One Name, Seven Frames of Reference: The Data Gap in Modern Martial Arts

On March 8, 2026, in an indoor arena in Osaka, I sat in the fourth row of the men's taolu final. The gold medallist scored 9.71; the runner-up scored 9.68. Those three hundredths lived inside Group C — difficulty — where two judges differed by half a point and were dropped from the average before publication. No sensor touched the floor that afternoon. No high-speed camera measured the height of a twisting jump. The only scoreboard in the hall was a hand-printed sheet of paper, passed through three pairs of hands before it reached the secretary's desk.

That same evening I opened my laptop to rewatch a kickboxing card streamed from Tokyo. There every kick existed in three data layers: output, accuracy, and recovery time between rounds. The winner landed fourteen fewer strikes than his opponent and won on the judges' decision.

Two events, seven hours and four hundred kilometres apart. One word covers both: martial arts.

In early January 2026, a sports data company in Tokyo sent me a file for an injury-risk assessment. The label at the top read "martial_arts". No sub-label. The file held 598 athlete records, each with weekly training sessions, weigh-ins, and a column left blank called "bouts".

I returned the file unanalysed. The reason sat elsewhere: the data was complete, the label was wrong.

Under the single name "martial arts", at least seven frames of reference run in parallel, and they cannot be averaged together. Karate kumite under the World Karate Federation is decided by an eight-point margin, with senshu awarded on a tie and hansoku penalties. World Taekwondo uses electronic body protectors, three two-minute rounds, and sensor-registered points confirmed by referees. Judo under the International Judo Federation revolves around ippon, waza-ari and shido, with golden score running without limit. Professional boxing uses the ten-point must system across up to twelve rounds. Muay Thai scores power and accuracy across every exchange. MMA applies the unified rules. And taolu, poomsae and kata — the performance group — are marked on a ten-point scale split into three parts: A worth five points for technical quality, B worth three for power and rhythm, C worth two for difficulty.

In karate kumite, the common injuries sit in the hand, wrist and head. In judo, the shoulder, neck and ribs. In taolu, the patellar tendon, ankle and lower back. Three risk profiles, three rehabilitation protocols, three different rest cycles — sharing one label.

The more striking question is how that label came to exist. Sports data platforms expand their catalogues by sweeping every combat discipline containing the word "martial" into a single folder, then pushing the classification work down to the user. Users are mostly journalists, team managers, or insurers — people with no time to open each rulebook. The result is a habit: any injury rate calculated from that folder gets diluted by disciplines that have nothing to do with each other.

Governance widens the gap further. The International Boxing Association was suspended by the International Olympic Committee in 2026, and a new body called World Boxing was created to carry the technical side. Karate kata appeared at Tokyo 2026 and was excluded from Paris 2026, meaning a complete injury data stream appeared and vanished inside a single cycle. Wushu was added to the Dakar 2026 Youth Olympic Games programme. Each time a discipline enters or leaves a programme, its recording system travels with it, and the athletes stay behind with a fractured file.

From June 2026 to December 2026, I maintained the Martial Arts Injury Dataset: records for 412 combat athletes across seven systems — UFC, ONE Championship, Rizin, World Karate Federation, World Taekwondo, International Judo Federation, K-1 — plus 186 performance athletes in taolu, poomsae and kata. The method was deliberately manual. Every injury required three independent sources: the organiser's official notice, a released medical record under review, and a source from the athlete's home market. Where the three disagreed, I recorded the disagreement rather than picking whichever looked most plausible.

Four results are worth pausing on.

Combat athletes competing more than six times a year showed a 31 percent higher rate of anterior cruciate ligament rupture than those competing fewer than four times, and a 24 percent higher rate of hand fractures. The gap held when I separated the disciplines. It did not depend on the rulebook. It depended on the calendar.

Performance athletes showed a 41 percent lower rate of acute injury, but a 2.3 times higher rate of knee joint degeneration by age 30. They are not struck in the head. They are struck against themselves, every day, on the same joint, for fifteen years.

Average retirement age in the sample: MMA 33.4; boxing 34.1; kickboxing 31.8; taolu 24.7; poomsae 26.2.

