An Empty Spreadsheet at Minute 90: The Discipline of a Data Sportswriter

The clock in the Hanoi newsroom read 11:40 p.m. on November 12. I opened the spreadsheet named after a match code, scrolled to row three hundred, and saw a field of white. The advanced-metrics column was empty. The minutes-played column was empty. The player-name column was empty. The header row still carried a date, a competition, a round, but beneath it not a single cell had been filled. In eighteen years on the job I had grown used to bad datasets: noisy data, skewed data, data missing three columns, data with player names misspelled beyond any hope of matching a profile. That night was the first time I met a completely empty dataset, and I had forty minutes to decide: write, or don't.

I chose not to write. But the reason behind that choice took six more months to explain to my desk, to a few colleagues, and to myself. This article is that explanation, told from the beginning.

In sports media people talk about two kinds of risk: the risk of saying something wrong and the risk of saying it late. Very few talk about a third, more dangerous than both — the risk of producing a conclusion with no root. Such a conclusion is not wrong in its wording. It is not late in its timing. It simply was not born from anything that exists. And in a news environment where every passing hour buries your piece under a thousand others, that third risk is the easiest to fall into.

How a dataset is born

To understand how a spreadsheet can be empty, you have to understand how it comes into being. In the pipeline I built and have run for years, every piece of sports analysis passes through a first layer I call the extraction layer. That layer does not analyse. It does one thing: it reads a source article and pulls out the title, the publication source, the timestamp, a list of information points, a list of named entities, the author's stance, and the time-sensitivity flag.

What is an information point? It is an atomic fact. A player scored 32 points. A team won four straight. A contract was worth 47 million euros. A coach was sacked after eleven rounds. Each such fact must stand on its own, carry a source, and be verifiable if challenged. When the extraction layer does its job, it returns a packet. When it fails, it returns an empty one: no title, no source, no assessed timestamp, an empty entity list, an empty information-point list.

On that November 12 night, I received exactly such an empty packet. On my screen it appeared as a spreadsheet with nineteen columns, all white.

Here is what is worth noticing. If someone hands you an empty spreadsheet and a nine-dimension analysis template — tactics, player data, team operations and salary cap, league landscape, rules and governance, coaching staff and locker room, risk, media narrative and expectations, industry ripple effects — the template itself creates pressure to complete. Empty cells do not sit still. Empty cells call out to be filled. And if the writer has no clear rule, the cells get filled with something that sounds plausible, reads smoothly, and is entirely untrue.

I remember sitting beside a young editor on a finals night years ago. He handled the live stat sheet. When the data provider's system went down in the second half, he spent four minutes doing what he thought was reasonable: he took the first-half figures and doubled them. Field-goal percentage unchanged. Turnovers doubled. Contested possessions doubled. He printed it, handed it to me, and said in a very confident voice that the system was back.

That sheet was not structurally wrong. It was merely mathematically correct and athletically false. Basketball does not run on multiplication. A finals second half runs on substitution rhythms, foul trouble, and whether the leading team chooses to slow down. A number generated from an assumption will never reflect that, and worse, it will become evidence for someone else's next article.

That is why, when the empty packet appeared on my screen, the question in my head was not "what do I write". It was "what do I have".

Context: the big season and the pressure to be present

November is the heaviest month of the year for a sportswriter in Vietnam. The domestic professional basketball league is entering its decisive stretch, international basketball competitions run in parallel, and across the ocean the professional season tips off with hundreds of games in six weeks. The calendar is so dense that a match analysis written at 11 p.m. will be read as obsolete by 7 a.m., because by then four other games have finished.

Under those conditions, speed becomes a form of credibility. Whoever publishes first is treated as having better sources. Whoever has the numbers the same night is treated as having a better system. I understand that psychology, and I once chased it. At twenty-eight and twenty-nine I believed fast reflexes were a professional skill. I still believe that, with one condition: the reflex must serve observation, not the filling of a blank in a template.

