An Eight-Part Chess Analysis With Not a Single Move in It

My notebook records March 14, 2026, at 2:07 a.m. Beijing time. A forty-one-page chess analysis had just landed in my inbox. The title was properly formatted. The structure had eight parts, numbered from 0 to 8. Eight data tables with bolded column headers, neat rules, not a single formatting error.

I read the opening and nodded. I scrolled to line thirty. Then I stopped.

No player names. No tournament name. No opening code. Not one Elo value. No move count, no average centipawn loss, no engine match rate. Not a single move.

The document had a dedicated cell for "classical rating." It read: insufficient information. It had a cell for "engine match rate." That one read exactly the same, word for word. It had a seven-row risk matrix, a three-stage transmission diagram running from youth training to derivative markets, a three-column expectation-gap table. All empty. All looking extremely serious.

It took me ninety seconds to understand what had happened: the data pipeline had failed, and it had failed in silence.

Chess is the most thoroughly digitized sport of them all. Every game, whether played at a world championship or in an online arena at three in the morning, leaves behind a PGN file recording every move along with thinking time. Rating lists are published on schedule. Every move can be checked against an analysis engine. Football has no such transparency, athletics has none, swimming has none.

That is exactly why the chess content industry believes it sits in a safe zone. If the data is already in the machine, all that remains is to pull it out.

Operations look different. The monthly volume of chess events is far too large for any newsroom to read by hand: hundreds of national tournaments, dozens of online arenas running continuously, continental federations issuing their own bulletins, and a total number of recorded games each year reaching into the millions. That pressure pushes sports newsrooms toward a pipeline model: ingest, extract, analyze, publish.

In football, a failed ingest leaves an obvious mark. No lineups, no scoreline, no match report; an editor spots it immediately. In chess, a failed ingest can still leave behind a properly formatted document with full headings and full tables. That document does not look like an error. It looks like a cautious piece of analysis.

My own tracking experience taught me something opposite to common intuition. In 2026, at fifty-five, I sat in the commentary booth in Nizhny Novgorod for the France-Uruguay quarter-final. Real-time player-tracking software showed me for the first time that France's midfield took an average of 5.2 seconds to close down after losing the ball, against a tournament average of 7.8 seconds. I put that measurement on air mid-match, even though the audience could not see my screen.

What I learned afterward had nothing to do with how interesting the measurement was. What I learned was this: a measurement is worth only what its provenance is worth. I had to review the entire match footage and confirm the alternating pressing pattern used by Didier Deschamps before I dared to write. Without that step, 5.2 seconds is just a pretty fragment of data with no foundation.

Since then, every piece I write opens with its own spreadsheet. In 2026, when every competition stalled and stadiums sat empty on screen, I spent six months digitizing handwritten notebooks covering 2026 to 2026, 2,400 matches in total. I did that not because I love spreadsheets. I did it because I wanted to know where my own data had holes.

Everything on a chessboard is data waiting to be read — if the reader is willing to sit down before opening the spreadsheet.

Where did that night's report fail? It failed at the lowest layer, and the layers above kept running as if nothing had happened.

A sports content pipeline has three layers. Layer one: ingest. Layer two: extraction and entity tagging. Layer three: interpretation and presentation. In this document, layers one and two had both broken. Layer three completed its task exactly as designed, meaning it still produced a complete eight-part text with an introduction, conclusions, and a methodology note.

The cause of the ingest failure was almost certainly an ordinary technical barrier: a paywall, a block on automated access, a single failed fetch, or a scraper returning an empty page shell instead of real content. That class of error happens thousands of times a day across the entire online content space. On its own, it is not worth mentioning.

What is worth mentioning is that layer two never sounded the alarm.

The extractor was built schema-first: it constructs the container first, then goes looking for content to fill it. When it finds no content, it does not raise an error. It fills the container with default values. And the default value of a sports analysis system is insufficient information.

The result is a document that looks entirely normal. It even looks more trustworthy than normal, because caution always creates an impression of professionalism.

Here lies a distinction most people in sports content have not drawn clearly: a document can be structurally valid and entirely empty in content, and the naked eye cannot tell those two states apart.

That forty-one-page report was correct in every formal respect. It complied exactly with the designed schema. It carried all eight analytical dimensions: game technique, player and rating data, tournament system, competitive landscape, rules and governance, risk, public narrative, and industry transmission. It had all the tables, all the cells, all the arrows.

It lacked one thing: information.

More precisely: no player was identified. No tournament was identified. No opening code. No rating value of any kind, whether classical, rapid, or blitz. No head-to-head record. No thinking-time data. No average centipawn loss per move, no engine match rate, no figures on opening preparation depth.

Across the entire document, the only surviving signal was a domain label: chess. That label says the router classified the subject correctly. It does not say that any chess content was read. A correct domain label is evidence that classification ran, and nothing more.

I have encountered this exact failure mode in manual form. In 2026, I published a prediction that the young Spanish midfielder Pedri would be the player covering the most ground at the European Championship, averaging 11.7 kilometers per match. When the tournament ended, he had covered 11.8 kilometers per match. A hundred-meter error per match, after I had re-measured my spreadsheet four times. A young talent exists before the tournament, but most of us only see him after the spreadsheet speaks.

