Chess and Data Discipline: Why an Unverifiable Number Is More Dangerous Than Silence

In September 2026, at the Sinquefield Cup in Saint Louis, a game ended and the entire chess world erupted into argument. Magnus Carlsen, the reigning number one, lost to Hans Niemann and then abruptly withdrew from the tournament. Days later, he made a statement implying his opponent had cheated. For weeks, millions of fans took sides, each convinced they held the data. But when I sat down, opened every scoresheet and cross-checked every indicator, I realised something simple and uncomfortable: that moment did not lack data, it lacked verifiable data. People call that a shock; I call it unread data. In thirty-seven years of observing the chess industry, from player to tournament organiser to data journalist, I have seen many cases where a conclusion was reached before the numbers could be verified. And almost every time, a wrong conclusion caused more damage than an information gap. An honest gap forces us to wait. A fabricated number waits for no one; it spreads and drags an entire system of false belief behind it. That is why this article exists. Not in the name of a specific game, not to judge any individual, but to dissect how chess data is created, circulated and distorted. I want to show you why, in a sport where every number can be looked up, the most dangerous thing is false certainty. Context: how chess became a data sport To understand why chess is fertile ground for data analysis, one must look back nearly two decades. In 2026, when I was still organising small tournaments and writing for a few regional outlets, a top-level game was evaluated mainly by a commentator's feel. Players were described with adjectives: sharp, solid, reckless. Nobody measured how much worse a move was than the best one. Nobody quantified an opening advantage. The turning point came when analysis engines became strong and cheap. Stockfish, one of the strongest engines today, can evaluate any position dozens of moves deep at a speed no human can match. From that came a new set of metrics. First the centipawn, a unit of advantage in which 100 centipawns equals roughly one pawn. Then ACPL, Average Centipawn Loss, the average centipawn loss per move. It tells you how much advantage a player concedes on average compared with the engine's perfect move. Alongside it stands the Elo system, a relative-strength measure created in the 1960s and used officially by FIDE, the International Chess Federation. Elo reflects the win probability between two players based on rating difference, not the absolute quality of each move. Besides classical Elo there is rapid, blitz and bullet Elo, corresponding to different time controls. A player can top the blitz list yet sit far lower on the classical list, and that gap says a great deal about their style. One more layer is game databases such as ChessBase and The Week in Chess, storing millions of historical games. They let you look up how often an opening variation appears, find a new move never recorded before, called a novelty, and measure a player's understanding of an opening system. And of course there are online platforms such as Chess.com and Lichess, offering detailed statistics on play frequency, win rates by format and even user behaviour. The key point is that each data source answers a different question. Elo answers who is stronger. ACPL answers whose moves are more accurate. The opening database answers who prepared better. Platform statistics answer who plays more. Mixing them, or using one metric to conclude about a question it was not designed to answer, is the most common mistake I see in the trade. That is precisely the foundation for my three-source discipline, which makes me famously the slowest writer in the newsroom: before drawing any conclusion, I must cross-check at least three independent sources, and ask whether the third is genuinely independent or merely copying the first. In chess this trap is especially subtle, because three different articles about the same tournament often all cite a single scoresheet from the organiser. The core: five levels of reading chess data Once you understand the data ecosystem, the next job is to read it by level. Below are the five levels I apply whenever I analyse a tournament or a player, moving from micro to macro. Level one: the game and technique, where ACPL never lies At this level the central question is how the game unfolded. To answer it, I re-evaluate the whole game with an engine and compute ACPL for each player. Note that a low ACPL does not automatically mean a win. A player can play with extreme accuracy and still lose because the opponent was even more accurate at exactly the decisive moments. What ACPL truly measures is consistency, not outcome. Numbers are asceticism: you must give up convenience to see the truth. When I read a game, I split it into three phases: opening, middlegame and endgame. In the opening, the interesting metric is not ACPL but engine match rate, the share of moves matching the engine's first choice. A high match rate in the early phase often reflects solid preparation rather than in-the-moment talent. This is where the media often errs: they call a perfect opening a sign of genius, when it is usually a sign of a good preparation team and a vast database. In the middlegame the picture changes. This is where a player's nerve and calculation are tested, so middlegames rarely match the engine completely. I usually flag moves where a player deviates from the engine beyond a threshold, then check whether they recovered. Deviations that still win often signal a human idea the machine did not see immediately, while deviations that lose are usually genuine errors. Telling the two apart requires reviewing what followed, not just the score at the moment. In the endgame the story differs again. As pieces come off, advantage is usually quantified with tablebases, and the margin is so small that one slow move can turn a win into a draw. This is the phase where technical metrics outweigh