The Empty Bus of Table Tennis Data

2:47 a.m., Busan. I open a handoff file from an analytics workflow a young editorial team had outsourced: nine sections, full headings, full tables, a complete risk-assessment frame. I read top to bottom. The assessment column is empty. The evidence column is empty. The severity column is empty. Exactly one line is fully filled in, and it sits in the highest-risk cell: the danger of fabricating conclusions on a foundation of zero data.

I sat still for a while. Eighteen years of reading table tennis data tables had made me used to incomplete files. Incomplete is fine — mark it incomplete, leave it blank, come back later. This file was different. It was formally complete, like a bus arriving at the terminal on schedule, headlights on, plate legible, driver in uniform — except no passenger is aboard and nobody knows where it is going. The bus pulled up in front of the goal, and this time the goal was a table tennis table; I began questioning both the driver and the passengers.

Sports analytics now runs on three layers: extraction, interpretation, publication. The first layer reads the source article and pulls out information points — player names, events, scores, rubber changes, coaching decisions. The second places those points on a tactical frame and draws conclusions. The third writes for readers.

Table tennis is the hardest sport at layer one, because nearly all of its value lives in small details: who serves a side-spin ball at 9-9, who flips the rubber mid-game, who drops half a step after the first backhand loop. Without those details, layer two has only two honest options: state that the data is insufficient, or stay silent. Every third option is invention.

This gets likelier in rolling-point systems. The WTT ranking runs on a 52-week window; old points expire, new points replace them, and a ranking figure pulled from a model's memory may already be weeks out of date. The file tonight has no figure to be out of date, because it has no figure at all. But the mechanism survives intact: a frame handsome enough that nobody bothers checking what is inside.

I read it a third time and found three traces worth more than the entire content. First, the source field does not contain a source name but an instruction: judge from the source fields. A data cell holding a command is firm evidence that the template was emitted before the source fetch completed. Second, the article type is recorded as unclassified, meaning the classifier also ran on empty text. Third, and this is where I stopped longest: the hidden-information section does have content, with medium and low confidence levels explicitly marked. The system can reason about its own failure, but not about the match.

Those three traces taught me something eighteen years of reading number tables had not: formal completeness deceives readers more effectively than any wrong number. A table with nine sections, bold headings and a ranked risk frame is automatically granted credibility that its contents do not deserve. A reader skimming a tightly structured document assumes the writer checked. Nobody checks the frame.

Based on my experience following matches, the worst analyses I have ever read were never the ones short on data. They were the ones with plenty of data that had nothing to do with the question. A piece citing 61 percent possession without naming the pass that produced the goal is exactly as empty as tonight's handoff file, differing only in being decorated with numbers that look real. In table tennis, the local version of that disease is the analysis reporting that a player won 4-1 without saying whether the fourth game turned on an opponent's rubber change or on a lost stretch of service points.

I have written that way myself. In 2026 I built a 3,200-word analysis around a draw at Old Trafford and drew furious pushback from a supporters' group over one remark about a player. I did not retract. I ran a two-hour livestream, redrew nine situations and challenged readers to argue back. The lesson I kept from that night was not about being tough; it was that when challenged, the only thing that saves you is concrete evidence. Tonight, in that handoff file, there is no evidence for anyone to challenge. That is why it is dangerous. Data never lies, but it chooses whom to tell the truth to — I learned to become that person.

Most of the debate about artificial intelligence in sport circles around machines inventing facts. I think that fear is aimed at the wrong target.

An inventing machine still leaves readers a chance of catching it, because invention tends to clash with what the eye has already seen. The more dangerous type is the machine that answers insufficient data in a rigorous, procedurally correct way, with confidence notes and a recommendation to block publication — and is then read by the newsroom itself as a sign of professionalism and shown off. Acknowledging limits has become a costume. Worn, it looks trustworthy; underneath there is still nothing.

I have worn that costume too. In 2026 I built a series on football in empty stadiums, using data from fourteen Bundesliga matches without crowds to test a hypothesis about crowd pressure. A K League data analyst caught me out: small sample, overreach. I accepted it and added a limits-of-analysis section to every piece. But if that section only buys me permission to keep talking without changing anything, it is a charm, not a confession. The pandemic stripped away the crowd but kept the match — and the obsession of the man holding the board.

So instead of asking how intelligent my system is, I will ask whether it is willing to stop. The minimum gate I now apply to any table tennis analysis: at least one concrete proper noun, at least three citable information points, and a source field that actually contains a source name. Miss one of the three and the piece does not leave the desk.

Tonight the bus still reaches the terminal, on time, lights on. I only want readers to start asking the driver one simple question: is anybody on board.