Volleyball and a Blank Data Sheet: When the Model Learns to Stay Silent

VOLLEYBALL AND A BLANK DATA SHEET: WHEN THE MODEL LEARNS TO STAY SILENT

Three screens, one blank space

2:47 a.m., September 14, 2026, an eleventh-floor apartment in Binh Thanh District, Saigon. Three screens glow on a long wooden desk. The left screen carries a live stream of a women's volleyball match inside a regional competition system. The middle screen shows a real-time statistics panel: live points, service point rate, blocks per set. The right screen is the output of a data-collection script I wrote in 2026, the thing I jokingly call "the counter".

The right screen is blank. Not blank because of a display error. Blank because the data structure returned empty.

I sat there another forty minutes. The match kept running. Players rotated. The referee blew the whistle. The stands stayed full. But my counter had not one line to count. And what I remember about that night is not the technical failure. What I remember is my own next reflex: I started asking myself whether I should "fill in" that blank space with feeling.

Thirty years ago I did exactly that without knowing it. Now I know. A blank space in a sports data table is the most dangerous thing an analyst can encounter, more dangerous than a wrong number, because a wrong number gets caught. A blank space tends to get filled with whatever the reader already has in mind.

Context: the volleyball data infrastructure few people ever see

Volleyball is a strange sport in Vietnam's team-sport landscape. Football has a data ecosystem two decades mature. Basketball has detailed box scores. Volleyball's audience still consumes the game through three things: the score, the name of whoever scored, and gut feel.

At the operational level, however, volleyball has a far denser data chain than people assume. I sort it into four layers.

Layer 0 — the referee's sheet. Score, errors, rotation position, service order. Almost never administratively wrong, but tactically almost empty.

Layer 1 — video coding. This is where professional volleyball actually lives. A coder rewinds every rally and tags it: wing attack, back-row attack, middle attack, set tempo, one-blocker or two-blocker, dig or no dig. Software such as DataVolley and VolleyStation is born here. Southeast Asian national teams now use this layer at least in official AVC events.

Layer 2 — derived metrics. Perfect-pass rate, sideout percentage, kill efficiency net of errors, digs per set, blocks per set, ace-to-error ratio.

Layer 3 — the public layer. News, broadcast graphics, arena stats boards, betting markets. This is where every error from the three layers below gets amplified, because there is no filter here.

The break point sits between Layer 1 and Layer 3. There is an intermediate layer rarely discussed: automated fetching and extraction. A background chain pulls data from competition pages, from organiser statistics systems, from small APIs someone wrote in a hurry, then decomposes it into discrete fields: events, teams, players, coaches, competitions, dates.

That layer is where the night happened to me.

It did not fail because Vietnamese diacritics were parsed wrong. It failed because a page changed structure, a new piece of JavaScript moved the content out of static tags, and my script read an empty frame. The structure remained intact: nine fields, nine headings, nine slots. Only the contents were empty.

And here is where I must say the hard thing about my own industry: a fully-formed structure containing nothing but empty content is the most dangerous input any analysis pipeline can receive, because it looks perfect.

The anatomy of an empty payload

"No data" versus "zero data"

This is the fatal distinction almost every automated pipeline gets wrong, and I have watched it repeat at least four times in a decade of regional volleyball coverage.

"No data" means the column is blank. No row. No value. The reader knows the thing was never measured.

"Zero data" means the column holds a 0. It was measured, and the answer was nothing.

The two mean completely different things, yet many systems process them identically. When a team records no blocks in a set, that is enormously valuable tactical information: it says their block system was neutralised. When a system records no blocks because the coder sat in the wrong spot or the cameras did not cover the angle, that is information about the data, not about the team.

Mixing the two produces a stat sheet that looks cold and objective while actually telling a fabricated story.

Nine analytical dimensions and how they collapse together

I split my volleyball analysis into nine dimensions. They have one thing in common: every one of them needs at least one original data point to begin. One match. One player. One competition. One date. One number.

When the payload is empty, all nine collapse at once, like a building losing its ground-floor columns. Tactics has nothing to analyse. Data has nothing to compare. Schedule has no anchor date. Landscape has no team name. Rules has no governing body. Personnel has no coach. Risk has no risk to list. Narrative has no headline to read. Industry chain has no link to connect.

What is frightening is that if you only look at the output format, everything sits in place. Nine headings. Nine tables. Nine conclusions. A downstream process that does not check carefully will read that and conclude that analysis was performed.

That is why I am writing this. A system can only reach a correct conclusion if it knows how to refuse to conclude when its hands are empty.

When the perfect-pass column disappears

Take volleyball's cleanest example. Perfect-pass rate measures the share of first contacts delivered to the ideal position, letting the setter run the full attack menu. It is an input to almost everything else.

