When the Data Table Returns Zero: Anatomy of a Failed Table Tennis Analysis Pipeline

03:41 in Beijing, and a Spreadsheet With Nothing In It

I opened the file at 03:41 Beijing time. Nine sheets, each one a dimension of a deep table tennis analytics pipeline: technique, tactics and equipment; athlete data and head-to-head records; event system and points rules; competitive landscape; rules and governance; coaching staff and talent pipeline; risk surface; public narrative and expectations; and industry transmission. Every sheet had its full table structure, its full column headers, its full summary rows.

And every data cell carried the same sentence: N/A – insufficient information.

The only cell with real content was a single process warning: the first-stage pipeline returned an empty result, and that failure would propagate downstream if it was not stopped.

I sat still for about four minutes. Not out of shock. Out of recognition. I was looking at something the sports analytics trade rarely admits exists: a document that is perfect in form and empty in substance, generated legitimately, correctly, without violating a single procedural step.

What Happened Inside the Pipeline

To understand how a nine-dimension analysis can return zero, you have to picture how the pipeline runs.

The first stage deconstructs the source article. It extracts discrete information points, summarises the core viewpoint, identifies the entities mentioned, assesses time sensitivity and evaluates source quality. Its output is mandatory fuel for everything downstream. Without it, there is nothing to analyse.

The second stage takes that fuel and builds a deep analysis across nine dimensions. Every conclusion at stage two must trace back to a stage-one information point, with the evidence attached. That is the core discipline: every judgement has a root. In the document I was holding, the root did not exist. Title blank, source blank, article type unclassified, entity list empty, information point list empty.

The correct stage-two output in that situation is a format-complete, content-null document that says plainly there is nothing to analyse, and asks for stage one to be re-run.

That is exactly what happened. And because it happened so procedurally, it became an event worth writing about.

Why an Empty Result Is More Dangerous Than a Wrong One

A wrong result is easy to catch. You read an impossible number, you check it, you find the error, you fix it. An empty result makes no noise. It drifts through the system without triggering anything, and if a reader only skims the headline and the formatting, they will believe they just received a complete nine-dimension analysis.

An empty result is not a neutral result. It is a false statement wearing the disguise of a correct format.

In sports data, this failure mode has a name I use with colleagues: propagating false negatives. You track a player, you record no errors, and you conclude he played well. But another possibility has not been ruled out: you were not tracking the right thing. No red flags does not mean assessed and clear. Those two states look identical on a screen and are entirely different in nature.

In the case of this table tennis analysis, the specific danger is this: a downstream system reads the output, sees no risk flags raised, and records that this record has been assessed as safe. Then that record flows into a dashboard, into a bulletin, into a decision. The whole chain is running on a void.

I have seen exactly this mechanism in a pre-season data meeting. A squad-wide fitness tracker was presented with every cell green. Nobody asked why that week contained four measurement sessions instead of seven. The green table was so pleasant that nobody wanted to ruin it with a question about sample size.

Nine Dimensions and the Price of Fabricating Them Full

Suppose someone decided the empty table was unacceptable and started filling it in. What would happen to each dimension?

The first dimension is technique, tactics and equipment. No athlete is named, which means no playing style, execution efficiency, physical fit or equipment factor can be identified. Filling it in would require inventing a player, assigning a style, and fabricating a serve-attack point-win rate. That number would look highly convincing. It would have exactly one drawback: it would not exist.

The second dimension is athlete data and head-to-head records. Head-to-head tables are the most fabricated type of sports data and the easiest to verify. A line like "this opponent is a nemesis, four losses in the last five meetings" reads as highly professional. But without a name, an event and a date, that line is not data. It is a sentence.

The third dimension is the event system and points rules. This is where errors cause the most damage, because the professional table tennis points system has a strict structure: ranking points per event, points value determined by event tier and round reached, points-defence pressure across a cycle, and each event's position within the Olympic cycle. Fabricating a number here leads to a wrong conclusion about qualification, and that conclusion can affect a real federation's real decisions.

The fourth dimension is the competitive landscape and the comparison between table tennis nations. No association is named and no event line is identified: men's singles, women's singles, men's doubles, women's doubles, mixed doubles or team. Each line has an entirely different power structure. Merging them into one general picture is an analytical error, not a simplification.

The fifth dimension is rules and governance. Here the central question is always: how does a rule change create winners and losers, and how long until the consequences appear. Answering it requires a specific regulatory text, a specific date, a specific affected group. Without those three, every scenario is fiction.

The sixth dimension is coaching staff and the talent pipeline. This is the dimension most easily filled with prejudice. The age structure of the main tier, conversion efficiency from junior ranks to the national team, generational transition: all measurable, all requiring multi-year data. Without multi-year data, writers drift into talking about "tradition" and "identity". That is the moment analysis stops and mythology begins.

