An Empty Model and the Discipline of Not Writing for the Data-Driven Sports Writer

Hanoi, 11:40 p.m., a Saturday at the end of January. The third monitor in my small office was still on. I had just finished running the analysis pipeline for a post-game report — the pipeline I have built and tuned for seven years: collect raw data, extract entities, cross-check sources, tag risks, and only then allow myself to write the first sentence. That night, the result came back empty. Headline: empty. Source: empty. Information points: not a single one. Entities: not one name of a player, team, or coach. A column of the words “undetermined” ran down the bottom of the table.

I stared at it for about ten minutes. Then I shut the machine down and wrote nothing. To this day I still believe it was the best decision I have ever made on a deadline night.

The gap that costs more than the conclusion

In this trade, people usually talk about the pressure to write well. I think the real pressure lives elsewhere: there is a hole on the page, and that hole is waiting to be filled. A basketball or football match has just ended, readers have opened the app, the desk has set the clock, and the clock does not care whether your sources are rich or thin.

I learned this fairly early. In 2026, when I was a young data editor in Hanoi, I wrote that Hanoi FC deserved to win 3-1 rather than scrape a fortunate 1-0 against Quang Nam in the V.League. My basis was the expected-goals figure for the whole match: 2.87 against 0.45; 68 percent possession; fourteen shots inside the box. The piece was mocked, and someone said football is not mathematics. A week later, the coach admitted he had rewatched the tape and adjusted his tactics based on that analysis. It was the first time I saw that data could shape real work on the pitch, rather than merely describe it.

Since then, I have set myself a rule: do not offer a judgment on a match without at least three advanced metrics standing behind it. It sounds simple. But that very rule is what forced me to shut the machine down on that January night.

That night, the source I received was an empty file. Not a thin source. Empty. No team name, no scoreline, no player, no context. A shell with nothing inside. And in the moment the screen filled with that row of “undetermined”, I understood I was facing exactly the temptation this trade always sets out: write the shell, and let the reader imagine the filling.

Three gates before writing

I run my sports analysis through three gates. The first gate is information extraction: the source article must yield discrete information points, and each must be given an index number, so that every later conclusion has to point back to a specific point instead of floating free. The second gate is entity extraction: there must be at least one correctly named entity, whether a team, a player, a coach, or a competition. The third gate is source labeling: who said it, when, and where.

These three gates may sound administrative, but they are a fence against one specific temptation. When an analysis contains no information points, every conclusion drawn from it is a product of the writer's imagination rather than of data. And the imagination of a sports writer is fertile. It knows how to conjure a match that never happened, a player who never played, a contract that was never signed.

That is why I never treat an empty input as “nothing to say.” It is a signal. It says that something broke at the very first stage — either the source was never loaded, or it was loaded but could not be parsed, or what was called a source was in fact navigation content, menus, ads, not one gram of basketball substance. Three different possibilities, but one shared outcome: the writer has nothing in hand.

I once thought I understood the limits of data well. In 2026, when the pandemic froze football, I threw myself into building a home-advantage dataset going back to 2026. When the Bundesliga restarted with empty stands, I bet that home performance would fall from 54 percent to below 50 percent. The results seemed to back me: Borussia Dortmund won only 3 of their remaining 8 home games, and the league-wide home win rate dropped to 48.7 percent. But my model for predicting the recovery failed badly, because I had not accounted for differences in training-ground quality and squad psychology.

Numbers show a trend, but they are not a prophecy. That sentence now opens everything I write. It also explains why an empty input unsettles me so much: I cannot build a trend out of nothing.

When the algorithm falls silent, people tend to talk nonsense

There is a kind of failure in analysis that outsiders rarely see, called silent failure. It happens when a system returns the words “not applicable” instead of a clear error message. Someone reading the results table will assume everything ran fine, while in fact nothing ran at all. “Not applicable” and “missing data” look almost identical on screen, but they sit a great distance apart in meaning. One says the item is irrelevant. The other says the item should have been there but was skipped.

For a writer, that ambiguity is poison. Because when the system goes quiet, the natural human reflex is to fill the gap. I have seen this many times in newsrooms. An empty stats table, a murky source, an unconfirmed name — and instead of waiting, people write around it. They build stories about spirit, about character, about the “class” of a team without ever knowing how that team actually played.

In basketball, the trap is subtler. A player who scores 30 points looks dazzling, but if he took 28 shots and his true shooting efficiency was low, then that 30 is hiding a problem rather than revealing a solution. To see that, a writer needs data on usage, on true shooting, on plus-minus when the player is on the floor. Without those, the “star shines” story is the easiest to write and the easiest to get wrong.

By contrast, a player like Nikola Jokić of the Denver Nuggets was underrated for years because he did not fit the flashy headlines. Advanced impact metrics told a very different story, and it took a long time for the media to update its view. So it was with Stephen Curry: his gravity off the ball — something only spacing metrics can capture — never appears in a box score. The box score is not wrong. It is simply not enough.

That is why I always demand of myself a clear distinction between “no data to assess” and “data indicates low risk.” These two states are entirely different. The first is an open question. The second is a grounded conclusion. Blurring them is the fastest way to turn analysis into propaganda for one's own carelessness.

