Inside the Taolu Judging Room: 5.00 – 3.00 – 2.00 and the Data Stream Leaving the Floor

5:12 a.m., a training hall in Longgang, Shenzhen. The lights on the second floor came on before the corridor lights. A sixteen-year-old girl stands at the corner of the mat, her left heel half a foot off the line. She jumps, rotates in the air, lands. The sound of her foot on the mat is like a breath. The sixty-first time that morning.

Her coach, a forty-eight-year-old man who once competed at provincial level in bare-hand forms, is not writing down scores. He is writing down heel height on landing. He is writing down the angle of the shoulder. He is writing down how many half-seconds she holds her balance before lifting the other foot. That notebook lives in his coat pocket. It never enters the judging room, and it never appears in any published table.

Every season begins on a morning when no one is awake yet.

Three months later, on the electronic board of an Asian arena, two lines of scores sit 0.03 apart. One reads 9.696. The other reads 9.726. The hall goes quiet. In the technical area, someone bends over a phone. Outside the arena, an app has already updated the scoreboard about two seconds before the referee announces the placings.

To understand why 0.03 matters that much, you have to understand how that scoreboard is built.

Competitive wushu divides into two branches. Sanda is full-contact: there is impact, there is counting, there is a result clear enough that arguing is pointless. Taolu is the other branch: performed forms. An athlete walks onto the mat alone, performs a bare-hand or weapon routine within a set time, and has no opponent to beat other than the scoreboard itself.

The familiar events include changquan, nanquan and taijiquan, along with short weapons such as straight sword and broadsword, and long weapons such as spear and staff. Major championships also include all-around events, where the scores of one bare-hand routine and one weapon routine are added together to find the best athlete.

The current taolu scoreboard has three groups, and those three groups are not equals.

Group A carries 5.00 points and is handled by five judges. This is the deduction group. Athletes start at the maximum and lose points against a printed list of faults: wrong posture, wrong direction, loss of balance, insufficient height in a required position, forgotten movement. The list is long, detailed, and leaves almost no room for interpretation.

Group B carries 3.00 points and is handled by three judges. This is structured impression scoring. There is no fault list. The criteria are power, rhythm, coordination of hands, eyes, body and stance, expression, spirit, and the overall structure of the routine.

Group C carries 2.00 points and is handled by three judges. This is the difficulty group. Each difficult element is assigned a code and a fixed value, plus a connection value when two difficult elements are linked directly. In the grouping commonly used in coaching circles, 3xx codes cover airborne jumps and rotations, 4xx covers balances, and 5xx covers falls and throws.

Group C appeared after 2026. Before that, nearly all the separation between athletes sat in Group B — that is, in the most subjective part of the scoreboard. Turning difficulty into its own group, with codes and fixed values, was a reasonable reform: it made part of the performance verifiable, countable, and explainable to an audience.

Some institutional context. The International Wushu Federation was founded in 2026 and is now headquartered in Lausanne, with more than 150 member federations. The first World Wushu Championships were held in Beijing in 2026 and the event has been staged every two years since. Wushu has featured at the Asian Games since Beijing 2026. At the Beijing 2026 Olympics, wushu appeared as a side tournament that did not count toward the official medal table. For the Dakar 2026 Youth Olympic Games programme, wushu was added to the competition schedule.

Each of those steps carried a requirement: the sport had to become easier for outsiders to read. Group C is the most readable tool taolu has.

Across six years of watching taolu at continental and world level, I keep a notebook of my own, exactly the kind I used to keep when I followed football teams. I record each group score for each athlete in a final, then calculate the spread of each group within the same final. The result repeats itself to the point of being awkward.

The group that decides a taolu final is not the group with the most judges, and not the group advertised as the most objective — it is Group B, the least transparent of the three.

In the finals I have logged, Group A typically spreads only about 0.10 to 0.15 points between first and sixth place. Group C spreads even less, mostly within 0.05 points, because difficulty is binary: land the code and you take its full value, miss it and you lose it, and at world final level almost everyone lands it. Group B spreads 0.25 to 0.35 points.

In other words: the section worth 20 percent of the total, advertised as modern and objective, barely separates anyone. The section worth 30 percent, scored by eye and experience, is where the medal is decided.

I still keep a habit from my football years: possession percentage is the most deceptive statistic in any table. A team holding 62 percent of the ball through sideways passes in its own half is not controlling the match, it is only keeping the ball. Group C works the same way. A maximum difficulty score says the athlete completed the required set of codes. It does not say the routine was sharper, tighter, or heavier.

This leads to a coaching consequence that has become standard in many places: young athletes are taught to optimise Group C first and Group B second. A 720-degree airborne rotation has a fixed value and can be measured on video. A breath in the right place cannot. When resources are limited, people invest in what can be measured.

And this is where the story leaves the mat.

