When the Tactical Map Is Empty: Lessons from a Broken Table Tennis Analysis Pipeline

In five years of following professional table tennis, I have learned something that seems obvious but is rarely stated: most analytical errors do not live in the conclusion, they live in the intake. A wrong tactical report usually begins with a misread fact, not a misjudged inference. This time, I have something close to an archetype on my hands: a two-stage analysis pipeline in which stage one — the extraction of information from the source article — returned a completely empty dataset. No title. No source. No information points. No entities. Only a template emitted with correct syntax and no content inside it.

What caught my attention was not the technical failure. What caught my attention was how stage two responded to it. Rather than filling the gaps with model memory, with familiar ranking figures or well-known names, the report chose to write "N/A — insufficient information" in every cell, across every analytical dimension, from technique to coaching staff, from event systems to risk warnings. For someone who makes a living reading matches through data, that is the most respectable thing an analytical system can do.

I want to tell this story my own way, because beneath its technical surface lies a very real tactical problem: how do you know whether you are looking at a real map, or at a blank sheet of paper in a beautiful frame.

Context: the two-stage architecture and where the break sits

The professional table tennis analysis model I and several colleagues use runs in two layers. The first — stage one — extracts raw text: it identifies the article title, source, article type, a list of discrete information points, the entities named (players, coaches, associations, events), time sensitivity, and source quality. The second — stage two — takes that dataset and deploys nine analytical dimensions: technique and equipment, player data and head-to-head records, event systems and points rules, the China-versus-world competitive landscape, rules and governance, coaching staff and talent pipeline, risk surface, public narrative and expectations, and finally industry transmission.

The architecture is sound in design. The problem is that it has a single, fatal break point: if stage one fails silently, stage two still runs, still emits a long document, with tables, section headers, and specialist terminology. Perfect form. Empty content.

In the case before me, every stage-one field returned an empty value or "not applicable". Article title: none. Source: none. Article type: unclassified. Information points list: empty. Core viewpoint summary: blank. Author stance: not applicable. Article purpose: not applicable. Entities involved: this field contained an instruction — "identify from the information points above" — while the information points list itself did not exist. Time sensitivity: not assessed in stage one. Source quality: this field also contained an instruction rather than data — "judge from the source fields" — while no source fields existed.

Read closely and a telling detail emerges: the data fields were empty, but the notes fields were full of operational instructions. That means the template was emitted successfully, but the step of fetching and parsing the article content never happened. This is not a case of an article genuinely lacking tactical content. This is a case of the pipeline stopping before it ever touched the text.

Core analysis: what happens when an analytical system meets an empty input

On the surface, the answer is simple: no data, no conclusions. But in professional reality, the pressure always runs the other way. A system asked to produce a report will find a way to produce a report. A language model asked to analyse will find a way to analyse. And the way it finds is usually to pull from training memory those numbers that sound plausible: a world ranking, a head-to-head percentage, a points milestone to defend, a name currently in the spotlight.

I call this the fame filter effect. When there is no specific data, the writer — human or machine — defaults to whatever is most common in collective memory. The result is an analysis that reads smoothly, sounds confident, and cannot be verified at all. In table tennis, this is the most dangerous kind of error, because the WTT ranking system operates on a rolling 52-week mechanism. A figure that was correct last month may have expired this week. A head-to-head record that held for the previous Olympic cycle may be meaningless after one season.

The stage-two report in this case took the opposite path, and that is why I am devoting an article to it. On the technical and tactical dimension, it states plainly: no playing-style label, no technical element (serve, receive, long rally, first three shots), no coach deployment, no single-match review was supplied. From that it draws three conclusions, and the third is the core one: this dimension cannot be scored on advancement, execution effectiveness, or physical fit, because any such score would be fabrication.

The same logic is applied consistently across all nine dimensions. On player data: no player entity is named, so no ranking deconstruction, no points-composition analysis, no points-defence pressure estimate is possible. On event systems: no event is named, so no tier positioning is possible across the ladder from the Olympic Games, the World Championships, the World Cup, through the WTT Grand Smash, Champions, Star Contender, and Contender tiers, to continental and domestic levels. On the competitive landscape: no association, no event line (men's singles, women's singles, doubles, mixed doubles, team), no result data, so no tier structure can be built.

What I want to emphasise is the discipline in how it handled familiar variables. In Chinese table tennis analysis, certain metrics are almost default: the win rate against foreign opponents, the title rate at the three majors over the last five editions, the number of seats in the world top ten, the depth of the U21 cohort. A poor report fills these cells with impressive numbers. This report wrote "not applicable — insufficient information" in all four cells, with a note that any landscape statement here would be background knowledge injected as if it were article analysis — precisely the failure mode this role is designed to detect.

The same applies to the rules and governance dimension. No governance level was named — ITTF, WTT, continental federation, or national association — and no rule type was referenced — competition, event system, selection, disciplinary, anti-doping. No officiating issue (service-fault enforcement, hidden-serve calls), no equipment-compliance issue (rubber thickness, surface treatment), no on-court discipline question. So no compliance-risk exposure can be estimated. And the third conclusion in this dimension is the sharpest: because no selection mechanism was described, the transparency ratio between published points criteria and discretionary room cannot be assessed — and that is the single most consequential governance variable in the Chinese team context.

