A “Basketball” Label and the Night I Almost Wrote With Thin Air

That night in Hanoi, my second screen showed exactly one line: Domain Label: basketball. Directly beneath it was an empty table. No team names. No player names. No metrics. Not a single shot recorded. The system still reported the processing status as successful. I sat still for four minutes, waiting for the data to pour itself into the blank cells. It did not pour.

For someone who writes about sport through numbers, that moment felt like walking into a post-game press room and discovering nobody took notes. You still have the memory of the game, the feeling at your fingertips of how a player moved, of rubber soles snapping against the wooden floor. But you have nothing to cross-check against, nothing to argue with yourself about, and worse, nothing to stop you from producing an analysis that reads beautifully and is hollow inside.

I have been in this trade for twenty-one years. Most of that time I have stared at stat tables. At twenty-eight, I wrote that Hanoi FC deserved to win 3-1 rather than scrape a lucky 1-0 against Quang Nam in the V.League, based on an xG of 2.87 against 0.45, 68 percent possession and fourteen shots inside the box. Back then I was mocked fairly heavily, with the familiar argument that football is not mathematics. A week later, the head coach admitted he had rewatched the tape and adjusted his tactics around exactly those numbers. That night, the media called them soulless. xG said otherwise, and I chose to believe xG.

But the night I am describing to you was different. There was no xG. No possession figure. No positional heat map. There was one label, and a gap the size of an entire game.

Ten years ago, talking about advanced metrics in Hanoi was close to talking to yourself. Sports desks back then lived on three things: the running commentary, the post-match quotes, and the writer's own emotions. Nobody objected, because it worked. It was fast, it was readable, and it demanded that no one know what PPDA stood for.

Then Vietnamese basketball began to change. The VBA expanded, clubs started hiring analysts, games were filmed from multiple angles, and event-tracking tables started appearing in locker rooms. Youth academies in Ho Chi Minh City and Hanoi began sending parents reports with indices attached instead of compliments alone. The national team prepared for SEA Games cycles and Asian qualifiers with a far denser data set than before.

I belong to the camp that believes this is progress, but not free progress. When a newsroom starts relying on data, it also starts depending on data. And everything we depend on can break.

That is why, for years, I have imposed an unspoken rule on myself: before offering any judgment about a game, I need at least three advanced metrics. Not because three is a sacred number, but because three metrics force me to cross-check. One metric can lie. Three metrics lying in unison is rare. But if all three are empty, I have no right to write.

That night, all three were empty. So I started asking where the system had failed.

Looking at the log, the problem was obvious to anyone who has worked with data: the classification layer had run successfully and written the label “basketball”, while the extraction layer returned an empty array. The label had been assigned from the source's metadata, possibly the section or the URL slug, not from the article's content. In other words, the system knew this piece belonged to basketball, but had not read a single word inside it.

In systems circles, this is called a silent failure. It is nothing like a crash. A crash is loud: there is an error report, an error code, someone gets paged. A silent failure is polite to the point of danger. It writes “success”, it packages the result, and it ships it. If no one downstream checks, that empty result travels straight into the next process and becomes the input for an analysis that reads very convincingly.

In basketball, the consequences of this failure mode are easier to picture than I expected. Imagine a game log where the points column for every quarter is left blank, but the totals row still displays fully. Someone on the coaching staff opens it, sees the totals add up, and concludes the log is complete. Nobody notices that the most decisive piece of information, the scoring rhythm by phase, has vanished.

Or imagine that PPDA, the number of passes an opponent is allowed before each defensive action, comes back as zero, not because the team defended loosely, but because the tracking system lost its signal for twelve minutes. That number does not sit at the bottom of the scale. It sits outside the scale. But in the table, it looks exactly like a data point.

This is the point I want to press, and it is also the hardest thing to teach a newcomer: in sports analysis, the most dangerous thing is not a wrong number but a number that looks right. A wrong number gets caught the moment someone checks it. An empty number pre-formatted into a cell quietly becomes the truth.

In 2026, I paid for a variant of this mistake. I built a model for the World Cup in Qatar on cumulative xG, goals scored and control metrics, and confidently said Germany would survive the group stage. Germany went out. Looking back, I realised my data set was missing an entire variable: Japan's defensive pressure, which registered a PPDA of 6.8 across their matches against Germany and Spain. That metric sat outside the system I had assembled before the tournament. I did not have a gap in my model. I had a gap in my own head, and the model merely reflected it back at me.

After that failure I spent weeks recovering my composure, then three months building a workflow that integrates multiple non-traditional data sources. Since then, every analysis I produce carries a mandatory section called risks and gaps. That night, that very section saved me.

