Blank Cells Are Not Alibis: Lessons From an Unfilled Sports Analytics Sheet

On the night of June 14, 2026, I sat in Los Angeles with two screens open. On the left was the Euro opener; on the right, the corner-kick spreadsheet I had been assigned to track for a national team. By the 70th minute I had logged fourteen set-piece situations, but three of them carried only a timestamp and no coordinates. I skipped those three blanks, compiled the numbers, and was one click away from sending it.

What stopped me was simple: if those three situations were inside the box, would my conclusion still hold? I did not know. And that “I do not know” was the most valuable content of the entire day. The first xG spreadsheet taught me: every goal has a hidden story. It took the summer of 2026 for me to understand one layer deeper — a blank cell in a dataset is not a zero; it is an unanswered question. Fill it with a zero and I have fabricated a fact with my own hands.

The most expensive mistake in sports analytics is rarely a wrong calculation. It is a blank cell read as a zero. And in most of the workflows I have passed through, nobody is assigned the specific task of checking for that.

In August 2026, I spent an evening rereading a second-stage deep analysis of the esports domain. The report carried all nine standard sections: patch analysis, tournament system, rosters and players, regional landscape, club finance, rules compliance, risk profile, public narrative, and industry transmission. Every section was laid out neatly, tables and cells complete. But the entire content inside collapsed into a single sentence: insufficient information to assess.

What is worth noting is that the report was still useful, in a way its author probably did not anticipate. Its strongest conclusion was a warning: never read an empty compliance checklist as a certificate of innocence. A blank field means no one has checked yet. It does not mean someone checked and found nothing.

I wrote that line into my notebook, then realised it described my own profession exactly.

Start with the most visible part. Every modern sports data platform — football, basketball or esports — runs on automated feeds. Provider A counts a shot as an attempt; provider B counts it as a misplaced pass. One system logs a duel as a tackle, another logs it as a loss of possession. When the two sources are merged into one table without a matching step, the unmatched data disappears — and it disappears silently, as blank cells scattered among thousands of fully populated rows.

That silence is the most dangerous thing in the entire spreadsheet.

In 2026, while still a middle-school student in Los Angeles, I hand-recorded shot data for all 64 matches of the World Cup in Russia. With no official xG source I could reach, I expanded my Excel sheet to more than twelve hundred shots, estimating chance quality myself from shot angle, distance and the number of defenders in front. When France lifted the trophy, the media praised a flamboyant attack led by Kylian Mbappé and Antoine Griezmann. My spreadsheet pointed elsewhere: that team won because it limited opponents to roughly 0.7 xG per match.

But there was a detail I overlooked for years. Eleven matches in my dataset had missing shot locations for one of the two teams. I still averaged, still arrived at 0.7, still concluded — and never annotated that those eleven matches had holes. If anyone reopens that sheet today, they will find no trace of the gap. It was swallowed into an average that looks perfectly tidy.

When home is no longer home, I am forced to rewrite every assumption.

In 2026, when the pandemic halted competitions, I compiled data from more than three thousand matches across Europe's five major leagues to build a home-advantage model. The result showed home teams were effectively “gifted” an average of 0.38 goals per match. When the Bundesliga restarted behind closed doors, I published a prediction: home win rates would fall sharply. The first three matchdays confirmed the model.

But that time I repeated exactly the same old mistake. Of the three-thousand-match set, nearly two hundred matches lacked information on actual attendance. I dropped them from the sample instead of flagging them as undetermined. Mathematically, the exclusion was defensible. Methodologically, it produced a sample cleaner than reality — and a sample cleaner than reality always yields a model more confident than it deserves to be.

Morocco 2026: when defensive data spoke first, the world listened later.

That year I was eighteen, launching my own analytics newsletter on Substack. I extracted PPDA and defensive-line distance for all thirty-two national teams and wrote that Morocco possessed the most proactive shield in the tournament, despite a possession share near the bottom, with Sofyan Amrabat as the spine of that structure. Everyone knows the outcome: Morocco reached the semi-finals. According to match data compiled from the 2026 World Cup finals published by FIFA, the team conceded only five goals across seven matches, one of them an own goal, and kept four clean sheets — a run that ended in the penalty shootout against Spain on December 6, 2026, where Achraf Hakimi took the decisive kick.