The coaching staff and talent pipeline dimension is the same. No coach, team, or administrative entity was named, so no coaching résumé, tactical philosophy, or power structure can be assessed. Pipeline health requires at minimum an age structure or a junior-to-senior conversion data point; neither is present. And no signal of generational-transition stress — all-veteran concentration, a vacuum in the 23 to 26 band, the "forcing seedlings to grow" phenomenon — can be detected from an empty input.

The point I find most noteworthy sits in the risk surface dimension. The report built a six-row matrix for six business risk groups: competitive, selection and qualification, generational gap, governance and public opinion, systemic, opponent. All six returned "not applicable". But it added a seventh row, and that row was rated high: analytical-integrity risk — the risk that downstream conclusions are fabricated on a zero-information base. Likelihood: high if analysis proceeds. Impact: high, both reputational and in decision quality. Mitigation: block publication, require a populated stage one before re-invoking stage two.

That is very much how I read a match: when there is no data on an opponent's running lines, do not guess the direction of the acceleration. Record that the camera missed that zone, and set up a second camera.

Contrarian angle: honesty about empty data is the most valuable product of all

The first reaction most readers have when they see a nine-dimension report full of "not applicable" is disappointment. We are trained to value content. A long text, many numbers, many names, many tables — that is the sign of serious work. A text that says "I do not know" — that is the sign of incompetence.

I think the opposite, and I have professional reasons for thinking the opposite.

In table tennis, there is a concept I use constantly when analysing video: the space behind the full-back. The main camera always follows the ball. When the ball goes to the right wing, the audience sees the right wing. But the decision in a rally usually lives in the space the lens is not pointing at — the position of the second blocker, the hip rotation of the receiver, the distance between the feet in the pre-acceleration stance. I have written many articles just to talk about the zones the camera forgets.

A report that is honest about empty data is a report telling you where the camera is. It does not say "this is how the match went". It says "the lens has not opened". The difference between those two sentences is the entire value of the analytical profession.

But there is a deeper layer, and this is the truly counter-intuitive part. The biggest risk of an automated analysis system is not that it fails. The biggest risk is that it fails in a way that looks like success. An empty report in a well-built frame will pass editorial review more easily than a short note saying "no assessable content" — because it looks more like work. And if it passes, it becomes the basis for the next decision: a commentary built on it, a prediction built on that commentary, a public expectation built on that prediction.

I have seen this chain in table tennis many times. A ranking figure cited wrongly in one article, three months later it appears in a broadcast discussion, six months later it becomes the premise for a judgment about a player's form, and a year later it is the reason someone is declared finished. Nobody in that chain lied. Nobody checked the starting point.

There is one detail in the report I want to linger on, because it shows the writer understood the failure mechanism precisely. The report distinguishes cleanly between two situations. Situation one: the extraction pipeline failed, and needs to be re-run. Situation two: the source article genuinely contained only non-analytical content — a photo gallery, a fixture list, a fetch error page. In the second case, the correct output is not a nine-dimension report, but a short note stating that no competitive content is assessable.

That distinction matters because it acknowledges something the analytical profession rarely admits: some articles are not worth analysing. Not because they are poor, but because they do not contain the kind of information tactical analysis needs. A fixture announcement is not material for assessing playing style. A podium photograph is not data for building a spatial map.

And this is the execution blind spot I want to name: in the sports news environment, production pressure always exceeds verification pressure. Writers are measured by article count, not by article accuracy. A system is measured by reports emitted, not by reports that can be verified. That incentive structure produces exactly the kind of product this report refused to produce.

There is one more technical point worth making, and it connects directly to how I read table tennis. The report proposed a hard control gate: block stage two if stage one does not contain at minimum one named entity, at minimum three information points, and a non-empty source field. Those three conditions are not administrative procedure. They are the minimum conditions for any conclusion to be traceable to a fact.

In match analysis, I apply the same principle: no data, no judgment on form; no video, no judgment on tactical decisions; no head-to-head data, no judgment on stylistic matchup. Those three sentences sound simple, but I have had to rewrite more than a few articles for violating them.

There is one further layer, and this is the part I consider most important in the whole story. The report states clearly: if a downstream model fills the gap from training memory, then every ranking figure, every head-to-head percentage, every points milestone may be obsolete under the WTT rolling 52-week calendar. This is not a theoretical risk. This is the actual operating mechanism of the professional table tennis ranking system: points have an expiry date, and a player can drop in ranking not because they lost, but because old points expired while new points had not yet been banked.

If you read an analysis stating that a player sits at number N in the world rankings, the first question is not "is that correct", but "as of what date was that number computed". An article that cannot answer that question is not data analysis. It is memory presented in the form of data.

What to verify next time

I do not believe in perfect reports. I believe in reports that know what they are missing.

The case I have just described is an analysis pipeline broken at the intake stage, and the way it responded to that break is the most valuable part of the entire document. It did not salvage the situation by inventing a story. It stopped, named the problem, proposed a control gate, and stated clearly that every sporting conclusion in the document could not be issued because no information point existed to serve as a foundation.

For someone whose profession is reading the spaces the camera forgets, that is the correct posture. If the camera did not record it, say the camera did not record it. Do not draw arrows into the dark and call it a trajectory.

Next time you read a table tennis analysis packed with figures, try one thing: check whether the article states where that data came from and when it was computed. If it does not, you are reading a map drawn from memory, not from the match. And the only question worth asking at that point is whether the person drawing the map knew they were drawing into empty space.