Ding Liren, Gukesh, and the Variable Every Chess Data Model Missed

Singapore, the night of December 12, 2026. Game 14 of the World Chess Championship had reached its endgame. The position was balanced to a degree that defied belief. In the analysis booths, the models had finished running long before, and nearly all of them agreed: this game would be drawn. The previous thirteen games had left the score at 6.5-6.5. One accurate move from Ding Liren would keep his crown for at least another cycle.

Then he made the wrong one.

Not in the opening. Not in the middlegame. Ding Liren — widely regarded as the finest endgame technician of his generation — lost the thread exactly in the phase the data said he was strongest. Gukesh Dommaraju, an eighteen-year-old from India, did not miss his chance. An hour later the match was over at 7.5-6.5. Gukesh became the youngest world chess champion in history, at eighteen years and six months.

What matters is not the result. What matters is this: the variable that decided the game had never appeared in any data model — including the ones I helped build.

The Data Had an Answer, But Only Half of One

Ding Liren took the title in 2026 in Astana, beating Ian Nepomniachtchi. It was a strange match: the classical games ended level at 7-7, and Ding won only through the rapid tiebreak. He took the crown without ever building a lead in the main contest — a detail most coverage skipped, but one that matters enormously to anyone who works with data, because it showed Ding was the kind of player whose own results understate him.

His 2026 was, on paper, a catastrophe. He endured the longest winless run of his classical career, at one point fell out of the top group of the FIDE rating list, and spoke openly about his mental health. Gukesh was the opposite: he won the 2026 Candidates Tournament in Toronto, his form climbing without pause, beating major names at an age when most players are still relearning basic opening lines.

To see why analysts were so confident, place the two men in historical context. Before Gukesh, the youngest champion in history was Garry Kasparov, who took the title in 2026 at twenty-two. Magnus Carlsen took it in 2026, also at twenty-two. That record stood for nearly four decades. Gukesh broke it at eighteen. For anyone working with data, a player that young reaching the summit is not merely an inspiring story; it is a signal of an unprecedented rate of development.

Seen through the raw data, the story was clear: a fading champion against a rising force. Any analyst would rank Gukesh as the favourite. I have followed elite chess for more than thirty years and work as a sports data analyst in Chengdu, and I ranked it the same way.

The final result proved the direction of that prediction right. The magnitude was entirely wrong.

My model, like most at independent analytics shops, projected a lopsided match — Gukesh winning by two or three games. Reality: 6.5-6.5 after thirteen games, decided only in the final one. Ding Liren, buried by the data before the first game began, played on level terms and at moments pushed Gukesh into difficulty.

At the elite level, being wrong about magnitude means being wrong about almost everything. A model that predicts “Gukesh wins 7.5-3.5” is not better than one that predicts “Gukesh wins 7.5-6.5”; it is merely more confident. And confidence, divorced from accuracy, has no predictive value. Data never lies, but it likes to test our patience.

The Evidence Chain: What Actually Made the Difference

Two questions must be separated. First: who wins? The data answered correctly. Second: how, and by how much? The data answered wrongly — and that error is where the lesson lives.

Across all fourteen games, the move quality of the two players hovered around comparable levels. No game saw Ding collapse technically in the way his 2026 form implied. Gukesh — the higher-rated side — was actually the one who missed more chances in the middle games. Rely on pre-tournament form data alone and you draw a one-sided picture; the in-match data draws a two-sided one.

This is familiar to anyone who has done sports analytics long enough. Accumulated form is a trend indicator, not an outcome indicator. It tells you who is rising and who is falling; it does not tell you who wins a specific match. In football this shows up constantly: a team on a winning streak can still lose to anyone on a given night. World Cup 2026 did not change the rules of the game; it merely showed us the pattern that already existed — that form and outcome are two different things.

I experienced this on a smaller scale. In 2026, ahead of a Clasico between Real Madrid and Barcelona, I published a model built on hundreds of shots from both teams over the previous fifteen rounds, and it showed Barcelona clearly ahead on expected goals. The result was right on direction. But had someone asked me then “by how much will Barcelona win,” I would have had no data-based answer — and admitting that was the most honest part of the job.