And the final index — a line of contact with an organisation after retirement. Among combat athletes, 78 percent still had at least one point of contact within two years of leaving the ring. Among performance athletes, that figure was 19 percent.

Two different data systems share one name, and that ambiguity means the athletes with the least voice pay first.

That 19 percent figure appears in no industry report. I had to build it by calling provincial federations in four countries and asking a single question: after your athletes retire, who do they call? Three of the four said they keep no such list. I entered that answer as data.

Every number tells the truth, but a fight never tells all of it. In the January file, 63 taolu athletes had a completely blank "bouts" column. My first instinct was to drop them from the sample. Instead I kept them and renamed the column "times counted". Their profession was never designed to be counted, so nobody counted. The silence of the data carries information here, and the writer's job is to name that silence rather than fill it with a tidier sentence.

One further detail. In Japan, where I live and work, professional combat sports operate under a mandatory medical licensing system, with regular eye and brain screening before every ring appearance. Performance competitions have nothing comparable. Same athlete, same body, two separate files. If she moves from taolu into combat competition at 22, her medical record begins at zero — nobody knows how many landings that knee absorbed over the previous decade.

Injury does not detonate in one match; it accrues quietly across seasons. The problem in martial arts is that two ledgers are being kept in two banks that never speak.

My first reaction to the dataset was to propose a single unified metric across the entire martial arts label. After testing it against specific cases, I withdrew part of that proposal.

More data does not automatically produce more safety. The World Taekwondo electronic protector is the clearest case I have tracked. When scoring depends on a force sensor, tactics shift: athletes learn to generate just enough force, at exactly the sensor's location, instead of optimising for the true power of a kick. The measurement system becomes part of the rules, and the rules get rewritten by the measurement system. The same mechanism appears in boxing with CompuBox: those figures describe a fight very well and hold no authority to decide anything. A fighter can win every statistical column and lose on three scorecards, and both results are correct inside their own frames of reference.

Forcing a single metric onto taolu would destroy Group B — power and rhythm — faster than anything else. Group B carries three of ten points, and it exists precisely because it cannot be measured algorithmically. When a routine is scored by three Japanese judges, two Malaysian judges and one Iranian judge, what is being measured is a cultural agreement about beauty. Algorithmising it means replacing that agreement with the standards of whoever writes the algorithm.

The counter-argument carries weight too. If every discipline keeps a fully separate label, the data loses comparability, and smaller federations will never gather enough sample to detect an injury pattern. Performance communities in many countries have fewer than two hundred national-level athletes. No single federation spots a knee degeneration trend on its own, because a handful of cases a year is not a pattern.

The balance sits in a two-tier label structure. The upper tier keeps "martial arts" intact to serve cross-discipline comparison. The lower tier requires a declared frame of reference: kumite, poomsae, taolu, judo, boxing, Muay Thai, MMA. The lower tier is the condition that makes any upper-tier rate meaningful.

A map is not the territory; data is not the fight. A map that prints "Asia" across an entire continent will not help anyone find a road.

In my dataset, athletes moving from a scored amateur system into professional competition carried injury files three times as dense in their first two years. Hiroki Akimoto, a World Karate Federation kumite world champion before he competed in ONE Championship, belongs to that group. Same body, same discipline, different scoring system. Rodtang Jitmuangnon in ONE Championship and Tenshin Nasukawa, who moved from kickboxing into boxing, sit in the same transition zone, where the old data cannot follow them into the new system.

The necessary work is clear and cheap: every martial arts organisation, including performance circuits without commercial insurance, should be required to keep a minimum injury record per athlete, and that record must travel with the person when they change competition systems. No sensors needed. Just one data column with a human name in it.

Reality always has the right to contradict, and this time it contradicted my own habit. I spent six years hunting a predictive index for each fighter, when what was missing most was one correct label. If a single name covers seven frames of reference, then every conclusion drawn from that name is a guess presented as data. Fans of martial arts in Vietnam, in Japan, or anywhere else deserve a more honest ledger — and the first step is not adding more data, but naming correctly what we are counting.