The context of that night was specific. I was assigned a post-game analysis for a match in an important round of the ongoing season. It is the kind of piece I write weekly, familiar to the point where I carry a template in my head. Open with an anomalous metric. Build the tactical setting. Present the evidence chain. Overturn the conventional read. Close with a signal for the next round.

But to do any of that I need to know which teams, who played, how the ball circulated, where each side pressed, who drew fouls, who sat out. The empty packet told me none of it. I could not even identify which league was involved, because the only populated field was a single broad sport label with no federation, no team, no player name.

I spent the first twenty minutes looking for a source by other means. I rechecked my inbox. I reopened my match log, a file I have kept since 2026, recording every game I actually watched from start to finish. I cross-referenced it with the day's fixtures. Those twenty minutes produced nothing except the conclusion that I had not been present at any relevant event. I cannot speak about something I never saw.

The core: four times I was wrong because I trusted a dataset

To explain why I decided to stop, I need to tell you about four times I kept going when I should have stopped.

V.League 2026: a correct metric with the context left out

In 2026, aged twenty-eight, I worked as a data editor for a football site in Hanoi. After a V.League match between Hanoi FC and Quang Nam, I argued that the home side deserved to win by three goals rather than the lucky 1-0 the media described. My basis was expected goals: 2.87 against 0.45 over the full match, 68 percent possession, and fourteen shots taken from inside the box.

I was mocked for a week. People told me football is not mathematics, that spreadsheets only look good on paper, that the stands do not award points for expected goals.

A week later, coach Chu Dinh Nghiem admitted at a press conference that he had rewatched the tape and adjusted his team's approach based on those very numbers. For the first time in my career I saw data not only describing a match already played but shaping one still to come.

But what I remember more clearly from that 2026 article is something else. I never mentioned that the away side fielded a patched-up back line because two centre-backs were suspended and a third was carrying a knock. I never mentioned that the match was played on a rain-soaked pitch, which made aerial balls pointless but grounded passes twice as dangerous. My expected-goals figure was correct. My explanatory chain covered only part of the event, and I stayed silent about the rest because it had no number attached.

That was the first lesson: a correct metric has never been enough to make a correct conclusion.

Croatia 2026: when a physical model overturned the media verdict

In 2026, aged twenty-nine, I travelled to Russia for the World Cup. While most colleagues picked Brazil or Germany for the title, I wrote that Croatia had the midfield with the highest average total distance covered, 112 kilometres per match, best in the tournament. The trio of Luka Modric, Ivan Rakitic and Marcelo Brozovic posted a PPDA of 8.2 — meaning opponents completed only 8.2 passes before being pressed, one of the harshest figures at the tournament.

The piece was dismissed as baseless shock value. The phrase I heard most was "a fitness model". People said a team cannot reach a final by running more than everyone else.

Croatia reached the final. Across three consecutive knockout rounds they played 120 minutes. In total they played nearly one full extra match in duration compared with their opponents. When extra time in the semi-final ended, I wrote a line in my notebook that I have used again and again: "Croatia did not reach the final because of luck. They reached it because of feet that do not know how to stop."

The lesson here is subtler than 2026. I was right, and right because of a physical metric. But within that very article I presented the conclusion as an inevitable consequence of a model. I wrote as though distance covered could decide a football match. In reality, what decided the semi-final against England was a header in the 109th minute, and behind it lay a set of factors my model could not measure: pitch quality, the psychological endurance of individual players after three prolonged matches, and whether the opposing coach dared make a substitution at minute 90.

2026 and 2026: half a year in the NBA, and empty stadiums

The period from 2026 to 2026 was when I worked more deeply on professional basketball at a major outlet. During that stretch I came to understand that basketball and football teach the analyst two opposite lessons.

Basketball teaches large samples. With eighty-two games a season, each team plays more than three hundred competitive minutes every two weeks, and the pace is so high that every metric updates continuously. Basketball is also the sport where a peak player can affect thirty percent of his team's possessions. One person changes the entire system. It is ideal ground for advanced metrics.

