An Empty Data Table Is Not a Clean Bill of Health

2:47 a.m., Monday, Brisbane. I opened a nine-tab spreadsheet a colleague had sent with a one-line note: “No risks detected.” Nine tabs. Not a single cell held a number. Not a single team name, player name, or competition name. The header row sat untouched at the top, bare as an empty frame.

I sat still for about forty minutes. Not out of anger. I had grown used to nights like this back when I was a mid-level analyst for a football site in Brisbane, in the years when editors struck out my numbers because “readers won’t understand them.” What kept me still was the phrasing of the note: an empty file presented as a safe conclusion.

A decent analytical pipeline runs in two stages. The first extracts from the source text: title, publisher, date, information points, entity list. The second builds nine analytical dimensions on top of what the first stage returned. When the first stage hands back a blank page, the second still runs — and it runs exactly as programmed. Every field falls to a null value, every risk item reads “insufficient information,” and the final report looks tidy, professional, and harmless. A hurried reader closes it and moves on.

A null value is not a zero, and a zero is not a conclusion. Those three things live on three different layers of the same spreadsheet, and my profession exists precisely because they get confused. When an xG model returns 0.00 for a shot, that is a finding about chance quality. When the model returns a blank cell, that is a finding about the recording system. A coach reads both as one sentence — “this player creates nothing” — and that is the moment a personnel decision gets made on the back of a technical fault.

The summer of 2026 taught me this the expensive way. Competitions froze, stadiums emptied, event-data providers nearly stopped transmitting. I lost two freelance contracts in three weeks. With no new matches, my spreadsheets stayed blank for months. An empty summer taught me this: with no match to watch, memory still shoots from distance. But it also taught me that a blank table can lead people to wrong conclusions about an entire football culture, if they cannot tell “nothing to measure” apart from “nothing to worry about.”

On the evening of 7 May 2026, I stayed back after work to re-measure Liverpool against Barcelona at Anfield. Andrew Robertson covered 12.4 km, 2.1 km of it at sprint pace — numbers I punched out of the footage myself because the broadcaster’s tracking sheet was missing two segments that night. Suppose the tracking system had failed completely. The spreadsheet would read 0 km. And someone would have written that the Scottish left-back stood still in a match where he was among the hardest runners on the pitch. The collective memory of that Anfield night stays correct; only the record lies. That is the most dangerous kind of error, because it makes no noise.

On 30 June 2026 in Kazan, Kylian Mbappe accelerated at around 37.6 km/h in the move that led to France’s goal against Argentina. I stayed up two nights stepping through frame by frame to understand that run, and then realised that if the speed gun dropped three frames, the number would vanish while the beauty would remain intact. A goal is a moment, xG is fate, and I choose to record both. But when frames drop, I am obliged to record what survives and to mark clearly what was lost — rather than plugging the gap with a number that sounds plausible.

In the A-League, I was called a rebel simply because I carried a laptop. In the 2026-18 season I re-watched 19 Melbourne City match tapes by hand to rebuild a shot-quality table, because the raw data feed I had bought was missing a set of situations outright. That table was not pretty. It contained cells where I wrote “unknown” instead of filling in a guess. Every number has a story, and my job is not to ruin it — and the fastest way to ruin it is to turn a blank cell into a round zero so the table looks full.

In esports the fault is harder to spot. I once reviewed a scrim block for a team in Australia’s domestic scene where the server-side log had been lost. The KDA column in the weekly report fell to zero. Three weeks later a mid laner was moved down to the academy roster on grounds of “declining form.” Nobody checked whether that match had ever been recorded. The same applies to PPDA: the metric depends entirely on the event feed, and when the feed breaks, the metric collapses to null rather than to weak. A pressing sequence gets read as a passive one simply because the data provider switched servers.

I still remember a press conference in Australia’s domestic league where a head coach was asked why his side had eased off its pressing. He answered with a feeling. The team’s analyst answered with a metric that had collapsed because the feed had broken. Both were wrong, but only one of them faced a challenge. Since then I annotate the source of every column, including the ones I punch by hand, so that whoever reopens the file can tell measurement from inference.

The irony is that the industry rewards decisiveness. A report brave enough to say “I cannot conclude” gets filed under indecision, while a report reading “no risks detected” gets filed under tidy. The silence of data gets misread as the calm of reality, and the confusion spreads layer by layer: a blank compliance checklist reads as “no violations,” a blank injury list reads as “a healthy squad,” a blank press-transcript reads as “no internal friction.” In all three cases, what is missing is evidence, and what gets extracted is a conclusion.

The correct report in this situation carries a large line at the top of the page: insufficient data, do not distribute. It is not handsome, but it is honest. A label like that protects the analyst and the decision-maker alike, because it blocks the exact moment when a blank cell is about to be translated into a fact.

There is one more layer I want to name, because it is a blind spot among data people themselves. We are taught that correlation is not causation, and we test that by comparing two columns of numbers. But when the correlation matrix is empty, we default to assuming the variables are independent. A blank table proves neither independence nor risk. It proves only that nobody has measured yet. At 39, I have learned that data also hurts when it is distorted — and it hurts most when distorted by absence, because nobody sees the bruise.

So the work to be done is not in the analysis stage. It is at the door. A decent system must reject inputs carrying zero information points, must fail loudly instead of proceeding with nulls, must force the operator to see the blank cell before it becomes a sentence of conclusion. Re-running the extraction on the original document is the cheapest action and the only one worth taking right now; the nine dimensions are already built, waiting for one game title, one name, and one date before they can start speaking.

The next morning I kept that nine-tab file in its own folder and labelled it “negative control.” Every time a new spreadsheet opens and looks too clean, I open it again. I want that hunger for data to outlast the false comfort a blank page provides. Next cycle, when I open a file and see nine empty tabs, I want the system to shout before I have time to nod.