With that metric you can read the system. You can see that below a certain threshold a team is forced into out-of-system attack, meaning individual ability replaces structured play. You can see how weak a two-attacker rotation really is.

Without it, you lose the ability to distinguish two completely different situations: a team choosing to attack out of system as strategy, and a team pushed there because its reception system has broken.

Two teams can post identical attack efficiency in one match while one controls the game and the other merely survives. Read only the final score and you will misjudge both.

Croatia was not a fairy tale; it was a problem that needed to be solved again from scratch. In football that problem lives in xG. In volleyball it lives in perfect-pass rate and sideout percentage.

Zero is not an event, it is a state

When a player records zero blocks in a match, it does not mean she played badly. It may mean the opponent avoided her position entirely, which is respect. It may mean the block system was designed for her to read the play while her teammate makes contact. It may mean the opponent deliberately set away from the middle.

A zero in a volleyball stat sheet is a question mark, never an answer.

That night I had an advantage: I knew what I was missing. I knew a website had changed structure. Had I not known, I would have read a sheet of zeros and told a story about a team that cannot block and a setter who cannot organise.

I have told stories like that. 2026 taught me to listen to what the model cannot measure. But only that blank-screen night did I learn to separate two states that sound alike: the model staying silent, and the model never being given data to speak.

The garbage-in, garbage-out loop

When nothing goes in, something still comes out, and it usually comes out more confident than when real data exists. With data, an analyst is constrained and must bend the story to the numbers. Without it, there is no constraint, and the story flies on the writer's own momentum.

During the pandemic I had to rebuild a domestic volleyball stat sheet after organisers published match data three days late. In those three days, analysis flooded social media: verdicts on a team based on score, blame assigned to individuals. When real data arrived, most of those pieces were right about the conclusion and completely wrong about the cause. In the middle of a pandemic, I counted history again and found every cycle wearing a familiar face. The familiar face is this: people always have an explanation ready, even before evidence exists.

The silent part of the model

I am a man who trusts numbers. But I do not believe everything can be measured, and I do not believe the unmeasurable should be denied.

In volleyball there are at least four things I have tried to measure and failed at.

The first is psychological pressure in a decisive fifth-set rotation. You can measure points, errors, efficiency. You cannot measure what a player feels at 12-14 down, standing behind the 9-metre line.

The second is the quality of a set. Some sets are logged as perfect but force the hitter to jump from a bad posture. Some logged as imperfect open a better rally. The coder sits two metres from the court; the hitter is in the air.

The third is the value of reading the game. A libero who steps early and is already in place never appears in a stat sheet. Only the rescue appears. The system records only what nearly failed.

The fourth is one person's influence on five others. That is what I call the silent part of the model, and I have no tool to measure it except rewatching the whole match chronologically, start to finish, without fast-forwarding.

What 2026 taught me was not how to measure more. It was how to notice when my data sheet is talking about itself rather than about the match.

The contrarian angle: more data is not automatically better

The whole industry runs on one assumption: more data means more accuracy. It does not follow.

The volume of data is growing faster than the speed of data verification. The share of unverified data in any stat sheet is rising, not falling.

With three metrics you can cross-check by eye. With three hundred you cannot. You trust the process. And the process, as that night showed, needs only one small change at the collection layer to turn three hundred metrics into zeros without raising an error.

There is also a correlation I must separate from causation: teams with the most data are usually the strongest teams. That invites the conclusion that data creates success. In reality, strong teams have the resources to build data systems and to recruit good players. Both come from the same cause, not from each other.

And the market does not pay for silence. An analyst who says "I do not have enough data to conclude" loses a piece, a click, a client. An analyst who says "Team A will win because X, Y, Z" gains everything. The incentive structure pushes this industry toward fabrication, and nobody calls it fabrication because it is dressed in technical language.

What remains after the blank space

A miracle, in most cases, is a maths problem that has not been solved correctly. Yet there is an exception: when you lack the numbers to solve it, calling it a miracle is a form of honesty. The problem is that you must say so, rather than pretend you finished the work.

Signals for the next cycle

In the current annual-season cycle I will track three signals at the data layer, not the standings layer.

First, whether domestic competitions publish perfect-pass rate and sideout percentage consistently. If they do, Vietnamese fans gain a far deeper way to read a match than the score alone.

Second, whether published stat sheets distinguish a blank from a zero. A sheet using a dash instead of a zero is far more trustworthy, because it shows the data people know what they are doing.

Third, whether analysts are willing to say "not enough data". That is the hardest signal to observe, because it lives in silence. It is also the most important.

My three screens still run every night. The counter has been fixed. I added one simple line of code: if the returned row count is zero, print a sentence, not a table.

The sentence reads: there is nothing to say.

To me, that is the most important line of code I have written in thirty years.