Table Tennis: The Open Audit of a Sourceless Data Table

At 2:40 in the morning I opened the analysis file that came back from the system. It had nine sections. Every section had a table. Every table had an assessment column, a benchmark column, a notes column. The framework was built down to the last cell: technique and equipment, player data, event system and points, competitive landscape, governance and rules, coaching staff, risk surface, public narrative, industry transmission.

Then I read the interior. Every cell carried the same line: N/A — insufficient information. No player name. No event name. No score. Not one ranking figure, not one win rate, not one head-to-head record.

The only concrete text sat in the closing note: identify entities from the information points above. The information point list above was empty.

I sat still for about two minutes. Not out of surprise. Because I recognised that I had written documents of exactly this shape before — the only difference being that the cells inside them had numbers.

Table tennis is a sport where data ages fast. The WTT ranking system runs on a rolling 52-week window: a result expires exactly one year later, and a player's ranking is a moving object rather than a fixed label. An analysis written in June can be wrong by July without anyone beating anyone — all it takes is for old points to drop out of the window.

The event structure is clearly tiered: from Grand Smash down through Champions, Star Contender, Contender, Feeder, each level carries a different points value and a different prize purse. The three majors — the Olympic Games, the World Championships, the World Cup — sit at the highest weight and are also where a player's legacy file gets written.

Reading those layers is my job. I have done it for seven years, I came up through football, and I carried one habit with me: before trusting a metric, trace where it was born, in which match, under which conditions. That habit makes me uncomfortable around numbers with no root.

The file was a document with no root. Not a weak root. No root.

The process I run has two stages. Stage one breaks the source text into discrete information points: who, which event, which result, which date. Stage two places those points into nine analytical dimensions and issues a judgment. Stage one returned an empty list. Stage two ran anyway, because the template was already built. The output was a nine-part report that, skimmed, looks very much like a real report.

This deserves a long pause, because it goes beyond the story of one broken file.

What does a decent table tennis analysis need in its first dimension? It needs to know which side a player serves from, the point-win probability across the first three shots, the distribution of rally lengths, and the win rate when pushed into a long-distance counter-looping exchange. It needs to know whether that player has just changed rubber, because the adaptation window after a rubber change typically runs three to six months, and inside that window every metric is noise. It needs to know whether the total thickness of the racket sits inside the 4mm limit the ITTF equipment rules allow — a small detail that has decided entire matches at professional level.

Without a player name, all four requirements hang.

The second dimension is player data. A ranking is not a flat number; it is an aggregate of results still in force, each with its own expiry. A player defending points from a major that is about to expire carries completely different pressure from one who has just won. Win rate against opponents from other associations, consistency at the majors, performance in deciding games — those three separate a steady player from one who is only good in the group stage.

The third dimension is the event system. How does a Grand Smash entry differ from a Contender entry, how many points separate them, where does the event sit in the Olympic cycle, does the draw push two difficult opponents into the same half. All of it needs an event name as an anchor.

Then the China-versus-the-rest landscape: the share of top-10 world places, the number of titles across the last five editions of the three majors, the depth of the under-21 cohort. Those three measures cannot be derived from the framework; they have to be loaded in from data.

Seven years sitting in front of match footage, I learned that the most valuable thing in a match is not the final score but the sequence of decisions before it: the serve at 9-9, the rubber switch mid-game, the footwork slowing after a long rally. Those only surface when there is footage and someone logging every rally. No footage, no log — only feeling remains.

What I want to state plainly: an analysis with zero information points is not a weak analysis. It is a document impersonating an analysis. The complete form makes a skimming reader believe someone did the work. Inside there is no claim that can be verified, and therefore no claim that can be refuted. A document that cannot be refuted is a document that cannot be audited — and to me that is the definition of professional worthlessness.

I once did the opposite and was still wrong. In 2026 I built a homemade xG model for a V.League match and predicted a 65% home win on the back of superior possession. The result was 0-3. It took a month of footage review to see the model was missing two variables: chance quality and the speed of central attacking transitions. The lesson then was that missing variables produce wrong conclusions. Today's lesson is heavier: missing all the data produces no conclusion to be wrong about, only a shell.

The counterintuitive part sits here. People usually assume the biggest risk in sports analysis is making a wrong prediction. I no longer think so. The bigger risk is issuing a document with the right form, the right sections, the right tables, the right jargon, and not one line that touches the reality of a match.

In 2026 I wrote that France could not beat Croatia, based on expected goals. The piece was read more than 200,000 times. France won 4-2. The error was not in the number; it was that I did not adjust for opponent strength in the knockout rounds. The empty-stadium season of 2026 taught me the same thing again: data without context is only half the truth.

But in both of those cases I still had data to be wrong with. This file had none. And if I simply kept writing, filling the empty cells from professional memory — half-remembering a player who once won some event, half-remembering a ranking — I would produce a fluent piece with numbers and names that could have been obsolete from its first line, because those WTT points had long since dropped out of the 52-week window.

A reader once told me my article said nothing about the defence. I brushed it off then, because I believed I had the data and they did not. A few years later, rereading that piece, I saw they were right. The data was not wrong, the reader was — and I had been that reader.

A 30% probability is not an excuse — it is the reminder that I am right only 7 times out of 10. But that 30% only means something when I have actually staked evidence. A report with no evidence has no 30%; it has a 0% chance of being audited.

I set one rule for myself, applied to every table tennis analysis that passes through my hands: at least one named entity, at least three citable information points, and a source field that is not empty. Fail any one of the three and the document is blocked — not because it is bad, but because it does not yet exist.

Every model of mine is built on mistakes that were once laughed at — the most genuine foundation I have. And that foundation begins with a very simple question: where was this number born.

If an empty data table can fit a nine-part report, how many analyses out there are wearing that same coat with nobody opening them to check?