Table Tennis: One Empty Payload, Two Schema Defects, and the Fabricated-Data Trap

Eleven Fields, One Survivor

The first-stage extraction payload for a table tennis article contained eleven fields. After auditing each one, exactly one survived intact: the domain label, recorded as table_tennis.

Article title: absent. Article source: absent. Article type: unclassified. Information points: an empty list. One-sentence summary of core viewpoints: blank. Author stance: absent. Article purpose: absent. Entities involved: "identify from the information points above" — while the information point list above is, in fact, empty. Time sensitivity: not assessed. Source quality: "judge from the source fields," and both source fields are absent.

Usable fields: 1 of 11. Analytical dimensions that can be developed without inventing content: 0 of 9.

I sat with that audit sheet for a while. My first reflex was not to ask who the article was about. It was a professional alarm: if I let my hand slip, I could write a perfectly fluent piece about table tennis, and it would be wrong from the first line to the last.

Operating Context: A Two-Stage Pipeline, and Where It Died

I work as a sports data analyst in Chengdu. Every deep-dive piece I write runs through two stages. Stage one decomposes the source text into discrete data: title, source, publication time, information points, named entities. Stage two takes those fragments and builds nine analytical dimensions — technique and equipment; player data and head-to-head records; event systems and points rules; competitive landscape; rules and governance; coaching staff and talent pipeline; risk surface; public narrative; industry transmission.

The rule governing stage two is one line long: say only what stage one put on the table. There is no exception for "I roughly know."

I learned that discipline early. In 2026 I entered the industry in a fact-checking role, and my only job then was to find which number in a draft had no traceable origin. Twenty years later I am doing exactly the same work, except now I also have to audit numbers that a machine produced.

This time, stage one returned an empty payload. Four common causes explain extraction failure at the ingestion stage rather than the filtering stage: the source is an image or video, the source sits behind a paywall, the text is truncated, or the source language is not on the supported list. A filtering-stage failure usually leaves debris — a name, a scoreline, a timestamp. Here there was no debris at all. That signal matters more than its surface appearance.

One more detail is worth noting: a pure results report, even under weak extraction, usually yields at least a few information points about the scoreline. A completely empty payload leans toward the hypothesis that the original was a commentary or analysis piece, where the value lives in the argument rather than in discrete facts.

Nine Analytical Dimensions, and Why Each Was Blocked

Technique, tactics, and equipment. No athlete named, no playing system cited, no point-win-rate or service-win-rate data supplied. To open this dimension, the minimum inputs would be: a named person, the playing-style system claimed to be in use, and whether the original described a technical overhaul, an equipment change, or a tactical review of a single match. Those three scenarios lead to three entirely different conclusions, so they cannot be pooled.

I keep the equipment section locked especially tight. In table tennis, claims about rubber, sponge hardness, blade ply count, or booster history are the most fact-checkable category of information and also the most frequently misreported by self-media channels. Without the source text, I have nothing to cross-check against. This is the dimension with the highest fabrication risk, and the correct handling is to close it, not to estimate.

Player data and head-to-head records. Player name, current ranking, points composition, and expiry calendar — nothing. A named head-to-head with a date and event tier — nothing. This dimension is the load-bearing wall of the whole framework: dimensions four, six, seven, and eight all inherit their conclusions from player data. Once dimension two is empty, six of nine dimensions are structurally blocked, not merely missing a few cells.

The analyst's signature check — measuring ranking against true strength — requires at minimum two inputs: a ranking figure and a record against opponents from other associations. Neither exists. And even with a ranking figure, whether it reflects true strength would still need another season to answer.

This is where I want to pause. The entities-involved field instructs the analyst to identify entities "from the information points above." The information point list is empty. Which means the entity field depends on something that does not exist. That is a self-referential loop, a schema defect, and it will recur on every future article until it is fixed at stage one.

Event systems and points rules. No event can be identified, so there is no tier to position against: the three majors, the top tier of the WTT system, the Challenger level, continental, or domestic. This is the dimension most sensitive to the missing time field. WTT operates a rolling points-deduction mechanism on a 52-week cycle. The same nominal points total means very different things depending on the month of the year. Based on my experience tracking matches and points-lock periods, a player can look solid one month and fragile the next without losing an additional match — simply because old points fell out of the window. Without a publication timestamp, every points inference floats.

