The Blank Analysis Board: The Discipline of a Chess Analyst

In March 2026, I sat in the press room of a chess tournament in Saigon with a blank analysis board in front of me. No Elo, no sample games, no opening system, no ACPL, not a single line of data. The organisers asked me what I made of a player who had just been handed a wildcard. I answered: “Not enough data to conclude.” The room went quiet for a few seconds, then someone laughed: “Some analyst, with no opinion at all.”

I did not argue. At home I left the page white and added one sentence: “No conclusion.” Years later I still keep that file on my drive as a reminder. A blank analysis board is not a failure on the analyst's part. Sometimes it is the result.

What a chess analyst actually owns

Unlike football, where one season hands me several thousand possessions to count, chess is a sport with a tiny sample. A top player plays roughly 70 to 100 rated games a year. A specific opening variation in a specific tournament may appear three times. Three times is not a trend. Three times is an anecdote.

The data set I work with daily includes the Elo rating and its deviation, number of games played, ACPL, engine match rate, opening repertoire, time control, head-to-head record and the age curve. Of these, Elo is the most misunderstood. The FIDE system only issues an official rating once a player has accumulated 30 games; below that threshold every number is provisional. Which means some young players walk into a major event carrying a figure that the system itself says should not be trusted.

ACPL and engine match rate have their own limits. ACPL punishes correct but unconventional moves, while engine match rate rewards playing like the machine, and playing like the machine against a weaker opponent is sometimes how you lose. That is why I always read a metric together with its context of time control and opponent.

There is another kind of data a board rarely displays: absent information. When a player avoids a certain opening line for six months, that absence is itself a signal, either they are hiding it or they have abandoned it. The weak analyst reads what is there; the good analyst reads what is not.

Most wrong conclusions do not come from wrong data, but from using correct data with the wrong sample size.

Three times the story outran the data

Number one: the 2026 Candidates Tournament in Yekaterinburg. On 26 March 2026, organisers suspended the event after round 7 because of the pandemic. At that point Ian Nepomniachtchi and Maxime Vachier-Lagrave shared the lead on 4.5 points. The media immediately built a “momentum” narrative: the leader is in form, nobody can stop him. I wrote that momentum after seven games is a concept with no statistical basis. Seven elite games, four of them draws, cannot separate signal from noise.

There is a detail few mention: once the event stopped, every player's rating froze. Thirteen months without a rated game means the number you are looking at no longer describes the person sitting at the board. It describes someone who no longer exists.

The event resumed on 19 April 2026. Nepomniachtchi won with 8.5 out of 14. The result was right. But I have to correct myself here: a correct result does not prove a correct argument. If I am right for the wrong reason, I am still wrong. What I defend is not a prediction about Nepomniachtchi, but a principle: you do not draw a trend line from seven data points.

Number two: London 2026, the world championship match between Magnus Carlsen and Fabiano Caruana. All 12 classical games were drawn, the first time in history that a title match went through the entire classical portion without a single decisive game. Carlsen won the rapid tiebreak 3-0. The common reaction: “chess is dying”, “both players were too safe.”

I read it the other way. Iran in 2026 were not parking the bus; that was how they rebuilt space metre by metre. Likewise, the 12 draws in London are a data set about defence at the highest level, not about decline. Both teams prepared so thoroughly that one system cancelled out the other. A draw at the 2800 level rarely happens because two players lacked courage; usually it happens because two players had calculated everything.

But there is another variable few people mention: the format. Deciding the title in a rapid tiebreak changed the incentive structure. If you know you are strong at rapid, the optimal strategy across 12 classical games is not to lose, not to win. The format does not merely record the result. The format produces the result.

Number three: the cheating conversation in online chess. After 2026 the game moved heavily online, online events multiplied, and a wave of suspicion came with them. In late 2026 a former world champion publicly questioned a young grandmaster over online rapid events. The platform that ran the events reviewed the games and announced it had found no evidence of a violation.

What interests me here is not who was right. What interests me is that both sides were forced to trust a black box. The anti-cheat algorithms used by online platforms are not published: no sample size, no sensitivity, no error rate. When a governing body does not publish its method, people stop arguing about data and start arguing about reputation. That is the environment in which rumour carries the weight of evidence.

Where the data is thinnest and most ignored

There is one area where the sample-size problem is more severe than anywhere else: women's chess. Fewer elite women's events exist, and open-event slots are scarcer still, so the total number of rated games for a top female player is usually lower than for a male colleague of comparable strength. The consequence is that every judgement about them, “she is rising”, “she is finished”, “she is not quite there”, is built on thinner data. I have gone back and checked my own work more than once and found sentences I would never have dared to write about a male player.

That is why I give at least a third of every piece to defensive structure and empty data zones, rather than only praising the winner.

The blind spot lies elsewhere

People assume the greatest risk in this profession is drawing the wrong conclusion. For me, the greater risk is drawing a conclusion too early simply because a blank board looks bad in an article. There is a very real professional pressure: you are paid to have an opinion. Say “not enough data” three times a month and the editor will go find someone else.

I have been on the other side of that pressure. In 2026, when I built a probability model for the return of league play, I discarded nearly half of the old data set because the long layoff had shifted the distribution of results. But I also learned that not every disruption justifies throwing the whole data set away. You have to separate two things: the noise that must go, and the principle that must stay. Chess is still chess; controlling the centre is still controlling the centre. What changed was the pace, the schedule and the way people prepare.

No tactic is ever old; only the way we read a game expires. And the fastest way for a read to expire is to trust a trend line drawn from three data points, or to trust a number that comes from a system you cannot verify.

In 2026, when football and chess both stopped, I threw away half of my old data set, because the world after the lockdown is a different sport. I do not say that to show off methodological courage. I say it to remind myself that discarding data is a decision, and every decision can be wrong.

What I carry into the next event

The blank board of 2026 taught me something simpler than any model: the analyst's discipline lies in knowing when not to say anything. A heat map can lie, but five consecutive failed presses cannot. In chess, “five consecutive failed presses” is the equivalent of five games repeating the same structural error, not one game lost through a lapse in concentration.

The next event, I will again sit in front of boards packed with data. But I carry a question more uncomfortable than any question about openings: does the number I am looking at have a large enough sample to answer the question I am actually asking?

If the answer is no, I will leave the board blank. And this time, I do not think I will be the only person in the room who understands why.