The Empty Report: Nine Analysis Sections, Not a Single Number

At 7:42 p.m., four minutes left in the third quarter, my screen showed a grey line: the positional feed had disconnected. No coordinates, no running speed, no record of the gap the defence left behind after each change of direction. Eighteen minutes later the post-game report still rendered in its exact frame — nine sections, full headings, tables aligned — and in every statistical cell sat an N/A, evenly spaced. That report was not wrong in a single word. It also said nothing at all. I kept the file, named it the empty report, and left it in the root folder, the place I open before writing any analysis.

Based on my experience following games in the NBA and in domestic leagues, a decent post-game assessment stands on three layers. The first is the play-by-play: who passed to whom, in which minute, where the ball went. The second is tracking data: the coordinates of ten players and the ball, twenty-five times per second. The third is context — schedule, injuries, who had just played thirty-eight minutes forty-eight hours earlier. In the 2026-14 season the NBA installed SportVU, a six-camera system, in all 30 arenas. By 2026 Second Spectrum had taken over the modelling layer. Since then the second layer has stopped being a luxury.

When the second layer dies, I can still write. When the first layer dies, everything behind it is only form.

In a newsroom the pressure does not come from missing data. It comes from the piece having to be live before 11 p.m. A deadline does not permit an empty section, so the empty section is filled with a harmless sentence. Then a second harmless sentence. Then the frame stands on its own without anyone holding it. Twenty-one years of watching this industry have taught me exactly one rule: whatever is easy to fill will be filled, whether or not it means anything.

The file I received that night had ten fields. Title: N/A. Source: N/A. Article type: unclassified. Number of information points: empty. Entities involved: empty — no team, no player, no coach. Time sensitivity: not assessed. Source quality: no source field existed to assess. The integrity check closed with an error code, INPUT-DEFECT-001, defective input, no analysable content.

What made me stop was the part that came after.

The system went on to render nine analytical dimensions: tactics and technique, player data, operations and salary cap, league landscape, rules and governance, coaching staff and locker room, risk, media narrative, industry ripple. Nine dimensions, a table for each. Every cell in every table was filled. All of it was N/A.

Not one section was left blank. Not one heading was cut. A perfect frame, matching bracket for bracket, in the right order of priority, with a glossary of terms at the end. Anyone opening that file and skimming the table of contents would see a professional analysis. They would not see that inside there was not a single event.

A report can be flawless in form and empty in content, and the reader's eye always grades form first.

What is worth noting is that the report did not hide anything. It listed its ten empty fields at the very top. But it was built for the reader to skip the top: bold section headings, neatly ruled tables, numbered conclusions. Visual hierarchy inside a document is itself an argument — whatever is large is what matters. And the largest thing in that file was an empty frame.

A real analysis contains at least one named entity and one link of evidence connecting a conclusion to an event. That file had neither.

The same week I went back to an old piece on Nikola Jokić. He was taken with the 41st pick of the 2026 NBA draft, at a point when every basic box-score column showed nothing worth noticing. What saw his value was not a scout's instinct but the second layer of data: the pass that creates space before the ball leaves the hand, the release angle, the speed of decision. Three MVP awards came later. An empty frame cannot do that work. It only pretends it did.

One small detail strikes me as the most expensive thing in the whole file: every conclusion carried a confidence tag — High, Medium, Low. The tag appeared even when there was no number to anchor it. A confidence tag without an anchor is not caution; it is an alibi, telling the reader the author weighed the risk when the author was only decorating.

The most honest line in the file sits at the top, where the check states that it cannot proceed to analysis. Everything after that line is performance.

Numbers never need us to defend them. We need them so we do not fool ourselves. That night, no number needed defending. Only a frame needed filling.

My first reaction was to distrust the frame. If a system can generate nine analytical dimensions out of nothing, the problem is the system, and the correct handling is to throw it away.

That conclusion is lazy.

What collapsed that night sat in the extraction layer, not in the data method. The extraction layer is where an article is read, events are pulled out, sources are tagged. The basketball domain label survived, which means the classifier read the headline or the section header and stopped there; the body never made it through the door. Such a system is not wrong because it trusts numbers. It is wrong because it has no mechanism to stop when there are no numbers.

The largest risk in that file sat on the reader's side: a glance of the eye and a belief already formed.

I have stood on the other side of this error and paid for it. In 2026, when the Bundesliga returned to empty stadiums, I bet that the home-win rate would fall from 54% to below 50%. It fell to 48.7%; Borussia Dortmund won only 3 of their remaining 8 home games. The number arrived exactly as predicted. Then my recovery model failed, because I had no variable for training-ground quality or for a squad's psychology. When the stands went empty, my model collapsed. I knew I had forgotten the human factor.

Two years later, at the 2026 World Cup, I built a group-stage model on cumulative xG and control indices and was confident Germany would advance. Germany went out. Looking back, the dataset I had gathered before the tournament contained no PPDA figure for Japan, who posted 6.8 across the two matches against Germany and Spain. That number was not outside reality. It was only outside my dataset.

Correlation is not causation, and a model with no variable for the people in the stands will always mispredict on the very night the stands are empty. I do not believe in hunches. But I believe in what a hunch turns out to be confirmed by data.

From this season, every report I file carries a tenth field, placed after the other nine: the gap. It lists what I do not have — injury data, the opponent's pressing numbers, the real minutes of the second unit over the past four weeks. It does not make the piece weaker. It is the only part of the piece I know for certain is correct.

The signal for the next cycle does not lie in a better model. It lies in labelling what we do not know as clearly as we label what we do. The empty report is still in my root folder, and I open it not to remember a night the feed died, but to remind myself that the most beautiful frame is the easiest one to be fooled by.