The Classification Trap: When Combat-Sports Analytics Chooses Silence

Three in the afternoon in Bangkok. The Lumpinee stands sit hollow after the weigh-in. I stay behind alone, ears still ringing with fighters' steps on the wooden floor, nose still thick with the hot liniment rubbed into calves, and I open the laptop. The screen shows exactly one line: unclassified. No fighter's name, no weight class, no ruleset. An analytics system built with serious money just returned an empty result, after three weeks of running and a stack of electricity bills. What makes my skin crawl is not the failure. It is how it happened. The machine ran all the way to the domain-labelling step, printed two words — combat sports — and stopped. It did not know whether the article was about professional fighting, about sanda, or about performance forms. Not knowing, it refused to speak. An empty stadium is the largest mirror a team's identity will ever get — and this time it reflected a machine that would not pass judgement. For a decade now, the combat-sports world has lived inside a data fever. Every punch gets a symbol attached. Every round is diced into hundreds of data points: strikes landed per minute, defensive rate, control time on the mat. Big newsrooms pour money into analytics desks, hire data engineers, and build tables before the referee has even announced the result. I understand the pull. A fight night lasts hours, but a statistics table reads in three minutes. Newsrooms need speed, need headlines that travel, need something reusable between events. And so a habit formed: classify first, understand later. But that pipeline has just exposed a fatal flaw at its very first step. Before any metric can be computed, the system must answer a question that looks obvious: what kind of martial art is this? The answer is not obvious at all. Modern combat sports include boxing, MMA, kickboxing, Muay Thai and wrestling. Sanda is its own line, permitting throws. And performance forms are something else entirely: they are scored on movement difficulty and performance quality, not on wins and losses. Those three groups demand three different readings. Applying one group's formula to another is not a rounding error; it is a systemic fault. An analysis of a performance form scored by finish rate can only reach one conclusion: that every athlete has never won. Wrong to the point of meaningless. This is where I want to linger longer than most people do. Take a concrete case. At successive SEA Games, Vietnam has repeatedly entered both wushu disciplines: sanda and performance forms. In forms, an athlete such as Duong Thuy Vi spent years bringing home medals through precision and the difficulty of her routine. If someone fed her entire file into a machine that scores by wins and losses, the screen would display a long row of zeros. Not because she is weak. Because she has never stepped onto a ring to fight — she steps onto a floor to perform. When the measuring stick is wrong at its foundation, every correct conclusion becomes meaningless. In sanda, the trap is subtler. The rules permit throws, so the timing of an exchange, the moment a throw is launched, and the way referees score differ sharply from boxing. There is no ten-point must system as in boxing, no title-challenge concept in the Western sense. Yet most online metric tables still tag sanda with the exact criteria of kickboxing. The result is that skilled throwers get undervalued, while heavy punchers get glossed. Then comes the heaviest issue, the one I consider the most important in the entire industry: weight cutting. In combat sports, weight cutting is the variable with the highest predictive power for death, and also the most concealed. A fighter can shed five or seven kilos in days by dehydrating. The danger is not the reading on the scale; it is what happens afterwards: acute dehydration kidney injury, rhabdomyolysis, and collapses right at the weigh-in where a few minutes' delay in medical response is paid for with a life. To build an alert screen for this variable, a system needs three things: walk-around weight, weigh-in weight, and a history of previous missed weights. No fighter name, no weight class, and all three vanish. A system that cannot screen weight-cut risk has not earned the right to call itself a safe analytics system. It is only a good-looking ranking machine. Money sits on the same fault line. In boxing, four major sanctioning bodies split power, so a single division can hold four champions at once. In MMA, the fighter revenue share has been measured in the high teens to about twenty percent, while top boxers can take more than half. An entry-level fighter in a lower tier may earn a few hundred dollars a bout, while monthly training costs swallow nearly all of it. Those ratios only mean something once we know which sport, which organisation, which weight class. Skip the classification step, and even a beautiful revenue table is just a string of characters hanging in the air. Compliance follows the same logic. Independent anti-doping bodies such as USADA or VADA only carry force once the discipline and governing body are identified. A negative test for one person cannot be inferred for another, and an empty result certainly cannot be read as a clean certificate. There is one more layer, and this is the layer I observed with my own eyes from the corner of the pitch. I once sat beside data people at a training session in Bangkok. They were good. Their tables were immaculate. But I noticed one detail: when a player ran to the touchline, they did not look at the player, they looked at the screen. Beside them, the coach only needed the sound of breathing and studs biting the turf to know who still had legs. I watch matches with my ears, and I hear the passes nobody made. A data desk's conclusions are usually right on the data but off on the rhythm. They measure what already happened, not what is still forming. And there is one more grey zone that very few newsrooms dare touch: brain health. Counting head strikes, counting knockdowns, tracking rest intervals after concussions, measuring weekly sparring load — that is the four-part set any serious combat-sports ecosystem should keep. No fighter name, no competition history, and the whole set evaporates. No one is found to be at risk, but no one is confirmed safe either. I used to think this was a dry technical story. On closer look, it is a story about power. Whoever holds the labelling power holds the concluding power. A wrong label stuck onto a fighter can follow him through an entire career, longer than any defeat. Media plays its part too: a rising face gets pushed upward with a halo faster than real achievements accumulate, and once the label is wrong, that halo is harder to remove. Now I will say what many colleagues will not like. The scariest part of this story is not the machine that chose silence. The scariest part is the remaining reflex of the whole industry: when data comes back empty, fill it with something that sounds plausible. The pressure to reach a strong conclusion outweighs the pressure to reach a correct one. And so people invent a handsome analytical framework, slap on a label at random, and present it as if it were fact. I once witnessed the opposite: an editor refused to publish an analysis built on unverifiable data. That piece never ran. Many in the newsroom counted it a failure. I considered it the most courageous decision of the week. In an industry where everyone fears a gap, whoever dares leave the gap standing is the most honest person in the room. That machine did exactly one thing right: it refused to lie to itself. It did not know what category the article fell into, so it did not pretend to know. Technically, that is a failure. On professional ethics, it is behaviour worth learning from. When everyone believes one truth, I start believing in error — but it must be an acknowledged error, not an error dressed up as truth. The question I leave for anyone building combat-sports analytics is not how to fill the empty data. It is this: do you have the courage to force the machine to answer the classification question before it computes the first metric? Because tactics never die, they are only forgotten until a madman dares revive them — and sometimes that madman is a blank line on a screen, daring to say it knows nothing yet.