The Empty Board and the Hunger for Analysis: When Empty Data Gets Read as a Game

On the night of November 14, 2026, the PGN feed from an international chess tournament broke mid-way through the fourth game. In the analysis room, the screen showed only an empty board — thirty-two pieces frozen on their starting squares, not a single move recorded. I sat there, holding a cup of coffee that had gone cold, waiting for the feed to return. Six hours later, when I reopened the forums, I counted four analysis pieces already published. Each described a different opening scenario. Each confidently pointed to a blunder in the middlegame. Not one of them — not even in a single sentence — admitted that its author had never seen a single move of that game. I opened the only piece of data I had: an empty shell, no title, no source, no information points. And I realized I was looking at the most serious blind spot in chess analysis in the digital age.

Twenty years ago, to commentate on a game you had to sit in the hall, writing down every move by hand. One mistake and the whole analysis collapsed. Now anyone can open a tab, download a PGN file, run an engine, and produce a three-thousand-word article in twenty minutes. That convenience created an illusion: that the data is always there, that having the tool means having the analysis. But the tool does not create data. The tool only makes empty data look full.

I came to chess in 2026, first as a young player then as a tournament organizer. I once sat in the commentary booth for VTC, narrating classic games from an era when fans still remembered individual moves by name. Back then data was scarce, so people did not dare to fabricate. Now data is everywhere, so they no longer fear fabrication. That is the paradox I want to dissect here.

Chess is the most tightly structured of all sports. A game has exactly one move sequence. There is no debate over whether the ball crossed the line, no referee's angle. Everything can be recorded, encoded, verified. For that very reason, chess should be the place where the empty-data problem occurs least. But the opposite is true. The tighter the structure, the easier the illusion slips through — because a text that looks structured is automatically trusted.

That night, the death of the feed was a test

PGN outages are not rare. Every season I witness at least a few server crashes on the live platforms. The question is not whether the outage happens, but how professionals respond when it does. There are two kinds of response. The first: stop, note that the data is missing, wait for the feed to return, and if it does not, publish a single line of notice. The second, far more dangerous: fill the gap with what you expect to happen.

The four analysis pieces that night belonged to the second kind. I read each one carefully. The first claimed White had avoided a familiar opening line. The second said Black blundered on move twenty-three. The third analyzed an endgame that never took place. The frightening part is that all four had data. They cited engine evaluations, win rates, head-to-head records. All of it flowed, all of it looked professional. And all of it came from an empty file.

This is not a new phenomenon. Across sports analysis, people call it the "data illusion": when a writer has no real data yet still produces a result that looks plausible. In chess the illusion is more dangerous, because chess allows absolute verification. A game has one truth, and any player can detect a fabricated move in thirty seconds. Yet that night nobody detected anything. At least nobody spoke up in the first six hours. Because most readers have no original PGN to check against. They only read fluent articles with numbers and charts. And their trust is stolen without their knowing.

The digitization of chess: from paper scoresheets to PGN files

To understand why this happens, we need to look back at the history of chess digitization. Before the 2000s, every game was written by hand on a scoresheet in algebraic notation. Large databases such as ChessBase emerged, gathering millions of games, turning chess into the sport with the largest data archive of all. By 2026–2026, online platforms like Lichess and Chess.com began streaming PGN live, meaning each game is transmitted in real time, move after move.

That change has two sides. The good side: analysts get instant data, engine evaluation of every move, continuously updated live ratings. The bad side: the entire analysis ecosystem depends on one feed. When the feed breaks, the line between "has data" and "has no data" blurs — because the interface still appears, the board still appears, only the moves vanish.

In that setting, an undisciplined analyst will not realize they are working with empty data. They see a board at the starting position and assume a game is in progress. They cannot distinguish "a game that has not begun" from "a game that took place but was not recorded". This is what I call the foundational perception error — it happens before any analysis, and it determines that every analysis downstream is meaningless.

For Vietnamese chess, the issue is more sensitive. Players like Le Quang Liem and Nguyen Ngoc Truong Son mostly compete abroad, and information about their games comes largely from international platforms. When the feed fails, domestic fans are the first affected, because they depend entirely on secondary sources. A fabricated analysis harms not only the media but also how fans understand the form of the player they love.

Three layers of a silent failure

When I sat down to analyze the incident, I saw it has a three-layer structure, like a game lost in the opening that one only discovers in the endgame.

The first layer is collection. The PGN feed crashed, and the system registered "no data". At this layer the signal should have been a red alert. But the system returned a valid empty shell: right structure, right format, no content. In engineering this is called "schema-valid, content-empty". It is the worst kind of failure, because it triggers no error. It stays silent.

The second layer is interpretation. Some automated tools, instead of reporting "no data", try to generate analysis out of nothing. They rely on probability, on common opening patterns, on expectations about the two opponents. The result is a text that looks plausible but is not real. Engineers call this a "hallucination". In chess it happens more easily because games have highly repetitive structure — a game between two strong players usually begins with familiar openings. The fabricator only has to stitch familiar fragments together.

The third layer is publication. An untrue analysis, once published, becomes a reference for later pieces. The next day people cite it. The next week it appears in a roundup. The next month it becomes "historical fact". This is the mechanism I call "negative contamination" — not false information spreading, but missing information disguised as full information.

The three layers together produce a paradox: the more data, the more analysis, and the less truth is verified.

No tactic is ever old; only the way we read the game expires.