The seventh dimension is the risk surface. A risk matrix has six categories: competitive, selection and qualification, generational gap, governance and public opinion, systemic risk, and opponent risk. Each needs a level, a likelihood, an impact and a mitigation. The only identifiable risk in the document I was holding was a process risk: the stage-one pipeline returned an empty result. That is a small but real finding, and in my view it is worth more than an entire risk matrix filled with guesswork.

The eighth dimension is public narrative and expectations. There is no narrative to measure, which means there is no expectation gap to analyse. In my trade, the gap between market expectation and objective reality is one of the highest-value predictive indicators available. But it only has value when both sides are measured.

The ninth dimension is industry transmission. The transmission map runs from upstream equipment, youth development and training, through midstream events, associations and clubs, down to downstream broadcasting, commerce and derivative markets. No link is referenced, so the map is empty. An empty map, drawn neatly, is still an empty map.

The Discipline of Refusing to Answer

This is the part I want to spend the most words on, because it is the hardest part of the job.

Saying "I do not have enough evidence" is a professional act. It is not timidity. It is the result of a process that has run every step and stopped at the correct point. Beginners tend to believe their value lies in the number of answers they produce. Veterans understand it lies in the proportion of answers they are willing to withdraw.

Refusing to answer when data is missing is not a gap in the analysis. It is an analytical result.

In that nine-dimension table, every line reading N/A – insufficient information is a real result, produced by a real process, and auditable. It tells the reader exactly one thing: on this dimension, there is no basis for judgement. A document composed entirely of such lines, plus one process warning, conveys far more honest information than a nine-dimension document stuffed with invented material.

I learned this in 2026, at twenty, after a specific mistake. I built a knockout-stage prediction model for the World Cup based on expected goals, and I trusted a feeling about one team's defence more than the model. The quarter-final on 6 July 2026 in Nizhny Novgorod finished France 2-0 Uruguay, with an expected-goals gap of roughly 2.8 to 0.4. The team I picked produced four shots inside the box; the opponent produced nine. My prediction was mocked, and it deserved to be.

Three weeks later I rewatched all twelve knockout matches and logged every scoring sequence. The lesson was not that the model was right and the feeling was wrong. The lesson was that I had not spent enough time asking whether the data I held could actually answer the question I was asking.

Since then, every piece I write has two sections readers recognise: the model, and the limits of the model. The second is usually longer than the first. That is deliberate.

Table Tennis, the Most Mismeasured Sport

I work with football and basketball data, but I have followed table tennis for a long time, and I believe it has the widest gap of any sport between the volume of data generated and the volume of data used correctly.

A single table tennis match generates thousands of data points: every serve with its spin type, placement and speed; every receive with direction and length; every rally with its stroke count, tempo, and both players' positions at the moment it ends; every point with the score state at that moment. All of it is measurable with commercial cameras and ball-tracking software.

Yet most table tennis analysis I read uses three numbers: the score, points won, and unforced errors. Those three answer who won. They do not answer why.

In football, the analytical community moved further ahead thanks to expected goals: instead of counting goals, it measures chance quality. A team can win 1-0 with an expected-goals figure of 0.6 and lose 0-1 with a figure of 2.4. Viewers see only the result. Analysts see the gap between result and process, and that gap is where every good forecast is born.

Expected goals does not judge the shot; it illuminates what your football refuses to look at.

Table tennis needs an equivalent. I call mine expected point value: for each rally, a model estimates the probability of winning the point from position, spin, tempo and score state, then compares it to the actual outcome. A large positive gap means a player is generating more value than the scoreboard shows. A large negative gap means they are winning with something unsustainable.

I have not published that model, because I need at least two independent data sources to confirm it, and I need to test it on a large enough sample to know where it fails. Until then it lives in my notebook as raw notes, not as conclusions.

One Process Error Is Worth More Than a Fake Data Table

Back to the only warning with real content in that 03:41 document.

High-level risk, as classified in the document: the empty stage-one result propagates downstream. Recommended action: halt publication of any analysis derived from this data, flag the record as stage-one failed, and prevent it from entering any dashboard, bulletin or signal pipeline.

The second risk, also high: downstream systems misread "no flags" as "assessed and clear". Recommended action: add a schema-level guard requiring a non-empty information point list before stage two is invoked.

Third risk, medium: source ingestion error. Recommended action: inspect the ingestion logs for this record, retry extraction from the original source.

Fourth risk, medium: a template-completeness requirement can be satisfied while conveying zero information, creating a false sense that an analysis has been delivered.

Reading those four lines, I saw what I consider the most important finding in the whole document: the system diagnosed its own death. That is the mark of a pipeline designed with care.

Most sports data pipelines I have touched cannot do that. They fail silently. A feed dies, a CSS selector changes structure, a page blocks scraping, and the result is an empty table pushed to a dashboard nobody questions. The reader sees blank cells, fills them from memory, and carries on working.

Here, the blank cell announced that it was blank. It sounds small. In operational reality, it is the entire difference between a system that can be fixed and a system that manufactures permanent illusion.