Football teaches the opposite. With thirty-eight rounds a season and eleven players on the pitch, an individual's influence is diluted to the point where advanced metrics typically explain only about half the variance in results. The other half sits outside the spreadsheet.

That difference is the foundation of the biggest shock of my analytical career. In 2026, when the pandemic paralysed European football, I had spent six years building a home-advantage dataset going back to 2026. When the Bundesliga restarted behind closed doors in May 2026, I staked a fairly bold prediction: home advantage would collapse.

The Bundesliga home-win rate during that suspended-and-restarted period fell to 48.7 percent, down from a long-run average of roughly 54 percent. Borussia Dortmund won only 3 of their remaining 8 home games. On the surface, my model was numerically right.

But when I tried to use it to predict when crowds would return in other leagues, it failed completely. I had not accounted for the vast disparities in training-ground quality and conditioning between clubs during lockdown. I had not accounted for sides choosing an ultra-defensive approach to limit infection and preserve squad depth. I had not accounted for coaches treating the period as a chance to rotate youth. My conclusion about home-win rate was correct. My recovery model was wrong. And I understood that I had built a model perfectly suited to a world that no longer existed. I wrote in a summary at the time: "When the stands were empty, my model collapsed. I knew I had forgotten the human factor."

World Cup 2026: when raw data hid a defence

In 2026, aged thirty-three, I was invited by a major Vietnamese outlet to serve as an analytical expert for the World Cup in Qatar. I built a prediction model on cumulative expected goals, actual goals and possession share, and arrived at a fairly confident conclusion: Germany would advance from their group as the side with the highest cumulative expected goals in it.

Germany went out in the group stage. Japan topped that group.

When I sat down to analyse my own failure, the problem surfaced quickly. I had not collected Japan's PPDA before the tournament. Recomputing afterwards, Japan posted a PPDA of 6.8 across their matches against Germany and Spain — a pressing level among the harshest at the tournament, and entirely absent from the dataset I had prepared.

I was devastated for weeks. Not because the prediction was wrong, but because I had been overconfident on a dataset I knew was incomplete. I knew I had not collected enough defensive metrics. I still wrote as though my dataset were comprehensive.

I spent the following three months rebuilding the system, integrating additional non-traditional data sources, and, most importantly, mandating that every subsequent analysis carry a dedicated section called "risks and gaps". That section lists plainly what I cannot measure, what my data does not cover, and what could cause my conclusion to collapse.

Since then I have stopped using the phrase "the decisive metric". Data decides nothing in basketball or football. People decide. Data describes a very small part of what they decide.

Anatomy of an empty table

Back to the night of November 12. Having lived through those four lessons, I sat looking at the empty spreadsheet and asked myself: what exactly am I missing?

I am missing a subject. No team name, no player name, no coach name. Not a single entity is referenced. This is the most severe gap, because every piece of sports analysis begins with a name.

I am missing league context. The only sport label in the file is a generic word that cannot distinguish the American professional league, international competition, Asian competition or domestic competition. That distinction matters far more than outsiders assume. Defensive three-second rules and zone rules differ between the American professional league and international play. Three-point line distances differ between systems. Paces of play differ so much that applying one league's benchmark to another produces a wrong conclusion at the starting point.

I am missing performance data. No points, no rebounds, no assists, no true shooting percentage, no efficiency rating, no plus-minus, no usage rate. Not one figure to cross-check.

I am missing salary structure. No maximum contracts, no mid-level tier, no rookie-contract surplus, no luxury-tax threshold, no cap position. In professional basketball, roster-building tools such as Bird Rights, the mid-level exception, traded player exceptions and the stretch provision all depend on a specific cap sheet. No cap sheet, no analysis.

I am missing a time stamp. The time-sensitivity field was never assessed. That alone blocks any analysis tied to trade deadlines, extension windows and tax-line decisions.