spirit. A player with good endgame technique can hold positions that look lost at a glance. Here is an example of how I read technical data: in many top events, thinking time is unevenly distributed. Some players think very long in the opening to recall prepared lines, then move fast in the middlegame, and vice versa. If you only look at total time, you miss valuable information. I always separate time by phase, because it shows whether a player is relying on memory or on in-the-moment calculation. In the classical format each side gets about 90 minutes for the first 40 moves plus an increment, while rapid and blitz compress time sharply, making decisions more impulsive and errors spike. This is why I always state the time control before comparing any metric between two games. Level two: players and data, where Elo and form often diverge This level answers where a player sits in the hierarchy. Classical, rapid and blitz Elo are the first three axes. But as I said, a rating means nothing unless placed against context. A 2750 player in an upward phase has a completely different predictive value from a 2750 player in decline. So I always add a metric I call recent form, computed from a player's results in their last ten games against their Elo expectation. If a player wins notably more than expected, there are two possibilities: they are genuinely rising, or they got lucky with an easy schedule. Telling the two apart requires examining opponent quality. This is where pure statistics easily mislead the reader. The second metric is performance rating, the Elo equivalent of a player's result in a specific event. A player can enter a tournament at 2700, score a 2850 performance, and that shows they played above their level. But a high performance in a short event may not be sustainable, because the sample is too small. I always ask: how many consecutive events has this player sustained that form? If just one, it is noise. If three or four, it is a signal. A tool I use often is the head-to-head table. In chess there are pairs whose head-to-head results diverge sharply from Elo expectation. This is often called a bogey relationship. But be careful: a few consecutive wins against a strong opponent may be pure luck, especially when the number of games is small. I only treat a bogey relationship as real when it spans years, formats and contexts. This is the correlation-is-not-causation principle I apply strictly. One more variable I never skip: age. Age is the only variable that never lies. In chess, peak form usually falls between the late twenties and early thirties, though some exceptions stretch to nearly forty. Notably, the rate of decline varies between players. Some keep their calculating strength to nearly forty, others slide after thirty. I always place form alongside the age curve to avoid hasty conclusions about an older player doing well, or a young player stalling. Level three: tournament systems, where the qualification path decides everything This level answers what an event means in the big picture. In elite chess, events are not equal. A world championship match carries a weight completely different from an open tournament. The Candidates Tournament, which determines who challenges the world champion, is where every top player longs to appear, and to get in they must pass through several routes. The first route is the World Cup, a large knockout event with hundreds of players, where a single bad game eliminates you. The second is the Grand Swiss, a long Swiss-system event rewarding stability rather than explosion. The third is the rating spot, for players who sustain a high ranking over a period. The fourth is a wild card, usually for players with special achievements or value to the event. Each route has its own character and pressure. Knockout events force players to accept high risk, while Swiss events reward patience. Understanding this helps me assess how much a Candidates spot is worth to each person. For a rising young player, a Candidates spot is a golden opportunity. For a former champion, it is a last chance. I also always check the field quality: the average strength of participants, prize scale, draw rate and watchability. A tournament with a high average strength but a huge draw rate tends to be less appealing to viewers. This is the paradox of elite chess: the more strong players, the more draws, because the margin is so small that everyone avoids defeat. Organisers often try to cut draw rates by changing rules, for instance banning early draw offers or applying the Sofia rules, forcing players to fight on in certain phases. Another factor I always watch is the schedule. In long events, the arrangement of rest days and the order of opponents can affect the final result. A player facing strong opponents consecutively will be exhausted, while another has a lighter path. I often draw a chart of opponent difficulty by round to see who enjoys a scheduling edge. Level four: competitive landscape, where generations collide This level answers who dominates and who is coming. In modern chess, the competitive picture can be split into four tiers. The first is the champion or the current number one. The second is the 2700-plus challenger group, players capable of winning any event. The third is the rising-star group, young players with fast-rising Elo. The fourth is the reserve pipeline, young players not yet at the top but with clear potential. What stands out in recent years is the new generational wave, especially from India. A string of young players has climbed into the top group within a few years, breaking the dominance of traditional chess powers. This is a quantifiable phenomenon: the average age of the leading group is falling, and the Elo growth rate of young players is far faster than the previous generation. When I chart age against Elo over time, the trend is unmistakable. But I always remind myself not to rush. A young player rising fast may reflect three things: real talent, playing more events, or luck in some games. Only when they sustain form