Nine Analytical Dimensions and an Empty Dataset: Notes from Berlin on the Limits of Every Model

2:40 a.m., August 13, 2026. Berlin has had an unusually cold summer this year — 14 degrees, light rain drumming on the roof in Kreuzberg. On my screen I have nine tabs open, exactly the nine I use for every deep-dive: patch and meta, tournament format, roster and players, regional landscape, club finance, rules and governance, risk profile, public narrative, industry transmission. Nine frames. Nine input fields. Not a single line of data.

In my inbox there is a four-line attachment. No tournament name. No patch number. No team name. No player name. No date. Four lines of prose describing something "very notable" and asking for my opinion before 9 a.m.

It took me forty minutes to write a single reply: the framework is built, the input is empty, every conclusion drawn from here would be fabrication. I sent it. The response came half an hour later, in a familiar voice: "Just estimate it from feeling."

I have worked in this trade for sixteen years, five of them sitting in Berlin reading transfer contracts and pricing players. If there is one lesson I paid for with real money, it is this: the most dangerous moment for an analyst is not when the data contradicts him. It is when the data is absent and he still wants to speak.

Numbers never lie — only the reader's heart turns them into lies.

Nine tabs open, and the price of filling a blank

To make clear what I am talking about, I should explain how those nine tabs work when they actually have data.

The first tab is patch and meta. In esports this is the foundation. A patch changes a damage coefficient, changes a cooldown, changes the economic value of a map zone, and the entire priority order of a tournament shifts with it. What I do is build a three-column table: characters or champions who benefit, sides that lose out, and the win-rate gap between the old patch and the new one. Without a patch number, the third column is empty. A two-column table is not analysis; it is a list.

The second tab is tournament format. Format determines the probability of an upset. A single-game series is a different animal from a five-game series. Schedule density determines which team enters the knockout rounds with legs two kilograms heavier. When the client does not name the tournament, I cannot say anything about upsets or stability.

The third tab is roster and players. The fourth is regional landscape. The fifth is finance. The sixth is rules and governance. The seventh is risk profile. The eighth is public narrative. The ninth is industry transmission — from publisher down to streaming platforms, down to sponsorship, down to derivative markets.

Those nine tabs are not ceremony. They are a prevention system. Each tab closes a door that an inexperienced writer walks through: the door of psychological speculation, the door of personal blame, the door of sanctifying a single win, the door of turning correlation into causation.

I learned that in a Berlin newsroom where the editor does not ask "is this piece good" but "where did this number come from, on what date, who published it, and did you check a second source." At first that standard drove me mad. Later it became my spine.

The problem is that the standard only works when there is a source. And in this trade, the empty-source scenario happens more often than outsiders imagine. A coach is sacked, a team is sold, a young talent is pushed onto the front pages — the news arrives first, the data arrives later, sometimes three weeks later. In that gap, everyone wants you to speak.

Empty is not zero

This is the technical part, and also the most misunderstood one.

In data analysis, "no data" and "data equal to zero" are entirely different things. When I record a team's early-fight win rate as 0%, I am saying something very strong: they fought, and they lost every time. When I leave that cell blank, I am saying something modest: I do not know yet.

Confusing these two states produces almost every bad analysis I have read in ten years. The writer sees a blank cell, and instead of leaving it blank, fills it with the cheapest available material: intuition, feeling, story.

With an empty dataset there are three possibilities, and they are not equivalent.

The first is data missing at random. You have the data, you simply have not collected it. The handling is obvious: go get it. Collect more, wait, or pay a source. This state is a matter of time, not of method.

The second is data missing systematically. It does not exist because nobody measures it. High-speed running distance in youth competitions, for instance, is not fully recorded. Here the analyst must acknowledge the limit and lower confidence in the conclusion. There is no other way.

The third, and the most dangerous, is data missing because the absence itself carries information. A club that does not publish medical data on a key player's injury, three weeks before a play-off, is telling you something through its silence. A club that does not publish its compensation structure after appointing a new sporting director is also telling you something.

This is where my language has to change. Instead of writing "this team will certainly lose its key player," I write "the probability of the key player's availability sits in a range I cannot verify, but eleven days of silence is itself a data point."

Every crisis is unlabeled data.

The hard part is not labeling it. The hard part is accepting that some datasets will remain empty for the entire duration of your analysis. And your job is to state that clearly in the final product, not to cover it with a smooth sentence.

Four times I nearly filled a blank

I will recount four. Not to boast, but because all four relate directly to a framework displaying "insufficient information" while the writer refused to leave it that way.