In chess the gap between form and outcome is even wider, because there is no randomness of the ball-hits-the-post kind. Chess is a discipline where quality determines results almost absolutely over the long run. But “long run” is the key phrase. A fourteen-game title match is too small a sample for accumulated form to do its statistical work. In a small sample, the un-modelled variable dominates.

And here, what was the un-modelled variable?

The Data Void

Ding Liren’s psychological state.

I tracked every game Ding played through 2026, and one pattern held with strange consistency: in round-robin events he played below his level. In head-to-head match play, he played above it. That is a behavioural pattern form data cannot capture, because it surfaces only in a rare and hard-to-sample format.

No rating measures this. No engine metric quantifies the feeling of “I have nothing left to lose” in a champion the whole world has written off. Before the match, no analyst had a data field called “probability Ding summons his best exactly when it counts.”

That is what I call the data void — the space where a model has nothing to say, yet the public assumes the model has said everything.

Modern analytics makes a systematic error: it confuses “no data” with “no problem.” When a variable cannot be measured, the default response is to ignore it, rather than to acknowledge it exists. Ding Liren is living proof of that error. What he produced in Singapore was in no spreadsheet, because it never had a chance to enter one.

None of this means data is useless. It means data has boundaries, and professionals must know where they lie. A good model does not only tell you what it predicts; it tells you what it cannot predict.

In an empty hall, data is the only spectator left — and sometimes even it does not know what to cheer for.

The Counterintuitive Angle: The Industry Has an Incentive to Fill the Void

This is the part few in the profession will say plainly.

When data is insufficient, the honest choice is to declare “insufficient information to assess.” But that declaration produces no content, no views, no sponsorship. So the sports analytics market has a structural tendency: to fill the void with stories that sound data-driven.

A model that returns “insufficient information” is commercially useless. It may also be the most professionally correct answer. This is the core contradiction of the trade, and one I have met head-on: there were moments when I knew my data was empty but still had to write, because the newsroom needed copy.

In chess the pressure is heavier, because the public holds an almost absolute faith in the computer engine. If the engine evaluates a position at +0.3, people assume that is “the truth.” But an engine evaluation is only a number about a position; it says nothing about which player will err under pressure, in Game 14, after four hours of play, with the world title hanging in the balance. That evaluation is data. The question of what a human will do with it is not.

There is a striking parallel in football. The resurgence of the back three is not a tactical advance; it is coaches managing reputational risk after their back fours were torn apart. In chess, the dominance of engine-driven opening lines works the same way. When reputation is what is being staked, people steer the ship into waters already charted, even when those waters lead nowhere near victory. What both phenomena share is the fear of being judged wrong, disguised as rationalism.

Ding Liren, at thirty-two, did the opposite. He accepted risk. And across fourteen games of a match the data said he should lose heavily, he came within a whisker of winning.

There is a subtler reason the void gets filled: career risk. An analyst who dares say “I don’t know” is treated as incompetent, while someone who offers a wrong but confident prediction is usually forgiven. The market rewards decisiveness, even when that decisiveness rests on nothing. The result is a system that incentivises disciplined fabrication — paradoxical to the ear, but exactly what is happening.

What to Watch Next

The 2026 World Chess Championship — with Gukesh as the reigning champion and a fresh challenger emerging from the Candidates — will be the next test of this question. The issue is not who is stronger on the rating list. The issue is whether models will learn to quantify the psychological variable, or will keep filling the void with plausible-sounding stories.

I do not believe in the second possibility. A human being’s mental state cannot be reduced to an index, and the harder we try, the further we drift from reality. What I do believe in is the first — learning to acknowledge the void, and speaking up when the data is not yet enough. A good analyst is not someone who always has an answer. It is someone who knows exactly when the only honest answer is: not enough to say.

Ding Liren did not keep his crown in Singapore. But he left the analytics profession a lesson no spreadsheet can teach: that sometimes the thing that decides everything is precisely the thing we cannot measure.