What I rarely mention is the portion of data I could not use. In my PPDA table, Morocco's seven matches were missing extra-time metrics. Against Spain, the team defended for a full one hundred and twenty minutes, while I only had data for the first ninety. My conclusion was right, but it was right thanks to luck more than to a rigorous model. Had Morocco conceded in extra time, I would have had to explain why my “correct” model omitted the most important thirty minutes.

In the summer of 2026, aged twenty, I interned at a sports data analytics company in California. I was assigned to evaluate a target striker for a mid-table club. My model showed his actual xG sat 4.5 goals below expectation over a season — too large a gap to call a decline in form, and too large to ignore. I concluded it was bad luck, not lost form. The club signed him. He scored on the opening matchday.

Then I rechecked my own raw dataset and found fourteen matches with no shot-location data. Forty percent of that 4.5-goal gap came from those matches — matches I had implicitly treated as “the player creates nothing”, when in truth I simply had no data. Once again, a blank masquerading as a zero.

That same summer I missed a deadline on the corner-kick report because I wanted a model that was one hundred percent perfect. A colleague told me something I still remember: a model that is eighty percent right and delivered on time beats a perfect model delivered after the final whistle. He was right, but I think the two of us were talking about different things. He was talking about speed. I was wrestling with honesty.

There is a natural reflex in this profession: when a conclusion must be delivered but the data is incomplete, we tend to stay silent about the missing part and present only what we have. The report looks tidier. The reader is happier. And nobody complains, because nobody can see what was left out. In basketball, the offensive rating per hundred possessions runs into exactly the same problem: possessions that end in a referee's error are often stripped from the sample, making the metric look more stable and skewing every team comparison systematically — in favour of low-pace teams.

I have seen the consequences of that reflex in the place where it does the most damage: transferring models between disciplines. My background is football, and I once believed a good pressing model in football could be applied directly to esports. Football and esports differ on the surface, but the same data layer sits underneath — that I still believe. Yet that layer is only structurally comparable, not contextually comparable.

A pressing action in football is bounded by stamina and pitch area. A pressure action in esports is bounded by map vision and cooldown timers. The same quantity, two meanings. If I carry evaluation thresholds straight from one discipline to another without rechecking the assumptions, I am doing precisely what I just condemned: filling a blank with a value that merely looks plausible.

And here is the flip side of the Morocco story. When a data-driven prediction becomes famous, people start trusting the model more than the model deserves. I was right about Morocco. But I was right on a sample of seven matches, three of which lacked extra-time data. If I applied that same confidence to another prediction with a smaller sample, I would fail — and in that case the person who pays is not me, but the reader who trusted me.

The same logic applies to compliance checklists, risk profiles, and internal reports nobody outside ever reads. A check item with no data gets marked as no issue found. That phrase sounds very safe. It sits between “checked and clean” and “never checked at all”, and in most meetings nobody distinguishes the two. For a referee, a supervisor, an analyst — the confusion leads to the same outcome: a decision made in a state of misplaced confidence.

I watch a great many matches that use VAR. The popular argument is that technology will make controversy disappear. What I observe is the opposite. VAR moves controversy off the pitch and into the review room, and there it places controversy onto a new grey zone: the grey zone of camera angles that are insufficient, timestamps that do not align, and rule definitions that have not been harmonised. Those are exactly the blank cells, except they arrive as video instead of tables.

The signal I want to track next matchday is not in the league table. It is in the number of blanks each report dares to disclose. An analyst with enough backbone will not submit a clean spreadsheet. They will submit one whose annotations run longer than its data, accompanied by a confidence interval they are willing to be held to.

I do not predict the future by intuition; I only read the traces the data left behind. If a trace has been erased along one stretch, the most honest thing I can do is state how long that stretch is — not draw a straight line across the gap and call it a trend.