Because when I reopened the empty table and asked myself what to write in it, the only honest answer was: everything. There was no metric to check. No player name to hold responsible. No tactical situation to dissect. Any conclusion about the game would be a product of imagination, dressed up in terminology.

In basketball, data gaps tend to appear exactly where we need them most: the transfer market. A twenty-year-old who has not played fifty top-flight games can still be priced at a figure that makes people in the industry hold their breath. I once sat in a meeting where the fee was agreed before anyone opened a stat sheet. In that moment, what filled the gap was expectation, and expectation has no sample size. A contract is only truly correct when the number signs alongside the signature.

I have seen the same pattern at a larger scale. In 2026, when the pandemic closed stadiums, I wagered that home advantage would fall from 54 percent to below 50 percent. When the Bundesliga restarted, Dortmund won only three of their remaining eight home games, and the league-wide home win rate dropped to 48.7 percent. The first half of my prediction held. But my recovery model failed badly, because I had not accounted for differences in training-ground quality and squad psychology. When the stands emptied, my model collapsed. I knew I had forgotten the human factor.

That night there were no empty stands. There was a different and colder void: a data gap at the root layer.

I spent most of that night doing something nobody sees and nobody pays for: checking whether the failure was isolated. I pulled the entire day's batch and audited it. I found three more cases with the same signature. Label present. Content absent. Three other articles in the same processing run had travelled through the pipeline inside that same empty shell.

That was when I understood the problem was not one article. It was a door with no lock.

The first reaction many people have to this story is to suggest abandoning automation altogether and going back to manual work. I disagree, and this is where I part ways with the room.

Removing automation does not make data more honest. It only makes the gaps harder to see. When a person types every number by hand, the gap does not disappear. It moves from the screen into memory. And memory, for a sportswriter, is the most biased data source ever created.

I have read a great many basketball analyses in which the author never once mentions what they are missing. They describe a team with three metrics, then conclude with twelve adjectives. Those adjectives are precisely where the gap gets filled. Readers cannot see the seam between three metrics and twelve adjectives, so they assume everything is grounded. I have written that way myself, and I know how comfortable it feels. It grants the sensation of having understood a game, when in fact you have merely restated the crowd's emotions in an expert voice.

There is a paradox I carry with me: an honest gap is useful, a filled gap is harmful. An article that says “I have no data for the first twelve minutes of the third quarter” will irritate readers, but it is true. An article that says “the team started slowly” with nothing behind it will please readers, and it is false. Between those two options, my trade has chosen the second far too often.

In Vietnamese basketball the pressure is even heavier. A VBA season lasts only a few months, the game count is low, and the sample is so small that every statistical conclusion is fragile. A player who scores thirty points in two straight games can be written into a phenomenon. But two games are not a trend. That is why I do not trust hunches. But I do trust what a hunch looks like once the data confirms it.

And for the same reason, I do not treat the human factor as noise. That is the easiest mistake for a data person to make, and I have made it. A player's emotional state, the pressure of a home crowd, an unhealed injury, a contract running down, none of those live inside my model, yet all of them live inside the game. My gap is not a place to stuff emotions into. It is a place to admit that I do not know.

There is a line I wrote years ago and still use: a number never needs us to defend it. It is the other way round. We need numbers so that we do not lie to ourselves. But that line only holds when there is a number. When there is nothing at all, the thing I need to protect is my honesty about the gap.

So what did I write that night? I wrote a two-page internal memo in which three quarters of the content was what I did not know. My editor read it and asked the first question: so can we publish anything.

We published one thing. A short note saying the newsroom's data system had failed and the analysis would be delayed by a day. Nothing exciting. Nothing that spread. But it was true.

The next day, when the data feed came back, I realised the most expensive lesson lay somewhere else entirely. The problem was not that I nearly published an empty piece. The problem was that I had been confident enough to write it, and the only thing that stopped me was a system log line appearing in front of my eyes. If that label had not shown up, if the screen had simply been white, I might have assumed I had forgotten to launch the tool and written from memory instead.

That is a category of risk no metric can measure.

The signal I am tracking over the next cycle is not on the scoreboard. It is in the validation gates that Vietnamese sports desks are building, or have yet to build. Three concrete things: a hard gate that rejects every empty result before it moves downstream; a field recording the provenance of each label, so we know whether it came from content or from metadata; and a batch-wide audit whenever a single anomaly appears, because silent failures rarely travel alone.

For readers, the signal is simpler still. Next time you read a match analysis and every sentence flows smoothly, try to find what the piece admits it does not know. If there is nothing, the gap has probably been filled, and the writer is selling you confidence instead of information.

For me, the lesson of that night is wrapped in a line I will keep repeating to myself through this season: data shows a trend, but it is not a prophecy. And when the data falls silent, the most decent writer is the one who knows how to fall silent with it.