain a meaningful share of boundary errors. Some shuttles recorded as technical mistakes were caused by a venue's air conditioning.

This is also the underrated explanation for home advantage. We credit the crowd and psychology. The crowd matters, but its effect is uneven and unmeasurable. What is measurable, though nobody has measured it systematically, is familiarity with a specific hall's drift. A player who has trained at Istora for a decade knows exactly how hard to push the shuttle from each corner. A visitor from Europe needs two games to learn it, and in a 21-point format two games is too late.

And in the middle of all those numbers is the pause.

Every game has a sixty-second interval at eleven points. Between games there are one hundred and twenty seconds. No shots, no points, no metrics are recorded. That is exactly when matches are usually decided.

When play stops, I hear my own pulse. I learned this from broadcast work: during intervals, people do not talk about tactics, they talk about breathing. A player who enters the interval unable to stabilise his breathing in the first forty seconds will walk out in the same tired body. Another, even while trailing, can reset and change the next game entirely. No column exists for this.

An Se-young's Paris 2026 is the clearest case of on-court data failing to explain a career. She beat He Bingjiao in the final by a clear margin, and the stat sheet will show dominance across most categories. The real story of her year sat off court: disputes over injury management, over national-team training structures, over an elite athlete having to fund her own support staff. No metric reflects a player fighting her own system before walking into a final.

And here is the side of digitalisation nobody puts in the press release.

The same courtside dataset, captured at the touchline with second-level frequency, has two customers. The first is the viewer, who receives a prettier stat sheet after the match. The second is the betting market, where those numbers become prices before anyone opens the stat sheet on their phone. I have watched courtside data travel along channels audiences never see: data providers in front-row seats entering each point by hand, transmitting within seconds, and bookmakers repricing between rallies. That is the darkest byproduct of sports digitalisation.

Meanwhile, the review system everyone assumed would reduce controversy has relocated it rather than removed it. Once, arguments happened on court between umpire and player and ended in seconds. Now they happen in a review booth, between experts and frame rates, and never end. A shuttle on a line with a millimetre-scale error margin will always produce two groups of people who see two different results in the same image, because images do not solve the problem of trust. Players also use challenges tactically: a failed challenge still buys thirty seconds, breaks the opponent's service rhythm, cools a scoring run. Perfectly legal, entirely invisible in the statistics.

There is one more trap I keep falling into and keep reminding myself about: over-reading a single match. Badminton produces far fewer rallies per match than football produces possessions, and each rally is shaped by drift, shuttle batch and accumulated fatigue. At that sample size, a beautiful shot proves nothing. I once built an entire argument about a player's rise on three matches and had to publish a correction six months later. The confidence interval on three badminton matches is close to useless.

For the same reason, I distrust conclusions of the form "team X's system works" after one team event. A Thomas Cup is four ties of at most five matches. A title is a tiny sample, and in tiny samples luck plays a bigger role than we like to admit. That does not diminish the medal. It only means we are not allowed to derive systemic lessons from it.

So what is worth tracking over the next twelve months?

First, shot-value models. If a data vendor begins publishing win probability by stroke type, contact position and score state, this sport will change how it evaluates players within two seasons. Second, a drift index per venue — buildable by anyone with a fixed camera and patience, before the big organisations get there. Third, interval physiology: breathing rate, heart rate, recovery time, if it is ever published at any level.

I walk into the data cathedral not to pray but to listen to the noise of truth. In badminton that noise still drowns the signal. The more precise the number, the wider the distance between the human being and the match — and in a sport where a ligament at minute 34 can decide a medal, that distance is what we must learn to measure before we learn to trust it.

The question I leave for next season is not who wins. It is: when a player walks into the interval at eleven points and no metric is being recorded, how much longer will we pretend we already know what happens next?