I am missing a source name. The original article has no source. No outlet, no date, no author. In my trade, information without a source is worth approximately nothing, however compelling its content.

This is the point where I want to linger, because it is the heart of this article. There is a widespread misunderstanding that the value of an analysis lies in its conclusion. In fact, the value lies in the chain that leads to the conclusion. If the chain is empty, the conclusion, however elegant, is a floating object. And when a floating object is published, readers carry it away and use it as the foundation for another floating object.

In this specific case, if I had filled those nineteen empty columns with league averages, with the player names most mentioned that week, and with tactical observations that sound reasonable, readers would not have been able to distinguish the product from one built on real data. The polish would have been identical. Only the roots would differ — and roots are the one thing you cannot see from outside.

An empty spreadsheet is not a technical problem. It is a professional-ethics test, and it appears precisely when you are tired, rushed and most in need of recognition.

The contrarian angle: this trade does not reward stopping

It is time to say plainly what most people in the industry know but few write down.

In sports media, the incentive structure leans toward being wrong. You are paid for published articles, not withheld ones. You are measured by readership, not by the share of conclusions that still stand three months later. A piece that is wrong but fluent and published on time will outperform a piece that is right but published two days late, and two weeks later nobody remembers which was which.

This produces a consequence I consider more serious than simply reporting something false: it produces a layer of analysis that is not real yet looks entirely real. This layer does not deceive readers with fully fabricated numbers. It deceives them with real numbers placed in a chain that does not exist. A correct metric attached to a wrong cause. A correlation presented as causation. A four-game sample presented as a season-long trend.

And the way it happens is rarely dramatic. It happens at 11:40 p.m., when you have a template to fill, a deadline approaching, and a spreadsheet refusing to give you anything.

In basketball this phenomenon takes a particular form. Because each season has eighty-two games, a writer always has a huge sample to select from. You can choose four games to prove a point and ignore the other seventy-eight without anyone noticing, unless someone spends two hours checking. In football it takes another form: because a season has only thirty-eight rounds and goals per game are low, writers tend to inflate the meaning of a single match, turning a moment into an essence.

Both forms lead to the same outcome: readers are given a feeling of understanding without being given the ability to understand.

I want to be clear here, because this is easily misread. I am not against writing fast. I am not against writing the same night. Most of my career has happened on such nights, and I believe speed has real value. An analysis delivered two days after a match has far less practical value than one delivered within hours, because by then the coaching staff have made their own decisions and the window to influence has closed.

What I oppose is using speed as an excuse to surrender to emptiness. Writing fast with real data is a respectable skill. Writing fast with no data is a harmful act.

In professional basketball, this incentive structure is amplified by the transfer market. Every summer, hundreds of rumours are pushed out at breakneck speed, most without verified sourcing, and most retold in assertive language. There is a line I still use with young reporters: "A contract is only real when the number is signed alongside the signature." Until the signature lands, every figure is a hypothesis, even when delivered in the most confident tone.

On the other side, the agency market and sponsorship contracts create another layer of silence. Modern athletes are bound by a web of commercial obligations that makes them speak less, speak more safely, and speak more formulaically. When that happens, the media lose their most important direct source and tend to fill the vacuum with speculation. Speculation is not methodologically wrong. It simply needs to be labelled as speculation, not presented as a data finding.

This is why I keep a short list of reminders before I write. When my data agrees with the popular read, I write normally. When my data contradicts the popular read, I check twice. When my data is nothing at all, I shut the laptop.

What I did in those forty minutes

I want to be concrete, because generic advice is useless.

In the first ten minutes I confirmed the empty file was genuine and not a path error. I checked the root directory, checked server logs, checked whether another version of the same file had been created that day. Nothing.

In the next ten minutes I verified that I had no alternative source. No press conference I attended, no footage I watched live, no match notes in my personal log. Had I had match notes, I would have had the right to write, because match notes are direct observation, and direct observation is a legitimate root for any analysis.

