Nine Sections, One Answer: Where Table Tennis Analysis Actually Breaks

Shenzhen, Tuesday morning. I opened a 4,200-word report a analytics team had sent over and set it beside a cup of tea that had long gone cold. The file had all nine sections: technique and equipment; player data and head-to-head records; event system and points rules; competitive landscape; rules and governance; coaching staff and the talent pipeline; the risk surface; the public narrative; and the industry transmission chain. Nine major headings, nine tables, nine sets of conclusions.

I stopped at the third line. The first cell read: “Insufficient information.” I turned to section two. Same sentence. Section three, four, five, all the way to nine — the same sentence, set in the same font size, aligned to the same margin.

Four thousand two hundred words. Not one player's name. Not one match. Not one ranking figure. Not one win rate. A perfect seal on an empty room.

I closed the file. What was worth writing sat inside that hollow frame.

A modern tactical report runs through two layers. The extraction layer reads the source article and pulls out information points — discrete, citable units of fact: a score, a name, a timestamp, a rubber change. The analysis layer takes that list and only then builds its evaluation dimensions. An empty list leaves the analysis layer with nothing to hold on to.

In the file I was holding, the extraction layer returned an empty list. Source title: none. Source: none. Article type: unclassified. Time sensitivity: not assessed. The notes cell still contained its technical instructions — the kind that read “derive entities from the information points above” — while the information points above were blank. The template had been emitted before the source page finished loading.

The fault sits in the operational layer, not in the tactics. But it opens a window worth looking through: table tennis analytics is being built faster than its own verification speed, precisely when the 52-week WTT rolling calendar means ranking points expire every week that passes.

I entered this trade at the fact-checking desk of Sports Illustrated in 2026. The job then was to phone and verify every number before a story went to page. Thirty years later, software has replaced the phone call; the question is unchanged: where did this number come from.

To grade a player, the analysis layer needs four input groups: a playing-style system and its technical elements (serve, receive, rally, the first three shots); a physical profile and body characteristics; match data such as point-win rate and serve-and-attack efficiency; and footage to check each rally against. All four groups were absent. Without them, any score for advancement, execution effectiveness or physical fit is just numbers lined up to look tidy. An honest report must be able to say “I don't know” at exactly the place where it doesn't know — and that is the hardest skill in this trade, because the industry's rewards fall to whoever speaks most smoothly.

The player data section was empty in the same way. No entity was named, so ranking could not be unpacked, points composition could not be broken down, and points-defence pressure could not be estimated — that pressure peculiar to the 52-week deduction mechanism. The single most discriminating metric in the China-versus-the-rest frame is the foreign-match win rate, held separate from the domestic win rate. Without a match, an opponent or a result, that metric is zero in the mathematical sense, not the professional one. The age curve could not be drawn either: under 22 is rising, 22 to 28 is peak, past 28 is veteran, and for women usually one to two years earlier. Placing a player on that curve requires a date of birth and a career stage. Neither existed.

The event system section was no different. Professional table tennis is tiered: the Olympic Games, the World Championships, the World Cup, then the WTT Grand Smash, Champions, Star Contender and Contender bands, the continental events and the domestic ladder. Each tier pays different points and creates different incentives for each player. Without an event name, that ladder does not exist. Without a draw, there is no half of death to find and no same-association separation to check against the regulations. These are not side details; they decide whether a player reaches the deep rounds at all.

The risk table is where I lingered longest. Six standing risk groups — competitive, selection, generational gap, governance and public opinion, systemic, opponent — all returned “not assessable.” A single row was graded High, and it does not belong to the arena: analytical-integrity risk, meaning the chance that a report looks substantial while containing no claim traceable to a source. In a table that should have been about opponents and points, the only gradable line was about the writer himself.

The real danger sits in the next step. When a data cell is empty, the reflex of both the writer and the machine is to fill it from memory. A language model trained on millions of table tennis articles will “remember” rankings, remember head-to-head records, remember form — and fill the gap with something that looks exactly like real data. Under a 52-week deduction cycle, a memory from three months ago may already be wrong. Quantitative figures must never be drawn from model memory; if it cannot be verified, mark it plainly as unverified.

The source-quality cell had never been graded either. The industry's source ladder is fairly clear: official announcement from the person involved, reputable press, average source, low-quality source. Without a source-tiering pass, no sensitive item may go to page: an undisclosed injury, a dispute over a selection slot, a rumour of a fixed match. This is also where I hold my old position: match data sold straight to betting companies is the darkest side effect of the digitisation of sport. The same dataset, used on one side to understand a match and on the other to price risk. The line between the two is thinner than it appears.

The counterintuitive point: this empty report file is the most honest document of the month. It deceives no one. What is worrying is the hundreds of other reports that look substantial, with tables, with percentages, with bolded conclusions, where not one line traces back to a real information point. Every tactical diagram is an organised lie told in the face of the chaos of a match — I wrote that years ago and still use it. But an organised lie needs material to build from. Without material it becomes a smooth lie, unorganised, and considerably worse.

There is another possibility few consider: the source article may never have contained analysable competitive content at all — a photo gallery, a fixture list, a failed page load. In that case the correct output is a short paragraph stating plainly that there is no competitive content worth assessing, not a nine-section report. The machine's failure here was not silence; it was noise far beyond what the situation required.

Shenzhen taught me that haste in a reform effort only produces a well-irrigated graveyard. A minimum threshold — at least one named entity, three information points, one non-empty source cell — sounds as dull as meeting minutes. Yet it is the most important governance variable in the entire chain.

Next time you read a post-match assessment, try counting how many of its claims trace back to a timestamp, a name, a sourced number. If that ratio falls below one third, what you are reading is a decorative table. When people replace the grass, they forget to replace what feeds the roots. And the root of every analytical table is, in the end, a very old question: where did this number come from.