At events with draws, one further layer is blocked: bracket difficulty, the possibility of meeting a nemesis opponent, and whether the organisers enforced separation of players from the same association. Those three questions can only be answered once the event is known.

Competitive landscape. This cannot be a "general background" section written apart from the source article. It is a positioning exercise, and it only has meaning when tied to a specific event line: men's singles, women's singles, men's doubles, women's doubles, mixed doubles, or team. A generic "China versus the rest" overview written from general knowledge would be padding, and this framework forbids exactly that kind of padding. One technical note: the men's and women's draws cannot be merged into a single sentence. The men's field is systematically more open than the women's, and any statement without an event-line label risks a category error.

Rules and governance. This is the dimension most vulnerable to irresponsible speculation, because rules and selection topics always invite the writer to fill gaps with narrative. Handling a governance topic properly requires discriminating source tiers: official announcement, reputable media, or self-media rumour. Stage one recorded the article source as absent and source quality as unassessed. An undetermined source tier by itself eliminates any possibility of repeating a governance claim here.

Coaching staff and talent pipeline. Also a dimension that sits behind names. No coach named, no age structure for the main tier, no conversion efficiency for the next generation. Generational-transition risk is normally a standing watch item in this field, but assessing it requires an age structure, and an age structure requires names. This dimension must not be filled with general knowledge about the Chinese, Japanese, or European systems.

Risk surface. This is the only dimension that ran, and its output deserves careful reading. The risk matrix has three items at high level, and all three are epistemic rather than sporting: fabrication risk when filling empty dimensions; stale-data risk from the missing timestamp; and source-tier risk from an unverifiable origin. Every remaining risk item — injury, schedule overload, points-defence pressure, generational vacuum, selection tension, Olympic programme adjustment, calendar inflation — is marked insufficient information. With no athlete named, there is nothing to screen.

Public narrative. Without a narrative, narrative analysis is impossible. Measuring the gap between expectation and reality requires two legs: a market expectation and an objective assessment. Neither leg exists here. The only defensible output is not a finding but a handling rule: with the source tier undetermined, all sensitive content must be left alone — not amplified, not adjudicated.

Industry transmission. This dimension is purely causal in construction: it needs an initiating event — a title, a rule change, a transfer, a commercial announcement — before the chain can be traced upstream to downstream, through the equipment market, youth development base, event commercial ecosystem, player commercial value, policy and capital flows, and finally the international ecosystem. Without an initiating event, the chain does not exist. This is also the dimension I most suspect the original article was actually about, based on the failure signature: no player entities, no event entities, no match data. A piece about the industry or about governance would naturally yield very few discrete information points, and could collapse entirely under an extractor that only recognises facts.

The Counterintuitive Angle: An Empty Payload Is a Signal, Not a Loss

The easiest thing to overlook here is diagnostic value. A full payload tells me the content of an article. An empty payload tells me the condition of an entire pipeline — and that information has a far longer shelf life.

The audit sheet exposed two schema defects that will recur on every future article unless fixed. First, the entity field depends on the information point list, which can be empty. Second, source-quality assessment is delegated to the source fields, while the template does not require those fields to carry values. Both are design faults, not analyst faults, and both have cheap fixes: decouple entity extraction from the information point list, and make the source field mandatory and non-null.

Then there is the timestamp. Missing content is a recoverable deficit: re-extract the source text and you are done. A missing publication date is different in kind. A publication date cannot be reconstructed from content with certainty, and under a rolling 52-week points system, an undated table tennis article is analytically ambiguous even when its content has been fully extracted. In this domain, a date is a functional input, not decorative metadata.

I stand on the side of the number, even when the number stands alone. Here the number genuinely stood alone — and it still had something worth saying.

Where This Stops

Data does not lie; we simply have not learned how to ask. This time the right question was not which player appeared in the original article, but where my pipeline broke, and how soon it will break again.

Three tasks sit at the top of the list for the next cycle. Re-run the first-stage extraction on the original source with a corrected template. Make the source field and the timestamp mandatory and non-null, and let the pipeline halt rather than emit empty values. And log the ingestion failure reason, so that image and video sources are routed to character recognition or transcription instead of returning a blank sheet.

For me personally, the next cycle's test is very concrete: if the new extraction returns three or more populated fields, the six blocked dimensions open on their own. If it still returns one field, then the problem lies with the source, not with the article — and that is a different conclusion, and also worth writing.