The Empty Spreadsheet and the Price of Conclusions Without Evidence

One Friday night in Shanghai, I opened an analysis file the newsroom had sent over, with a short message attached: approve it, please — it goes up tomorrow morning. The file looked good. Nine sections, each with tables, subheadings, and a confidence rating cell.

I scrolled down and read carefully. The information points field was blank. The entities involved field was blank. The source headline read N/A. The source read N/A. The article type read unclassified. Time sensitivity read not assessed.

The only living thing in the entire file was a single label line: football.

I sat still for a few minutes. Not because the file was hard to understand. Because it was too easy to understand in another sense: a perfect skeleton with no flesh. Anyone skimming would see sections, charts, and assume someone had done the work. That is the worst kind of error in this trade, and it has nothing to do with football until someone decides to publish it.

Football is a business of empty frameworks, and people still pour money into them every week.

I have worked this trade since my years in Vietnam, then moved to China to work for sports platforms and betting companies. After many years of watching number tables scroll across a screen, I learned something that sounds paradoxical: the most dangerous thing is not a wrong metric, but a metric that does not exist yet is treated as if it had been verified. A model short on data still prints an output. A table short on sample still draws a trend line. Software has no shame. It only knows how to export.

In 2026 I was a senior analyst for a newly launched sports platform in China. Before round 18 of the Chinese Super League, for the match between Shanghai SIPG and Shandong Luneng, I published an analysis built on xG: SIPG had generated 2.8 expected goals, their opponent 0.4. I predicted 3-1. The traditional pundits picked a draw, because they were looking at form tables and head-to-head history. The final score was 3-1. The piece hit 50,000 views in 24 hours.

What I remember is not the 50,000 views. I remember sitting there checking every situation again to confirm that the 2.8 had been built from real shots, in real positions, under real pressure. If I had entered one shot location wrongly that day, the model would still have produced an output. Still 2.8. Still a 3-1 prediction. And the analysis file would still have looked flawless.

That is what an empty data file always reminds me: the framework was never the evidence. Nine sections, twelve tables, three confidence levels — all of it can be decoration.

In the summer of 2026, a betting company hired me as lead analyst for the World Cup. My model combined two variables: PPDA — the index measuring an opponent's pressing intensity — and average defensive height. On 27 June 2026 it predicted South Korea to beat Germany 2-0. I posted it on social media and told people to follow it. The result was 2-0. I thought I had found the formula.

Then came the round of 16. The model said Brazil would beat Belgium, based on better defensive quality and better defensive xG. I said so on live television. On 6 July 2026, Brazil lost 1-2. Clients lost money because they listened to me. I argued bitterly with a colleague online, then spent three weeks rewriting the code, adding competition variables and a noise coefficient.

Every model is wrong, but a few are wrong in a useful way. My 2026 World Cup model was wrong in a useful way, because it forced me to look at where it had no data: knockout football is a different game psychologically, and psychology does not live inside PPDA.

Since then, every piece I write carries a fixed warning line: the model is a probability, not a prophecy. I write it for myself rather than the reader, to remind myself that between a table that looks complete and a conclusion that holds up, there is always a gap the analyst has to bridge alone.

Back to that Friday file. The worrying part was not the nine empty sections. It was that an automated process had run across the whole document, reached the most important field, and returned exactly one phrase: insufficient information. And at the next layer, every conclusion was generated from that phrase rather than from football.

In the analytics industry this is called null propagation. A missing value upstream does not disappear. It flows downstream, puts on a new field's clothing, and surfaces as "no risk detected". That is the moment the trade fools itself without anyone pushing it.

I have seen that exact mechanism in football many times, only nobody writes it into the table.

A club goes quiet for two weeks. The press reads it as a stable dressing room. In reality the reporters may have had their accreditation withheld, or the club is keeping silent over a serious injury. No news does not mean nothing happened.

A player has no metrics for three matchdays. People note a dip in form. But if he is out with an injury and the club only discloses the part that protects its share price, then the empty cell reflects a communications habit rather than the player's legs.

A transfer market story with no source, no agent's name, no date anchor. Readers will still build a complete narrative in their heads. The human brain cannot tolerate a gap. It fills it with the most familiar assumption available.

Data going missing is not the loss of data — it is a type of data. The gap has its own content, and that content is usually a process defect rather than a football event.

On injuries, this is the largest blind spot. No system discloses player fitness in full. Clubs disclose when disclosure pays: reassuring sponsors, cooling public opinion, or pushing a player's value down before renewal talks. When a name vanishes from a matchday squad with no announcement, the reader receives an empty cell. And that cell gets filled with guesswork, usually the worst guesswork.

On youth development, the mechanism is even clearer. Academy projects named after former stars appear on schedule, with photographs and press releases. Most stop there. What is genuinely scarce is a corps of grassroots coaches trained systematically, with curricula and periodic evaluation — work that produces no headlines but produces players.

After years living between two football cultures, I see something hard to put into words: the same metric can carry two different meanings in two places. In China, player data is sold as investment evidence. In Vietnam, data is read as commentary. The same xG line, used on one side to price a player and on the other to start an argument. An analyst has to know which frame of reference they stand in, or they will apply one standard to the other standard's data.

This is the point I consider most important, and also the most easily misread. When a field reads "cannot be analysed due to insufficient data", a mechanical reader translates it as "no risk". Those two sentences are worlds apart. One is a conclusion. The other is an admission that there is no basis for a conclusion.

Over my years in this trade, I have learned that the biggest analytics disasters do not come from overconfident models. They come from models treated as finished that were never actually loaded with data. Nobody checks the gate, because the gate is not glamorous. The results table is glamorous.

Every spreadsheet is a meditation session, except that when you finish meditating you have lost money. I say this to young people entering sports analytics as a reminder: the hardest part of the job is not building charts. The hardest part is daring to say there is nothing to analyse today, and surviving the fact that your boss does not want to hear it.

In the Vietnamese market, where millions read football news every day, the data gap is more dangerous still. Publishing speed is high, secondary sources are dense, and the habit of re-translating from aggregator sites means one wrong detail can replicate across dozens of outlets before anyone verifies it. The odds of being caught are lower than the speed of transmission.

Here is the paradox: the more data gets produced, the less able readers are to tell real data from data presented as real. A file like that Friday one does not need to lie. It only needs to stay silent in the right place.

xG does not score goals, but it makes people argue more than the ball itself does. I still use xG. I still use PPDA. They are useful witnesses, as long as I remember that witnesses need to be cross-examined, not worshipped.

So where does this trade go?

I think about a role that has no official title in sports newsrooms yet: the data gatekeeper. Not an editor, not an engineer, not an analyst. This person has one job — to block files that lack sufficient data before they become articles. This person is paid to say no.

For models, I would add a mandatory inspection layer: every conclusion must trace back to at least one specific, verifiable information point. No point means no conclusion, however beautiful the framework. For readers, I would speak more plainly about the limits of each piece.

And for myself, I keep an old habit: every time I write the word random, I ask which intervening variables I have ruled out. If I have ruled out none, I am not allowed to use the word.

That Friday file, I did not approve. I sent it back with one line: correct label, empty content, do not publish. The next morning the newsroom re-ran the extraction process. By noon the new file had a title, a source, seven information points, and a specific match to talk about.

Football is still there, waiting to be written about properly. My job is to stop writing about it on a blank page.