The Empty Audit: How Thin Is Vietnamese Table Tennis's Point-Level Data?

There was a Tuesday morning in Binh Duong when I opened the deep-analysis file I had waited two days for. Eleven pages. Nine sections. Forty-seven check items. Every one of them in the same state: insufficient information, cannot assess.

No player names. No head-to-head table. No point-win rate on serve. No tactical breakdown, no industry transmission map. The summary section held a single line: no assessment is possible.

I sat in front of the screen for a while. Seven years in this trade had taught me to live with missing data, matches with three games missing, hand logs that drift fifteen seconds apart. This time was different. The result was entirely empty, and what caught my attention was how it was reported: the pipeline refused to invent. It did not fill the gaps with adjectives.

Context

My process runs on two tiers. Tier one deconstructs the source text: extracts information points, identifies entities, dates, provenance. Tier two performs deep analysis: technique, tactics, head-to-head, event system, risk, industry transmission.

Tier one returned an empty file — no source title, no source, not a single information point. Tier two therefore had exactly one job left: to declare it had nothing to analyse.

To outsiders, that is a failure. To me, it is correct behaviour.

My work in Vietnam runs into a paradox. Table tennis is played a great deal and watched a great deal, but recorded very little. A national championship contains hundreds of matches; the number logged point by point — enough to compute serve-point win rate — can be counted on one hand across a whole season. National team matches at the SEA Games are better, but records usually stop at game scores, with occasional live point statistics on the electronic board. When the board goes dark, the data disappears.

When I was tasked with building a model for a domestic sports data platform, I spent six weeks answering one question: what do we actually have? The answer was blunt. We have scores. We have fixtures. We have names and years of birth. Point sequences, shot direction, placement, rally duration — almost nothing in computable form.

That empty audit is a miniature of the entire system.

Analysis

A table tennis analysis only earns its keep if it answers five questions, and all five need point-level data.

Point-win rate on serve and on receive, split by game and by opponent. Rally-length distribution. Receive-error rate, split by spin type. Efficiency on the third ball after serving. Placement distribution by zone and by score situation.

Without those five, any claim about a player is a description of memory. And memory cannot be audited.

Let me use what I actually recorded. At the SEA Games 31 held in Vietnam in May 2026, I sat courtside logging by hand through the competition days. My format then: one cell per point, with server, serve type, winner, number of contacts, finishing zone. A men's singles match ran nearly forty minutes on average. I spent roughly fifty minutes logging and thirty minutes cross-checking against video.

Two systematic errors surfaced in the first week.

The first lived in the time anchor. My clock and the electronic board drifted about seven seconds per game, compounding. Once matched to video, rallies slipped off their markers. I had to reset an anchor point: the sound of the ball striking the table on the first point of each game, captured through the audio waveform. After re-anchoring, error dropped below one second — the threshold needed to compute rally-length distribution.

The other error lived in definition. I counted a "service error" as a ball into the net or long. Coaches here count a "service error" as a serve that the opponent attacks first. Two definitions produced rates differing by a factor of three on the same dataset. Publish either one without stating the definition and I have manufactured a false fact.

That is why my three-step method exists. Record the raw number without embellishment. Trace where the number came from — who logged it, under which definition, in what conditions, with which games missing. Conclude with a confidence level, always leaving an exit door open for new data.

For Vietnamese table tennis datasets, most of my conclusions stop at the level of signal. I still publish them, clearly labelled.

Players such as Nguyen Anh Tu, Dinh Quang Linh and Tran Mai Ngoc each carry a distinct technical profile. The problem is that the profile lives in the heads of spectators, not in a file.

Look to another sport for the standard. In 2026, while doing analysis for a football outlet in Binh Duong, I published a home-made expected-goals model for Becamex Binh Duong against Ha Noi FC, predicting a 65% win probability for the hosts on the strength of superior possession. The result: a 0-3 defeat. Ha Noi held 38% of the ball but fired eleven shots from inside the box. I spent a month pulling tape to find the two missing variables: chance quality and central-attack speed. I then rewrote the whole algorithm, adding PPDA and the receiving positions of the holding midfielder.

That lesson transfers directly to table tennis. With only serve-point win rate and no finishing zone, I would repeat the old mistake: concluding from a variable that merely looks strong.

One more layer. In 2026, before the Euro final between Italy and England, I published an analysis built on PPDA: Italy at 9.2, England at 13.5. The numbers showed Italy pressing very early while England sat deep. I wrote that Italy would not let England breathe. The game unfolded exactly that way in shape, even though England led and Italy only won on penalties.

I do not conclude that PPDA predicted the outcome. I conclude that a metric only means something when it measures a repeating behaviour, and that behaviour must be visible to the naked eye so readers can verify it. In table tennis, the observable behaviours are the serve, the receive and the third ball. Without data, there is nothing to say.

In 2026, when competitions paused, I analysed four hundred matches across the Bundesliga and K League 1 to measure the effect of playing without crowds. Home teams won only 31% of matches instead of 44% under normal conditions. The empty stadiums of 2026 proved one thing: data without context is only half the truth. Vietnamese table tennis sits in exactly that state — except it is missing the first half too.

Contrarian angle

There is a professional pressure I have to name. Newsrooms need copy. Readers need conclusions. An analysis that reads "insufficient information" across all nine sections sounds like a defective product.

But transparency is not dumping every dataset. Transparency is keeping the decisive numbers, publishing the method, and stating plainly what has not been measured. An empty audit is more accurate than three thousand words padded with adjectives.

I have stood on the other side. In 2026, before the World Cup final, I wrote that France could not beat Croatia, based on expected goals: France at 1.8 per match, Croatia at 2.4. The piece drew more than two hundred thousand reads and a heavy backlash. France won 4-2. My error was failing to adjust the data for knockout-round opponent quality — Croatia had faced weaker sides in the group stage. I sat down and wrote a three-thousand-word self-rebuttal, published on the same page, with open data attached.

The data was not wrong; the reader was — and I had been that reader.

A thirty percent probability is not an excuse — it is a reminder that I am right seven times out of ten. Every model I have was built on mistakes that once got laughed at — that is the most honest foundation I own.

One more thing needs saying: reader feedback is a data layer. When someone tells me a player is serving better than last season and I cannot measure it, that person is pointing at a variable I omitted, not arguing with me.

Next-cycle signals

Starting next week, I change my logging standard. Every match I follow will ship a public point-level raw file, with variable definitions and confidence levels. Three signals I will track all season: serve-point win rate split by game, rally-length distribution, and the error margin between hand logs and video.

If an analysis returns nine empty sections, the correct response is not to fill those nine sections. The correct response is to go and rebuild tier one. The question I leave for myself: of all the conclusions I have published about Vietnamese table tennis, what share were really just memory delivered in a confident tone?