The first: Hannover 96

At 23, fresh out of a Berlin lecture hall, I took a content job at a sports data startup. My first assignment concerned the 2026-18 Bundesliga relegation race and the story around André Breitenreiter at Hannover 96. I rebuilt the sequence of matches using expected goals, and what I found ran against the general feeling: the quality of chances Hannover created and the quality of chances they allowed were not as far apart as the table suggested. Their position had been dragged down by a run of results that xG did not support.

I wrote that changing the coach at that moment rested more on feeling than on data. The editorial desk called me naive. Hannover 96 then took 11 points in the final five matchdays and survived — a figure I can still verify in the archived Bundesliga table.

I retell this not because I was right. I retell it because I nearly erred in the opposite direction: I nearly used a single metric as a verdict. Had I only had xG that day, without the squad list, without the fixture list, without fitness data, I would have written a piece with the right conclusion and the wrong method. A right conclusion from a wrong method is the most dangerous kind of rightness, because it teaches you to trust reflexes instead of process.

The second: Germany at the 2026 World Cup

A year later I looked at Germany's PPDA — the number of passes a team allows the opponent before each of its own defensive actions. Their figure sat outside the historical band I had logged for them. I paired it with contested-zone data and the rate at which they recovered the ball after losing it.

The picture was not "Germany are weak." The picture was "Germany have lost the ability to convert pressure into possession." They pressed in volume, but their recovery rate within five seconds of that pressure was abnormally low. I wrote that Germany risked elimination in the group stage, that a patient and disciplined opponent could make them pay. On June 27, 2026, Germany lost 0-2 to South Korea and went out.

The whole newsroom called me a "data prophet." I hated the label. It encourages people to believe I predicted the future. I did not predict. I merely refused to fill a blank with feeling, and in that case the blank was filled at the right moment by real data.

Based on my experience following Bundesliga matches in that period, I drew a rule I still use today: a pressing metric unaccompanied by a ball-recovery metric says nothing. It only says the players are running a lot.

The third: the summer of empty stands

In 2026, when the season froze because of the pandemic, I sat down and watched all 263 Bundesliga matches of 2026-20. I counted home win rates and found they fell from roughly 46% to roughly 29% when played without spectators. Union Berlin, the club tied to the fan wall at their stadium, lost around 61% of its points compared with when fans were present.

That number first pushed my reflex toward psychology: this team loses its nerve without a crowd. I stopped. To say that, I needed in-match behavioural data to back me: skill-error rates, aimless long passes, how far the team's shape stretched by minute. I went looking and found partial support, but most of the variance lay elsewhere: home teams lost advantages tied to the tempo controlled by referees, and advantages tied to added time.

From that dataset I built the Decay Coefficient — a way to measure how vulnerable each team is when off-field conditions change: crowds, schedule, patch, rest periods. It is a model, not a truth. I stress that word.

In the summer of empty stands, I hear data falling drop by drop.

A transfer consultancy in Berlin bought the 40-page report and invited me to become a transfer market administrator. That was the turn that made me a valuer rather than only a writer.

The fourth: EURO 2026 and 1,400 data points

In 2026, aged 30, I was leading an analysis team. A Bundesliga club asked me to value three transfer targets. The first was a player who had exploded at a major continental tournament, who had played only six matches at the top level and scored in three consecutive games. The second was a striker in the French top flight averaging 0.52 expected goals per match across three consecutive seasons. The third was a defender returning from a long-term injury.

Six matches is one sample. Three seasons is another. I told the club that a six-match sample lacks the variance to separate talent from luck, and that in my data the confidence interval for the first target was wide enough to encompass the possibility that he was a mid-table player.

I built a regression model on roughly 1,400 data points and chose the second target. The decision was judged boring. Three months later the tournament breakout suffered an injury, the defender's form collapsed, and the chosen striker scored 14 goals. I wrote about how we rejected a star with 1,400 data points.

But I must be honest about one thing: that time I also nearly filled a blank. My model had no detailed medical data on the third target's previous injury. That cell was empty. Had I filled it with a guess, I could have delivered a stronger, more attractive and more probably wrong conclusion.

A transfer is not buying a person, it is buying a probability distribution.

The framework is also a hype machine

Here I need to speak plainly about the prettiest paradox in this trade.

People assume a framework returning "insufficient information" is a weak framework. The opposite is true. A weak framework always returns an answer, whatever the input. Nine tabs, all empty, in which seven still produce a conclusion — those nine tabs are decoration.

I have reviewed a great many transfer reports over five years. The sad rate is that most of them describe the relationship between two variables in causal language while the data only permit talk of correlation. A team spends more and climbs the table. A player changes clubs and improves his numbers. A new coach arrives and the team wins four straight. All three can be true, and all three can be coincidence.

The issue is not that people write incorrectly. The issue is that causal conclusions sell better than correlational ones. The sentence "this team wins because of the new coach" spreads many times further than "across those four matches, their opponents had defensive metrics below the league average and the sample is insufficient to separate the two effects." The hype machine is not on the reader's side. It is on the writer's side, in the instinct to be recognised as someone who understands.

