Nine Dimensions of Table Tennis Analysis and a Blank Data File in Shenzhen

One in the morning in Shenzhen. I opened the table tennis analysis file my colleagues had sent over and got nine blank pages: no source headline, no outlet name, no information points, no player names, no tournament names. Every cell of the nine-dimension framework we have spent two years building came back with a single line: “insufficient information, cannot assess”. I was not annoyed. I felt lighter. Our table tennis framework has nine dimensions. One: technique, tactics and equipment, which requires point-win rates by rally type, serve and receive structure, and any change in blade or rubber. Two: player data and head-to-head records. Three: the event system and its points mechanism. Four: the competitive landscape, in particular the balance between the Chinese national team and the rest of the world. Five: rules and governance. Six: coaching staff and the talent pipeline. Seven: the risk surface. Eight: public narrative and expectations. Nine: industry transmission, from equipment manufacturing through events to broadcast rights and player commercial value. It sounds impressive. But every dimension needs an anchor. Dimension two cannot run without a player’s name. Dimension three cannot run without a tournament’s name. Take dimension three. We need to know a player’s points-defence pressure under the WTT rolling 52-week ranking. That mechanism is not the tennis one: a tournament’s points expire exactly 52 weeks after they are credited, so last season’s run to a Grand Smash final becomes this season’s debt due. To say who is under pressure, I need to know which points are about to drop, in which week, and what the schedule of the direct rival looks like. Without those three data points, “under pressure” is inference dressed as reporting. Dimension five, rules and governance, also needs concrete dates. The ITTF moved from 38mm to 40mm balls in 2026, moved to the plastic ball in 2026, and the ban on glues containing volatile organic compounds took effect in 2026. Each change redistributes advantage in its own way: a bigger ball reduces spin speed, the plastic ball reduces it again, and the glue ban removed the speed-boosting layer that once sat under the rubber. Pure spin loopers lost value; players who take the ball early and vary pace gained it. Without knowing which era an analysis describes, I cannot tell whether it is describing adaptation or innovation. This is where I want to slow down. In recent years a school of table tennis analysis has tried to import football metrics straight onto the table. It sounds modern and it fails at one basic point. xG exists because football has territory and shot location. Table tennis has no territory. A loop from the left corner does not carry a higher “xG value” than a loop from the middle, because both are struck across a table 2.74 metres long. A defensible substitute for xG here is point structure: the share of points won on serve in a deciding game, average rally length, and the share of points won inside the first three exchanges. A good analyst of this sport has to know when to refuse a borrowed tool: a metric only means something when the mechanism that produces it exists. Head-to-head records, dimension two, are also routinely done badly. A 7-3 win ratio looks clear until you examine the sample: three matches at low-tier continental events, two in which one player had not recovered from a shoulder injury, and two from the era before the serve rule changed. A head-to-head record is only a map. We do not hunt treasure; we hunt the ability to read the map. And a map drawn from four matches is drawn from memory, not from survey. Now the hard part. In this trade the temptation greater than inventing numbers is filling gaps with content that sounds perfectly reasonable. Anyone can do it. You have a nine-dimension framework, you open a language tool, you ask it to write about a player’s ranking pressure, and it will write. Fluently. With numbers. With names. All of it wrong. The analysis I received tonight chose the opposite path. It kept the framework intact and wrote “cannot assess” in every cell. In the risk section it refused to score high or low, with a reason that is exactly right: scoring a null input does not produce a risk rating, it produces an arbitrary one. That is the most professional act in the entire document. Four information-value categories — competitive value, industry value, timeliness, reference value — all received one star out of five. One out of five. There is nothing to argue with. Then the most interesting part: that blank file still generated three risk warnings, two of them high. First, an upstream failure: the extraction stage produced nothing usable and the whole chain behind it was blocked. Second, the risk of fabricated analysis. Third, a medium-level design flaw: the entity-extraction field instructs the reader to identify players, associations and events from the information points above, while the information-points field is empty. An instruction that refers to itself. I met this exact situation in May 2026, when the Bundesliga restarted after the pandemic. My prediction model kept failing, and I found out why: home-win rate fell from 45 per cent to 38 per cent across 26 matches played without crowds. Across five years of data, the “crowd” variable had never been entered into the system, because nobody thought it could disappear. I held the report back for three weeks to refine it, and the desk had to use the old edition. In the end I published the revised version with a 0.82 adjustment for home advantage. When the stadium is empty, the data sits down and cries alone. But that time I learned that after the crying you add a new section to the report called “data limitations”. Since then every conclusion of mine carries a condition: under current conditions, at 85 per cent confidence. Data does not save a match, but it shows why the match died. Tonight there was no match to describe, no loop to measure, no player to price. There was an intact nine-dimension framework, a list of three signals to track, and one technical question still unanswered: if the upstream extraction stage failed because the source article could not be retrieved, is the fault in the collection system or in the source that vanished? Based on my experience following matches, blank files like this are rarely accidents. They are usually the trace of a source that was taken down, a page that was blocked, or a process that has never been tested under pressure. Numbers do not lie; they keep secrets. My job, as every night, is to sit with the spreadsheet and wait for upstream to return the first anchor. The framework is ready. When the anchor arrives, all nine dimensions can run in a single pass.