over two or three consecutive seasons do I treat it as a genuine generational shift. Chess history is full of players who exploded for one season then vanished. This is why my long-term cross-check discipline matters. One must also look at national systems. Chess powers do not merely produce exceptional individuals; they build an entire training system. India, China, Russia, the United States and Uzbekistan each have a different model. India has a strong mass chess movement with many junior events. China has centralised training. Russia keeps a long-standing chess-school tradition. The United States ties chess to universities. Each model produces a different kind of player, and that shows in their playing style. In women's chess the picture is also shifting. For decades a few women proved they could compete with men at the highest level, and major names left deep marks. Recently the number of young women entering open events has risen, though the average Elo gap remains. This is a long-term indicator I track, because gender change in chess will come slowly but surely. Level five: rules and governance, where rulings must rest on evidence This level answers who sets the rules and how they are enforced. In chess the highest governing body is FIDE, alongside continental federations, national federations and event organisers. Common legal issues include anti-cheating, format and tiebreak rules, eligibility and registration, and governance procedures. The most sensitive issue is anti-cheating. In chess, cheating usually involves engine use, and detection requires statistical models and on-site security measures. This is an area where I am especially cautious, because a false accusation can destroy a player's career. My principle is clear: never speculate without verifiable evidence. The silence of data does not mean innocence, but it does not mean guilt either. In recent chess history there have been cheating cases that caused shockwaves, and each left lessons about process. When an accusation is made without public evidence, it turns into a media war in which the truth is sidelined. Conversely, when evidence is fully published and the process transparent, the chess community can accept the outcome however harsh. Another legal topic is federation transfer. A player may change the national federation they represent, and this sometimes triggers disputes over rights and waiting periods. The issue often concerns opportunities to play team events such as the Chess Olympiad, where each country fields a team. Federation-transfer rules aim to prevent a player from switching nationality purely to gain a spot. Finally there is tiebreak resolution in special formats. When a classical match ends level, organisers often use rapid or blitz games to decide. In a deadlock, the Armageddon format may be used, where one side gets more time but must win. These rules directly affect player tactics and how I read results, because a win in a tiebreak is not fully equivalent to a win in a classical game. A contrarian angle: the silence of data does not mean innocence This is the part I want you to read slowly. In my trade, the most dangerous mistake is not a lack of information, but turning that lack into a conclusion. When a data table is empty, a database returns no results, or a source is blocked, the natural human reflex is to fill the gap with guesswork. In chess, where every number can be looked up, such guesswork can be caught at any moment. I once witnessed a typical case. A magazine published a piece about a young player with impressive figures, but on checking, the number came from a single unverifiable source, and two other articles simply quoted it. Three sources, but really just one. My three-source discipline is not about counting sources, but about checking whether they are genuinely independent. This is also what modern search algorithms seek to reward: content with information gain, writing that brings the reader something they did not already know. If an article merely repeats what exists, it adds no value. If it invents a number to look original, it causes harm. Honesty about the limits of data is a form of information gain few consider. In chess this matters more than in other fields, because data here is verifiable. If I write that a player scored a 2800 performance in an event, readers can look it up and check. If I write that some player cheated, I need evidence, not a feeling. Precisely because verifiability is high, the damage from error here is also highest. A fabricated number in this field is not merely wrong; it betrays reader trust and harms the very subject it describes. I call this the fabrication pressure of a fact-dense field. When everything can be verified, the writer is tempted to insert specific details to look credible, even when they lack them. The storyteller's instinct pushes us to fill the void. But in my work, an acknowledged gap is worth more than an invented detail. There is another fear I want to name: the fear of appearing inadequate for admitting you lack data. In sports media, where everyone wants to seem knowledgeable, saying there is not enough information to conclude is often seen as weak. But I have learned, over many years, that real strength lies in tolerating uncertainty. A good analyst is not one who always has an answer, but one who knows when the answer cannot yet be given. The takeaway In elite chess, data has become the common language. Elo measures strength, ACPL measures accuracy, performance measures form, the head-to-head table measures relationships, and age measures time. A good chess reader knows what each number answers and what it does not. A good data journalist knows when to stop when the number cannot yet conclude. The next-round signal I am watching is not a specific player, but how the chess community handles data gaps. If more people dare to say they lack information instead of guessing, the sport will be stronger. For in the end, what keeps chess alive is not a perfect move, but the belief that the truth can be verified.