That match did not come from me being good at prediction. It came from me refusing to fill in a value I had not measured.

The real damage of an empty report is not in the report itself. The damage is in the next step.

A data repository that does not record extraction failures will convert ignorance into an assertion.

The mechanism is simple. If the empty report is stored with a status of processed, then three months later, when someone queries the repository for cheating controversies in chess, the result returned for this article will be zero. That result does not carry the label could not be read. It carries the label does not exist.

Chess is especially vulnerable to this kind of distortion, because so many conclusions in the game take negative form. No novelty was found in this system. This line has been refuted. There is no drawing tendency at this level. Each of those sentences can be a genuine finding, or it can be a PGN file with missing moves.

I have seen the same consequence in numerical form. Average centipawn loss is the most widely used index for judging a player's accuracy, and it only means anything when the move list is complete. If ten moves are missing from the middle of a game, that index does not produce a gap. It produces a value that looks real. That value will be cited, compared, entered into rankings, and eventually used to conclude that one player played more accurately than another.

The same mechanism applies to engine match rate, to thinking-time data, and to every aggregate index nobody bothers to verify move by move.

There is a list of things that report should have said, and I refuse to fill them in on its behalf.

It should have raised the gap between the world's number one player by rating and the world champion. This is the single largest structural feature of the chess world today. Magnus Carlsen holds the number one position by rating, while the world championship has belonged to Gukesh Dommaraju since December 2026, after he defeated Ding Liren. Those two positions do not coincide, and every serious discussion of elite chess has to pass through this point.

It should have analyzed the dividing line between over-the-board chess and online chess. Online results cannot be directly extrapolated into classical strength, and any ranking that blends the two sources without a note is being unintentionally dishonest.

It should have calculated the qualification path to the Candidates Tournament, compared points among direct rivals, and placed the result at the correct stage of the championship cycle.

It said none of that. And I will not write it out as though it had. The silence of a report proves nothing, including the things it ought to have said.

If I had to rebuild that pipeline, I would install four safeguards. I will state the method plainly, because hoarding the formula is a bad habit I managed to drop after many years.

The first safeguard is a minimum threshold for counting an ingest as successful. A source counts as having content only when at least one of five signals exists: a player name, a tournament name, a rating value, a reference to a game or move, or a reference to a governing body. No signal, no article.

The second safeguard is a yield check before the analysis layer is allowed to run. A minimum of three information points and a minimum of one identified entity. Below that threshold, the system must halt and raise an alarm, rather than being permitted to emit a valid-looking schema.

The third safeguard is retaining the raw text alongside the structured data. When the extractor performs poorly, the reader must still have a path back to the original source.

The fourth safeguard is recording provenance at the moment of ingest: URL, publisher, timestamp, and access result. Without those four fields, nobody can audit the reliability of any conclusion drawn later.

All four of these are cheap. The cost of not doing them is not cheap at all.

The counterintuitive angle sits here: that empty report was the most honest document I received that month.

A system willing to print forty insufficient-information cells did exactly one thing most sports content systems do not: it did not fabricate. The paradox of this industry is that rewards flow toward fluent documents, while null results are deleted from the folder before anyone reads them. No editor publishes a piece titled we could not read anything. They publish one with player names, percentages, and charts.

In 2026, at the Paris Olympics, a women's basketball team asked me to analyze their rebound positioning. I agreed on two conditions: no television appearances, no listing on the coaching staff. My fourteen-page analysis showed the team was losing an average of four points per game because players chose their waiting positions against the referee's direction of movement. That document had value because it measured every single player. A document that measures nothing has no value, however equally thick it may be.

Structured emptiness is a form of honesty. Ungrounded richness is a form of fraud.

A sports content pipeline and a rating list differ only in the screen; the operating system underneath is identical. Both live on input data, both die from input data, and both are capable of hiding their own death behind a handsome interface.

There is one assumption I want to remove. People often worry that machines will fabricate chess analysis that never happened. In this case, the pipeline did not fabricate. It did something harder and rarer: it refused. The fabrication would come from a writer under deadline pressure, or from a language model asked to analyze an empty container — and that model will return plausible-sounding chess content rather than return blank space. What needs guarding is the system that decides what counts as readable.

Chess carries a particular temptation. Because an analysis engine can verify every single move, people assume the numbers in this game cannot be faked. That is true of move evaluation. It is not true of aggregate indices. Nobody goes back to re-check a published average centipawn loss table. Engine evaluations are auditable; aggregate tables are not.

I no longer believe in miracles on the chessboard; I believe only in the conversion rate of an advantage. And that rate exists only when the data exists.

What I want to see in this major-tournament season is a mandatory note at the foot of every automated analysis: how many signals this piece read, from which source, at what time, and how many cells were left blank. A nutrition label for sports data, printed publicly, with nobody's permission required.

Football, chess, athletics, swimming all speak the same language. That language is trustworthy only when the speaker knows where he is being silent.