Eight Sections, Empty Core: A Warning From a Chess Analysis System That Returned Nothing

Late one November afternoon I opened a forty-page report. It had a title. It had a table of contents. Eight chapters, neatly numbered, each one broken into charts, conclusions, evidence, risk warnings. At a glance it was the product of a serious process. By page three I noticed something wrong: every data cell read "insufficient information". No tournament name. No player name. No game count. No Elo rating. The information-points list, the spine of the whole document, was completely empty. Eight chapters. Not one fact. I had felt that before. In 2026, aged twenty-five, I was an assistant analyst for the SHB Da Nang coaching staff. In the round-12 V-League match against Hanoi FC, I misidentified the position of a corner kick in the 63rd minute. The error was not in the number itself, but in the fact that the head coach read my notes and misread the defensive shape. I filed the wrong SHB Da Nang tape that year, and I learned that football does not forgive carelessness. Vietnamese football now runs on data. V-League clubs use GPS vests to measure distance, sprint counts and the gaps between lines. Chess squads use Stockfish and Leela to compute ACPL, the average centipawn loss per move, along with engine match rate. Two different sports, one shared stage: extracting information from raw sources before any analysis happens. That process has two layers. Layer one reads the article and pulls out facts: names, events, dates, numbers. Layer two takes those facts and does the expert work. The fragile point is layer one. It can return something that looks perfectly valid while containing nothing at all. The report in my hands was the cleanest possible example of that failure. Its eight chapters asked, in order: what is technically notable about the event; where does the player stand; what is the tournament format; how is the competitive landscape tiered; what rules and governance risks exist; what is the overall risk level; where is the public narrative pushing expectations; and how does the chess industry transmission chain propagate. Eight correct questions. Eight empty answers. In the player chapter, the rating table read: classical Elo insufficient information, rapid insufficient information, blitz insufficient information, recent form insufficient information. The head-to-head section: no opponent could be identified. In the landscape chapter, the tier diagram from the throne down to the reserve pipeline carried a single line: insufficient information at every tier. The format chapter stated plainly that the event could not be placed anywhere in the hierarchy — world championship, Candidates, qualifier, elite invitational, open or online. The rules chapter stated that no governing rule system could be identified. The industry-transmission chapter split the chain into three segments: upstream youth training, midstream events and platforms, downstream content and commerce. All three carried the same two words. Time sensitivity was never assessed either. For a news piece that is the biggest loss of all. An analysis that does not know which week it is describing cannot prepare anyone for any match. One detail deserves a pause. In the information-value assessment, the score was one star — far from the usual five-star scale. The stated reason: that single star is awarded purely for the existence of a structure. Semantically, the value is zero. That is where the report was honest. It did not invent. But hand that file to an automated system that insists on filling every field, and the outcome changes completely. A language model asked to "analyse" an empty input tends to compose chess content that sounds entirely plausible. It will assign a player an Elo rating, construct a game, cite an opening. Not one line of it is real. A report with a full skeleton and an empty core is harmless on its own. A report with a full skeleton and an empty core that has been filled in with guesswork is the dangerous thing. What I kept from that file was not the chess analysis but the operational warnings at the end. Three warnings, ranked by priority. Highest: silent failure propagating downstream. Layer one breaks, returns an empty shell, and layer two runs as if nothing happened. The recommendation was concrete: gate layer two behind a minimum threshold — at least three information points and at least one named entity — and fail loudly rather than emit a valid-looking empty shell. Second highest: fabrication risk. Any conclusion that cannot be traced back to a specific information point must be rejected. Third, medium: false-negative contamination. A dataset may later "show" that a topic was never mentioned — not because the topic did not exist, but because extraction broke. This is the hardest error to catch, because it manufactures evidence for a wrong conclusion. Reading that, I thought of the empty stadiums of the COVID season. In the middle of the pandemic, sitting in an empty ground, I could hear the tactical system breathing. That day I had no crowd to lean on, so I had to reconcile GPS data for eleven players in the match against Ho Chi Minh City FC. Young forward Nguyen Phi Hoang covered an average of 46 metres per minute inside the box but touched the ball only 21 times. Those two numbers contradicted each other, and the contradiction was the story. If the GPS units had returned a complete but empty file that day, I would have written a twelve-page piece about a match that never happened. Based on my experience tracking matches, the prettier the metric, the more suspicion it deserves. An unusually low ACPL for a young player can signal form, or it can signal missed moves at the data-entry stage. An overly smooth GPS trace can be a failed antenna, not a clever runner. One bad tape early in a career is the most expensive lesson there is: the eye always needs verification. Here is the paradox. In analysis we fear blank space. Blank space looks lazy. So we fill it. But a controlled blank is safer than a stuffed conclusion. A report with no data tells the truth. A report with wrong data lies to its reader. In chess, "no novelty found" is routinely read as "the opponent has no novelty". Those are worlds apart. The first sentence is about a database. The second is about an opponent. Football is the same: a team with no goals from corners across three rounds may be bad at corners, or your note-taker may have recorded the wrong column. The counter-intuitive point sits here. We usually blame the machine when a system returns nonsense. In this case the machine did its job: it read, it found nothing, it recorded that it found nothing. The fault belongs to people at the design stage — designing a process with no stopping condition. A system that is never allowed to say "I could not read this" is a system that will always say "analysis complete". Italian pressing is music that must be played together; one beat off and the whole symphony collapses. Data works the same way. Reading and analysis must share a rhythm. If reading slips a beat, everything behind it falls. The lesson from that eight-chapter file belongs to neither chess nor football alone. It belongs to anyone building a sports analysis pipeline in Vietnam, where data is becoming something people pay for. The fix is small. Set a minimum threshold. Keep the raw text alongside the extraction. Record the source, the publisher, the publication time and the result of the fetch itself. Check three times before publishing, as I have required of myself since round 12 that year. An empty pitch turned out to be the truest mirror of modern football. And an empty data file, read the right way, is a mirror of the same kind. Next matchday, before I open the metrics sheet, I will ask a different question first: does this file have anything inside it, or only a frame.