Information gain, what readers actually need

There is a concept I borrow from search work and apply to writing: information gain. Put simply, every article must give the reader at least one thing they did not already know. If my piece merely repeats a scoreline everyone has seen, retells action everyone has watched, then I have added nothing but a few lines of text.

But information gain can only come from the source. An empty source cannot produce gain. A murky source produces only fake gain, something that looks like insight but is really a guess dressed up in confident prose. And in sport, where crowd emotion runs stronger than in most fields, fake gain spreads like fire in wind.

I once watched a colleague build an entire tactical analysis on an “informed source” that turned out to be a deleted status update. The piece was good, the reasoning coherent, and completely wrong. Readers had no way to know. Only when one detail was accidentally cross-checked did the whole newsroom realize it had just published a building on sand.

So I spend most of my time not writing, but checking sources. It sounds unglamorous. But that is the work that decides whether a piece will stand after a week, after a month, or collapse the moment someone bothers to check.

When the stands emptied, my model collapsed

I still remember the feeling in 2026, when I was invited to serve as an analysis expert for a major tournament. I built a model on expected goals, goals scored, and control metrics, and confidently predicted that a big team would advance from the group on the strength of the highest accumulated data. The result: that team was eliminated in the group stage. Looking back, I realized my model was missing a piece — the opponent's defensive pressure, which lay outside the dataset I had collected before the tournament.

That fall left me shattered for a few weeks. But it taught me something I will never forget: the most dangerous thing is not missing data, but not knowing you are missing data. A confident model with a blind spot is worse than a model that knows its own limits. Since then, every analysis I do has a section called “risk and gaps,” to say plainly what I cannot measure.

When the stands emptied, my model collapsed. I knew I had forgotten the human factor. But precisely because I knew I had forgotten it, I could write about it instead of ignoring it.

Croatia at the 2026 World Cup is the reverse example. That year I went to Russia to report, and while most colleagues picked the big teams, I wrote that Croatia would reach the final. My basis was the midfield's average total distance of 112 kilometers per match, the highest in the tournament, together with the trio of Luka Modrić, Ivan Rakitić, and Marcelo Brozović registering a very low PPDA. At first the piece was dismissed as baseless shock. Only when Croatia actually beat England in the semifinal did a group of international reporters acknowledge it. Croatia did not reach the final through luck. They reached it on legs that did not know how to stop.

What do the two stories have in common? In both, data mattered. But in the second, I had enough input to build an argument. In the January night story, I had nothing at all. And knowing you have nothing turns out to be a skill.

Counterintuitive: the value is in the times you do not publish

In this trade, people celebrate controversial pieces, the times public opinion is overturned, the numbers that make a whole arena go quiet. But if you look closely at those who last in the profession, you realize their real value is not in the times they published. It is in the times they decided not to publish.

An article built on an empty source can be very fluent. It has a striking opening, a climax, a conclusion. It is missing exactly one thing: the truth. And readers, especially sports readers, often have no way to verify whether what they just read was woven out of thin air.

I do not believe in hunches. But I believe in what a hunch confirms through data. The line between those two sentences is the line between an analyst and a storyteller who makes things up. When the source is empty, my hunch still whispers all kinds of versions. My job is not to let it outrun the evidence.

There is another trap a data writer easily falls into: using counterintuition as a gimmick. After a few pieces succeed by reversing the crowd's judgment, the writer begins to believe the crowd is always wrong and he is always right. But counterintuition is only valuable when it has evidence. Otherwise it is just contrarianism for show. Before every piece, I ask myself: what if this year's data agrees with the crowd? If the answer is “I still have to write the opposite for the sake of being different,” then I am working as an actor, not an analyst.

And the final trap, the one that nearly pulled me in that January night: using data's silence as an excuse to write on feeling. “No numbers, so let's just write from the gut.” I have heard that sentence many times. But no numbers means no basis, not permission to be arbitrary. Emptiness is not a license. It is a barrier.

Numbers never need us to defend them. Rather, we need them so we do not deceive ourselves. Even when the only number we have is zero.

A signal for the next cycle

I wrote nothing from that empty file. But I spent the next morning adding a fourth gate to the process: an input-completeness gate. If an article enters analysis and yields no information points and no entities, the system will return an error instead of a table full of “undetermined.” I want the error to be loud. I want it to shout, so that no one — including myself on a tired night — can accidentally turn silence into an article.

In sport, people often talk about matches decided by moments no one sees: an off-ball run, a well-timed substitution, holding the ball an extra half second. In data-driven writing there are similar moments, except they happen before the article is born. A source file returned empty. An error line skipped over. The words “undetermined” misread as “all is well.” Those moments never make the page, but they decide whether the page can be trusted.

I will keep tracking the rate of empty inputs in my process. Not to eliminate them — I cannot control whether sources are always full — but to understand them. Each time a source comes back empty is a chance to ask myself what I am relying on when I write about a match. And each time I can answer that question without inventing anything, I know I am still standing on the right side of data.

That January night, on my desk there was a gap and a deadline. I chose the gap. The reader lost nothing. If I had written, they would have lost the most expensive thing a person in this trade can give: the belief that the words on the page are words that have been verified.