I wrote that line for football, but it applies to chess painfully well. The way chess analysis reads the game has expired at one very specific point: we trust the tool more than the source. We check whether the engine is strong, but not whether the PGN file is real. We argue about Stockfish versus Leela, but nobody asks: where does the input data come from?

I once saw an editor reject an analysis of a game simply because the author could not cite the PGN source. Back then I thought he was too strict. Now I think he was the only one doing it right.

The principle I drew after November 14 is simple, and it is exactly the principle I have applied to every football analysis for eight years: before analyzing, confirm that the object of analysis exists. An empty PGN is not a chess game. It is an invitation to fabricate.

The minimum threshold for starting an analysis

In my data work I set a hard threshold: a record is eligible for analysis only when it contains at least three independent information points and at least one identified entity. In chess, that equals: at least two player names, an event and round, and at least one reference to a move, an ECO code, or a rating value. If all three are missing, the system must raise an error and stop, not return an empty shell that looks valid.

It sounds like a dry technical rule. But it is the only barrier between analysis and fabrication. The problem with most current systems is that they are designed to "never fail" — meaning, when they meet empty data, they still return a result, even a meaningless one. In software engineering this is called "silent failure". And in analysis, silent failure is many times more dangerous than loud failure.

A good system must be able to say "I don't know". A good analyst must too.

I remember 2026, when the pandemic stopped football worldwide and my data company tasked me with building a win-probability model for the first ten rounds after the Bundesliga resumed. I used the previous three seasons and found an anomaly: teams with an xG differential better than 1.5 before the pandemic won only 4 of 10 opening games after restart, 23% below the historical average. Management doubted it because there was no precedent. I persisted in explaining the method and added a new variable: days of rest between matches. My model predicted 7 of the first 10 games correctly, while the old models got only 4. The lesson was not in the number. The lesson was this: when data clashes with expectation, the right response is not to discard the data, but to question the context.

In 2026 I discarded half of my old dataset, because football after the lockdown is a different sport. But I did not discard the principle. The principle "verify the source before analyzing" does not expire. It only needs to be applied more strictly.

What nobody wants to say about the anti-cheating craze

The entire chess world is pouring energy into one threat: engine cheating. Tournaments install scanners, check latency, ban electronic devices, analyze engine-match rates. Those efforts are right and necessary.

But there is another, quieter threat almost nobody mentions: analysis content with no underlying data. Engine cheating corrupts one game's result. Empty analysis corrupts an entire generation's understanding. The first can be detected and punished. The second cannot, because nobody checks a fluent article.

I am not saying every analysis is suspect. I am saying we are measuring risk wrongly. We fear a player using an engine to win. We do not fear an analyst using imagination to fill a gap. But in the long run, the second is what erodes trust.

A heat map can lie, but five consecutive failed presses cannot. In chess this becomes: an evaluation chart can lie, but a verifiable move sequence cannot. The problem is we are so used to charts that we forget the move sequence.

Esports taught me what chess is relearning

There is one field where this problem was solved long ago: esports. In esports every match is recorded at the frame level. There is no such thing as "I don't have the match data". If a match is not recorded, it does not exist in the ranking system. Major esports platforms have mechanisms to refuse to register a result when the original data is missing.

Esports taught me what football hides well: a reaction within 200 milliseconds can break an entire system. And in analysis, a decision made in a moment — the decision to publish before the data exists — can also break an entire system of trust. Chess, as the most tightly structured sport, should have been the pioneer in fighting the data illusion. Instead it is falling behind.

So what should be done, concretely

First, every analysis must carry a source declaration: where the PGN came from, when it was downloaded, who provided it. Without this line, the piece is not analysis, only commentary.

Second, every automated system must have a minimum threshold. If the input data does not meet it, the system must stop and raise an error, instead of returning a valid empty shell. "Loud failure" is always better than "counterfeit success".

Third, every analyst must record their own data-collection status. When I write about a game, I save the original PGN file, and note any part I could not verify. Readers have the right to know what I see and what I do not.

It sounds like a lot of procedure. But remember: the price of a wrong analysis is not a bad article. The price is a contaminated dataset, and from it, hundreds of later articles inheriting the original error.

The game that never took place

Back to November 14. By the next morning the PGN feed was restored. The real game appeared, and it resembled none of the published analyses. White avoided no opening line at all. The middlegame blunder did not occur on move twenty-three. The endgame was entirely different. Those four analyses, checked against the record, were all wrong — not wrong in perspective, but wrong in the most basic fact.

What troubles me most is not that they were wrong. It is that they existed. Someone, while waiting for the feed, decided that an empty shell was enough to write three thousand words. And in a field where determinacy is the most sacred property, that decision slipped through.

I kept the empty file from that night on my hard drive, naming it "game_04_empty". Sometimes I open it and look. It is a reminder that, in this work, the greatest courage is not making a bold claim. The greatest courage is admitting: tonight I have nothing to write.

What I will check in the next game

Whenever I begin a new analysis, I will do something I once took for granted: confirm that the game I am about to analyze actually exists — with player names, event name, and a verifiable move sequence. If not, I will write a single line: not enough data to analyze.

Chess is entering an era of unprecedented data. But more data does not mean more understanding. Sometimes the biggest step forward for an analyst is learning to stay silent while the board is still empty. Because a game that never took place has nothing to analyze — only fabrication, and fabrication is never analysis.