What Happens If Stage One Runs Again Successfully

This question deserves an answer, because it turns an engineering incident into a reusable lesson.

When stage one has fuel, everything downstream unlocks. Technique and equipment become real analysis: a rubber change or a blade-structure change can be evaluated across an adaptation period, with clear dates and point-win rates inside that window. Athlete data becomes a head-to-head table with dates, event tiers, and a separation between phases and achievements. The points system becomes a points-defence pressure model across the cycle. The competitive landscape becomes a layered picture: the dominant tier, the chasing group, emerging forces, the rest. The risk surface becomes a matrix of likelihood and impact instead of a list of anxieties.

That is why I say re-running stage one is the only productive step in this entire story.

The Contrarian Angle: What Actually Broke Was Not the Pipeline

This part I want to separate from everything above, because it runs against the natural reflex of almost everyone who works with data.

The first reflex on seeing an empty result is to hunt for a technical fault: a dead feed, failed scraping, bad routing. Those hypotheses are reasonable. But stopping there means missing the more important diagnosis.

A pipeline can produce a format-complete document containing not a single information point. That means format and content were separated at the design stage. Structure was valued above truth. The template was valued above the material. And once those two are separated, there will always be pressure to fill the void with anything that looks like content.

Correlation is not causation. A document that looks complete is not a document with information. A table with nine rows is not a table with nine dimensions of analysis. An article with enough subheadings is not an article with enough argument. This is the error large data models commit most often, and people are no different, except that people still know how to be embarrassed.

In sport, that pressure is amplified by the news cycle. A match ends at eleven at night, and analysis is expected by seven in the morning. There is no window to re-run stage one, to cross-check two sources, to ask one simple question: do we actually have enough data to say this.

An empty stadium does not create ghosts; it creates the cleanest data a practitioner could dream of. But a newsroom empty of time creates something else: content written to fill space rather than to answer a question.

My contrarian view is this. The empty result at 03:41 was not the incident. It was the only moment in the entire operation when the system told the truth. It said: I do not know. And instead of treating that as a failure, sports analytics should treat it as the correct default state until there is enough evidence to say otherwise.

Four Gates I Recommend for Every Sports Analytics Pipeline

I took four things from this episode, and I apply them in my consulting work.

The first gate is a non-empty content check at the schema level. The information point list must contain at least one element before the analysis stage is invoked. It is a necessary condition, not a sufficient one, and it blocks most propagation failures.

The second gate is a clear distinction in every dashboard between two states: not yet assessed, and assessed and clear. Those states need different symbols, different colours, different downstream handling.

The third gate is record-level ingestion logging, so that when an empty result appears, it can be traced to a dead source, a mis-routed request, or a page-parsing failure.

The fourth gate is monitoring schema conformance over a rolling window. If the share of records with empty stage-one output rises above baseline, that is a systemic defect, not a one-off.

None of these gates requires a complex model. They require a design decision: to accept that an honest pipeline will regularly return empty results, and to design for that instead of hiding it.

Dents on the Chart, and What an Empty Table Leaves Behind

I do not write about football; I write about the dents players leave on a chart. In table tennis the dents sit elsewhere: the sag in tempo across a long rally, the bounce in point-win rate after falling behind, the spin of serve efficiency in a deciding game.

Those dents only appear when data flows through. Today the chart is empty. No dents. No athlete. No rally. No point.

And inside that emptiness there is one thing worth recording: a pipeline chose to tell the truth instead of choosing to look useful. That is a rare choice, and I think it deserves a line in the notebook.

The Blind Spot Nobody Wants to Measure

If I had to name the single biggest blind spot in how sport uses data today, I would point at this: we measure the quality of conclusions very carefully, and we almost never measure the quality of the ground those conclusions stand on.

An analysis of one athlete can be argued over for a week. But the more fundamental question is rarely asked: is the data table that analysis rests on still alive, when was it last updated, how many cells have been empty over the past seven days.

This is why I keep a private indicator I have never seen elsewhere: the ratio between signals emitted and signals traceable to an independent source. I call it the traceability rate. In a good week it sits above eighty percent. In a bad week it drops below fifty, and in those weeks I write nothing.

Silence is a valid output. That is what I want young people in this trade to hear earlier than I heard it.

The Next Cycle Will Be Decided by Input Quality, Not Model Quality

I believe that over the next few seasons, the competitive advantage in sports analytics will not come from better models. Models have become a commodity. Anyone can build a decision tree, a regression, a small neural network. What separates good practitioners from poor ones will be the ability to prove that their input data exists and can be trusted.

A record like the 03:41 one will become an asset rather than an accident, because it proves something most analytics tables cannot prove: the system knows its own limits.

When a pipeline can say "I do not know", the people who use it can believe it when it says "I know".

That is the entire bargain. And that is why I filed the 03:41 document into a folder called lessons, instead of deleting it as an error worth forgetting.