I call that instinct white cheating. No number is altered. No source is invented. The writer merely selects blanks to fill with what he wants to see, then presents the result as a discovery.

There is another layer I rarely write about, but will here: sports data sold directly to betting companies is the ugliest side effect of the digitalisation of sport. A metric created to describe chance quality is repackaged into odds before the audience has understood what it measures. In that chain, the honest analyst carries a specific responsibility: never sell a model as though it were a prophecy.

In esports the pressure is greater. The pace of data collection here is many times faster than in football. Every patch creates a new dataset, every round creates a new sample, and every small sample can be packaged into a large story. Professionalisation turns players into assembly-line products and simultaneously turns every one of their metrics into tradable goods before the player himself knows the metric exists.

I do not believe in intuition — I believe in the decay coefficient of intuition.

Four cases where an empty dataset held the answer

Since moving into valuation, I have noticed something under-written: in a meaningful share of transfer questions, data is missing not because people are lazy. It is missing because people have reasons not to publish it.

When a key player's contract has ten months left and the club is not negotiating an extension, I log the date the silence began. The gap between that date and the day the window opens is a variable. In the data I have assembled, sudden departures are usually announced by that very silence.

When a team changes its coaching staff mid-season, team-level analysis becomes meaningless for the first three to four rounds. Ball-progression structure changes, midfield starting positions change, defensive line height changes. I still see people citing full-season data to conclude something about a team that changed coach two weeks ago. That is a segmentation error, and it is surprisingly common.

When a young player is promoted to the first team after one good match, his top-level sample may be 90 minutes. No model, including the best I have ever built, can separate signal from luck in 90 minutes against the pressure of an elite defence.

And the fourth case, the one I encounter most: a club does not publish medical data. Every transfer model has a black hole here. You are valuing a player without knowing what his knee cartilage looks like. The only approach I know is to lower the expected value, widen the variance, and state clearly to the client that there is a variable I cannot observe.

There are matches that end when the referee blows the whistle — and there are matches that only begin when the data speaks.

What an analyst must say when there is nothing to say

When a client sends you an empty dataset and demands a conclusion, there are four things to do in exactly that order, and the order cannot be reversed.

The first is to describe the input in neutral language. No judgement, no speculation, no adjectives. "The document is four lines, names no tournament, no date, no subject."

The second is to list what can be inferred from that input, however little. Even if the result is three lines, they are still three real lines.

The third is to list what is strictly required to answer the question. Specifically enough that someone reading it can go and fetch it for you. Tournament name. Patch number. Publication date. Publication source. For transfers: current contract length, fee structure, release clause, injury history.

The fourth is to set a confidence level and a probability range for every conclusion, even very small ones. No probability range, no conclusion.

Those four steps sound simple. The difficulty is that they run against every writerly instinct. A journalist is judged by published work, not by work he refused to publish. An analyst is judged by the number of conclusions delivered, not the number withheld.

But if you look closely, every large wall in this industry was built from exactly one brick: one person, at one moment, deciding not to say what he did not know.

I still remember a morning in Berlin when I sent a client a twelve-page report and nine of those pages presented what I did not know. The client called back and said it was the most transparent report he had ever received. I did not ask what he did with it. I only logged that day as a data point about myself: stating a limit does not reduce the value of a product. It increases it, for exactly the kind of user who can use it.

The same holds at a larger scale. Esports leagues are entering a phase in which data exceeds the capacity to understand it. I call this the resolution imbalance. People record everything and understand very little. In that phase, the advantage does not belong to whoever collects the most. It belongs to whoever can identify precisely what he is missing.

Signals for the next cycle

Three signals I am tracking for the coming cycle.

The first: I am counting the days between a patch being published and a team publicly changing its roster structure because of it. In the major leagues, that gap is narrowing. It is a better measure of a team's analytics department than any coach interview.

The second: I am watching leagues begin to publish positional data at a level of detail previously held by only a few large platforms. Each time a league publishes another data layer, a group that built a competitive edge on it loses that edge within eighteen months. That is a law every data analyst should internalise.

The third, and the one I await most: I am counting the number of clubs that publish confidence intervals alongside statements about a signed player's metrics. Today that number is close to zero. When it starts to rise, this industry will have grown up.

As for the night of August 13, 2026, I closed the nine tabs and went to sleep at 3:20. The next morning I sent the client a single message: with a tournament name and a patch number, I can answer within two hours. Without them, the most honest answer is the one I already sent.

If you have read this far and find those two answers different in value, it may be worth reading again from the beginning. Because this entire trade rests on whether you dare to leave a cell blank.