The Empty Table Tennis Data File and the Discipline of Not Inventing Numbers

Three in the morning in Shenzhen. I open the extraction file for a table tennis article and count exactly one field carrying data: the domain label. The other eleven — article title, source, article type, list of information points, core viewpoints, entities involved, time sensitivity, source quality — all return empty or "undetermined".

A system runs for four hours to return a single tag.

People outside the trade call that a technical failure. I call it one of the most honest outputs my system has ever produced. Ten years between the table and the keyboard have taught me that bad data presented beautifully is more dangerous than missing data, because readers at least know to guard against a gap. An empty file says one true thing: this time, we know nothing.

So I chose to write about the gap itself, because it exposes two schema defects that will repeat on every subsequent article if nobody fixes them.

Our pipeline runs in two stages. Stage one deconstructs the source text into discrete data fields: who played, which event, which round, what score, which quotes were pulled. Stage two takes those fields and expands into nine analytical dimensions — technique, tactics and equipment; player data and head-to-head history; event system and points rules; the landscape between the Chinese national team and the rest; rules and governance; coaching staff and the talent pipeline; the risk surface; public narrative and expectation; and finally the industry transmission chain.

Nine dimensions sounds impressive. All nine grow from one root: stage one has to extract something first. This time stage one extracted nothing, and the whole structure above hung in the air.

The first defect sits in the "entities involved" field. Its instruction reads: identify from the list of information points above. But the list of information points is empty. The system asks me to find players by reading a list containing no players. That is a circular dependency at the design level, and the consequence is not small: with the entity field empty, six of the nine dimensions are blocked at the first step, blocked structurally rather than substantively.

The second defect sits in the "source quality" field, which asks for an assessment based on the source fields of the information points. No information points means no source fields to read. An instruction that cannot be executed, and it leaves every downstream claim unweighted for credibility.

Then comes the place I want to linger longest: the timestamp.

A publication date in table tennis belongs to the arithmetic, not to the decoration. The WTT ranking system — the commercial events arm of the ITTF — deducts points on a rolling 52-week cycle. The same nominal points total means entirely different things in January and in November. A player holding 2,500 points with 1,800 of them expiring in six weeks is no longer the world number five in the sense the ranking intends.

A ranking figure severed from its date is nothing but noise wearing the shape of data.

In this report, the "time sensitivity" field reads exactly two words: not assessed. My signature check — setting ranking beside true strength, measured by foreign-match record and major-event consistency — requires at minimum a live ranking figure and a head-to-head table with dates. Neither exists. The check cannot run, and I will not replace it with a feeling.

The report rates three risks as high. Fabrication risk: the analyst fills empty dimensions with plausible-sounding table tennis content that has no source. Stale-data risk: with no timestamp, any recovered figure cannot be anchored to the 52-week deduction cycle. Source-tier risk: both source and source quality are blank, making every claim unverifiable and unweightable. All three are epistemic risks, sitting on the writer's side, not the sport's.

The landscape dimension carries its own trap. Men's and women's draws cannot be pooled. The men's field is materially more open than the women's, and any sentence of the form "world table tennis is closing the gap" that fails to name which draw it means is a category error. The report refuses to issue a threat assessment on Japan, France, Sweden or Germany, because doing so would substitute general knowledge for article-specific analysis — precisely the trap a data person must avoid most.

The equipment dimension is the easiest to backfill and the most dangerous. Rubber sheets, sponge hardness, blade plies, booster history — those are the most fact-checkable claims and also the most frequently misreported by self-media outlets. Without the original text, this dimension stays closed, not estimated.

I could sit down and reconstruct that article in forty minutes. Pick a name, assign him a playing system, describe a six-month technical overhaul with its accompanying trough, add a few point-win rates, build a head-to-head history of three real matches and two not-quite-real ones. Readers would not tell the difference. I would, and that is the entire difference between analysis and merchandise.

This report chose the opposite path: lock every content field, open only the diagnostic section. The information-value table gives itself one star across four categories — competitive value, industry value, timeliness value, reference value. A table that scores itself one star is evidence the system still knows how to count, not a confession of failure.

One line in the report I read over and over: the dominant risk sits on the analyst's side, not the sport's. In table tennis commentary, what is permanently missing is source discipline; fabricated data is always in ample supply. Style descriptions that sound entirely reasonable, rankings that sound entirely reasonable, selection stories that sound entirely reasonable. None of them sourced, and none of them catchable, because there is nothing to cross-check against.

The whole industry is being trained to fear the gap. Missing a number, go find a number. Missing a source, cite "according to multiple reports". Missing a timestamp, write "recently". Those three habits compound into an ecosystem where errors never surface.

But there is a paradox worth noting: a missing timestamp is more dangerous than missing content. Content can be re-extracted from the original source at any time. A publication date cannot be reconstructed from the text with absolute certainty. An article that loses its date loses one analytical dimension forever; an article that loses its words only needs the pipeline run again.

That is why I do not treat this as a data failure. It is an ingestion failure, and it left a very characteristic fingerprint: exactly one field survived. When an extractor returns the domain label pulled from a URL or from metadata, the odds are high the source was an image, a video, or paywalled content. The original article may be perfectly ordinary. The machine is the blind one.

There is one more possibility I am obliged to raise. If the source piece was about the event system, about governance, about the economics of the sport, then producing very few discrete information points is natural. An institutional analysis does not hand you a scoreline; it hands you an argument. An event-oriented extractor will strangle that kind of text, and in that case the three productive dimensions are the event system, governance rules and the industry transmission chain — while the two technical dimensions should be closed on the record rather than padded.

Three fixes, in priority order.

Make the source fields — URL, publisher, author — mandatory and non-null. If they are absent, fail the pipeline outright; do not emit "undetermined". A system willing to stop is more trustworthy than a system that always has an answer.

Decouple entity extraction from the information-point list, so an empty list does not drag the entire entity field to zero.

Capture the timestamp with its time zone at ingestion. In this sport, the date is an arithmetic input; decorative labels save no number.

The signal to watch in the next cycle is clear: the field-population rate on the re-run. Three or more populated fields unlocks the technical, head-to-head and landscape dimensions immediately. If the source turns out to be institutional in nature, the analytical spine has to shift axes entirely.

I once wrote a piece about a striker whose expected-goals figure ran nearly double his actual goals, and an entire generation of veteran writers called it a mathematical farce. A year later the number held, and the lesson I took was about discipline, not about maths: say only what you have counted, and say plainly where and when you counted it.

This time I counted exactly one thing. And I am writing anyway, because an honest piece about not knowing is still worth more than a confident piece about knowing everything.

Data does not answer your question. It teaches you to ask the right one.

Trust is the only commodity this market misprices — until the data corrects it.

And numbers are the match's love letter — learn to listen and you will see everything. Even when the love letter, this time, is only silence.