Taolu scores today are entered on judges' tablets, transmitted to the competition's central server, displayed on the arena screen, and simultaneously pushed out to live data feeds. The interval from a judge's finger touching the screen to the number appearing on a data feed outside the arena is measured in milliseconds.

Those feeds are sold to sports data aggregators, and those aggregators sell on to betting companies. For taolu, betting markets are concentrated mostly in Asia, and the common markets are medal handicaps and head-to-head matchups between two athletes in the same final group.

Live data supplied to betting companies is the darkest side effect of the digitisation of sport, and taolu is the clearest example I have watched. A sixteen-year-old performs her routine. She does not bet. She does not receive a single cent from the data stream her body produces. But that data stream left the arena ahead of her.

In Group C, leaked information is nearly harmless, because codes and values are public. In Group B, leaked information has value. A whisper that tonight's second judge rewards spirit and dislikes disjointed rhythms is enough to move a head-to-head line. The biggest risk in taolu is not a fixed match. It is a predictable scoring tendency.

There is a parallel dimension here, and I watch it with two heartbeats.

In Vietnam, competitive wushu entered the sporting scene in the 1990s. The country's most prominent taolu athlete is Duong Thuy Vi, a name tied to several Asian medals in short-weapon events. If you read those generations through my notebook, a pattern shows up clearly: strong Group B, weak Group C.

The reason lies in coaching resources. Difficult technique requires a teacher who can operate airborne rotations at international level, a training floor with padding thick enough to allow thousands of repetitions without injury, and a continuous youth competition system so athletes get used to the pressure of being scored on codes. The beauty of power and rhythm, meanwhile, can be built in an ordinary gym under the eye of a good teacher.

From around 2026, Vietnamese national teams began inviting difficulty choreographers from China on short cycles. Those visits did not change the foundation, but they narrowed the Group C gap from heavy defeat to moderate defeat. The 0.03-point margin in the opening of this piece belongs to that generation.

The same routine, read two ways on two sets of stands. Spectators in Shenzhen watch Group C first, because that is what they were taught to recognise. Spectators in Hanoi, through a phone screen and a Vietnamese-language commentator, watch rhythm first, because that is what their eye catches naturally.

People remember medals. I remember the hands that wrap tape.

In the corridor of the technical area, there is a person whose job is to wrap athletes' wrists before they step onto the mat. There is a person who checks the shoelaces three times even though they are already right. There is a coach who, the night before competition, sits down and rewrites the whole difficulty design by hand on paper, even though the file is already on his laptop.

The dressing room does not know how to lie. It only knows how to whisper.

What I hear in those corridors is not a story about a stolen medal. It is a story about a sport that built a layer of transparency on top and left the deciding layer underneath, in the dark.

The first thought many people blurt out when they hear about the scoreboard is: we need more cameras, we need each judge's score published, we need more transparency. I think that argument is pointed the wrong way, and Group C is the proof.

Group C has been the most transparent part of taolu for more than twenty years. It has codes, fixed values, video confirmation, and it can be recounted by eye. Precisely because it is transparent, it has almost no capacity left to separate athletes at final level. Adding cameras to Group C only makes it more transparent, not more decisive.

Transparency does not touch the layer that decides. It only makes that layer quieter.

The second thought concerns home-nation bias. In taolu, home bias exists but it is not the largest force in the scoreboard. The larger force is familiarity bias: judges reward the movement vocabulary they were trained in. In changquan, the movement system is highly standardised, so familiarity bias has little room. In nanquan and taijiquan, where stylistic authenticity is judged more strictly, familiarity bias surfaces — and it surfaces under the label of style, not the label of nationality. That is the blind spot in most online arguments about taolu judging.

The third thought concerns the Olympic path. Many argue taolu needs to become more like artistic gymnastics to be welcomed. Wushu's inclusion at Dakar 2026 shows the Olympic movement can read the sport, and it reads it precisely through the difficulty code system. But that same system produces a layer of fourteen- and fifteen-year-old athletes selected for rotation ability rather than expressive ability. The difficulty arms race makes taolu more legible to outsiders and more hollow to insiders.

Three things are worth watching next quarter.

The first is the age of athletes posting the highest Group C scores at national youth events. If the average age keeps falling, this sport is changing profession.

The second is the Group B spread in major finals. If the spread narrows, judges are being squeezed, and the soul of the sport is being standardised in a different way.

The third is scoring data disclosure policy. So far, most events publish only the group totals. On the day individual judge scores are published in real time, both the betting market and the federations will have to rewrite how they work.

The sixteen-year-old in Longgang still has three months before competition. She has never heard the term data feed, has never seen a betting line on her event, and will most likely never receive a cent from the data stream her body produces.

The only thing she controls is the next jump, the sixty-second, and the height of her heel on landing.

Which of us is really keeping time for this sport — the person writing numbers into a notebook at 5 a.m., or the system reading them back in milliseconds?