In the third ten minutes I wrote on paper what I knew for certain and what I did not. The first list had exactly two lines: a file was created for a round, and the file contained no data. The second list was much longer, and covered everything a normal analysis needs.

In the final ten minutes I drafted a message to the section editor: no data, cannot write, need a new packet, otherwise please pull the piece from the weekly plan. I sent it at 12:22 a.m.

I recount this not because it is interesting. I recount it because it shows that stopping is not a professional impulse but a process. It has steps, an order, and it can be executed even when you are tired and under pressure.

There is one thing I have learned across years of working with data models, true of basketball, football and any sport with enough data to analyse: a system is only as good as its input, and no model can rescue an empty input. You can build a perfect model with hundreds of variables. Hand it an empty table and it returns a meaningless result — and it will not raise an error.

Risks and gaps

No article of mine ends without this section, and neither does this one.

The biggest risk in the story just told is not that I failed to write. The biggest risk is that I might have written. A piece built on an empty packet would have had full structure: opening, context, analysis, contrarian angle, conclusion. It would have read smoothly. It would have had numbers, player names and tactical judgement. And it would have been wrong at every layer, starting from the lowest, without any reader able to detect it. This is the hardest risk to see in my trade, because a defective product looks exactly like a sound one.

The second risk is source opacity. When a packet carries no source, no date and no outlet, nobody can assess its reliability, including me. The correct response is not to lower the standard but to raise it: require source, date and timestamp as mandatory fields.

The third risk is sport ambiguity. When the sport label says only "basketball" without specifying the competition, the analyst loses the ability to apply the right rule system, the right pace benchmark and the right landscape frame. Technically small, consequential in practice, because any cross-league comparison needs a conversion baseline.

And the largest gap, honestly, is the gap in self-checking. Every writer is the sole judge of his own source quality. On a busy night, without a mandatory process, that standard drifts downward — not because you want it to, but because you need a piece before 6 a.m.

I keep this section in the article not to congratulate myself for stopping, but because I believe readers have a right to know what an analysis stands on. If I offer a conclusion, I want you to know what I left behind to reach it.

Signals for the next round

Three months after that night, I rebuilt the extraction pipeline with two changes. First: every data packet must carry a source, a date and an outlet; otherwise the file is marked invalid and routed to manual handling. Second: every packet must name at least one concrete entity — a team, a player, a coach, or a competition organiser. If a file names none, it cannot become input for any analysis.

Both changes sound small. They cut roughly twenty percent of the system's input volume, and in exchange, the share of articles needing post-publication correction fell to a third of its previous level. In the work of a data sportswriter, that is a trade I will take.

Looking ahead, I see three signals worth tracking through the rest of this big season.

First, the quality of public data on domestic competitions. If basketball and football leagues in Vietnam keep publishing metrics at their current level of detail, domestic analytical capacity will depend mainly on direct observation rather than spreadsheets. That is not a bad thing, but it demands that writers be present, take notes, and accept they cannot cover every game on a given day.

Second, the ratio of data-backed analysis to speculative analysis during the transfer window. Whenever a major contract is announced, people tend to forget that its true value can only be assessed after the player has played enough top-level matches. A price not confirmed by a performance sample is a hypothesis, not a contract.

Third, and the one I care about most, whether newsrooms build source-verification processes for analytical content. In eighteen years I have never sat in a single meeting in Vietnam devoted specifically to verifying sports data sources. I have sat in many about headlines, cover images and publishing times. If that changes, the quality of domestic sports analysis will change with it, with no need for new equipment or new staff.

I do not believe in hunches. But I believe in what a hunch confirms when data backs it. On the night of November 12, both were silent: the data empty, the hunch with nothing to hold on to. So I did the only remaining correct thing — closed the laptop, sent the message pulling the piece, and went to bed.

Some nights, the highest value a data writer can deliver is an article that never gets born. I know that is a hard sell to a newsroom. But numbers never need us to defend them. It is the other way around: we need them so we do not deceive ourselves.