The Empty Report from Berlin: Why 'No Data' Is Never Good News

At 3:12 a.m. in Berlin, the last U-Bahn had long stopped running, and the only sound left on Rosenthaler Platz was the bell of a late-night courier's bicycle. On my second monitor, a forty-page PDF had just finished rendering: a preliminary assessment of an esports organisation a Munich investment fund was considering backing. I scrolled down. Page one, empty. Page two, empty. By page seventeen, the section headed "competitive risk" read N/A. "Roster structure" read N/A. "Player form cycle" read N/A. "Sustainability of the media narrative" read N/A. Nine analytical dimensions, not one quotable conclusion. That file was not an analysis. It was an empty box with a careful label.

My German colleague, a former transfer analyst at a second-division club, glanced over my shoulder and asked exactly one question: "So there's nothing to worry about?" I closed the laptop and took about four seconds to answer. The correct answer is no. N/A does not mean safe. N/A means not yet measured. And in my line of work, those two things are separated by exactly the distance between a club that survives relegation and a club that goes bankrupt.

My name is Hoang Hao, I am thirty-two, I live in Berlin, and I work as a transfer market administrator for a sports consultancy. For five years I have spent most of my time valuing footballers, valuing esports players, and valuing everything that sits between those two worlds. But there is one kind of valuation nobody ever taught me properly: valuing silence. When the data does not arrive, when the spreadsheet is blank, when the report comes back empty, most decision-makers quietly read it as a clean bill of health. It took me years to understand that this mistake is more subtle than any modelling error.

The two-stage pipeline, and the gap in its first stage

In our system, every analysis passes through two stages. Stage one deconstructs: it extracts the title, the source, the named entities, the list of information points, the author's core stance, the time sensitivity, and the source quality. Stage two is the professional layer: nine dimensions, from patch and meta analysis, tournament format analysis, roster and player analysis, regional landscape analysis, club finance, rules and governance compliance, risk profiling, public narrative and expectation analysis, all the way through to industry transmission.

What very few people outside the industry realise is this: no matter how strong stage two is, it is meaningless if stage one returns an empty list. All nine analytical dimensions are functions that take entities as their input variables. With no entities, no function runs at all. However elegant a regression model may be, it cannot compute the variance of a column of data that never existed.

So why does stage one return empty? In my operations log, four causes recur. The first is technical: the page is JavaScript-rendered and a static extractor sees only a skeleton. The second is a non-text source: video, image, infographic, where all the content lives in audio or in pixels rather than in paragraph tags. The third is a source locked behind a paywall. The fourth, the rarest but intellectually the most dangerous, is an article that genuinely contains nothing: a piece written only to fill a slot on a page, with no number, no name, no event.

These four kinds of emptiness are completely different, yet to a downstream reader they look identical: the same N/A, the same blank space, the same implicit conclusion that "there was nothing there". That is the first architectural flaw I want to address here. A mature sports analytics system must distinguish these four kinds of emptiness, because the remediation for each is entirely different. Technical emptiness demands re-running the extractor. Source emptiness demands switching to a non-text channel: listen again, watch again, transcribe. Paywall emptiness demands buying access. Only true emptiness permits the conclusion that the article had no content. Collapsing all four into one N/A is collapsing four different diseases into one prescription.

Hannover 96 and the first lesson about trusting numbers over feelings

I learned this principle not in a meeting room but in a real football season. At twenty-three, fresh out of a journalism and communication degree in Berlin, I took a content-writing job at a sports data startup. My first assignment was to analyse the Bundesliga relegation battle of the 2026-18 season. The club I was assigned was Hannover 96, and at that moment the board was under pressure to sack head coach Andre Breitenreiter after a run of defeats.

I sat down with the expected goals data, xG, for that entire stretch. What I found ran completely against the general feeling in the stands. Hannover 96 were creating chances of equal or better quality than their opponents in most of those defeats; the gap between xG and actual goals sat in the region any model would call noise. In other words, the team was losing for reasons that could not repeat, not because of a systemic fault. I wrote a piece opposing the sacking, published it, and was called "naive" by the editorial desk.

The season ended the way the data had already whispered: Hannover 96 took eleven points from their final five matches and survived. I do not tell this story to praise myself. I tell it because it taught me something that later became a professional pillar: when the crowd's feeling and the data say two different things, in most cases I choose to check the data three more times before I choose the feeling.

A year later, at the 2026 World Cup, I applied the same lever to the German national team. The metric I used was PPDA, the number of passes a team allows its opponent before it performs its first defensive action. The lower the PPDA, the more aggressive the press. At that tournament, Germany's PPDA was 8.7. It was a strange number: it said the team was still charging forward a great deal, but charging forward without structure behind it, and that structure was being left behind the midfield line.

I wrote that Germany would be eliminated in the group stage, and I named South Korea as the opponent that could do it. Everyone knows the result. The newsroom called me a "data prophet", a nickname I dislike, because it turns a process into a talent. That process has only two steps: choose a measurable metric, and ask it three times before believing it.

From that season on, I abandoned entirely the style of "feeling the match". Every analytical piece I write must carry at least one verifiable metric, and before filing I have to cross-check it against the StatsBomb database. To this day, every piece I submit to a partner carries a source note at the bottom. No exceptions, including the ones I know nobody will read closely.

The empty-stadium summer, and how a data gap produced an entire model

In 2026, world football froze because of COVID-19. I was twenty-six, stuck in Berlin, and I did something I would recommend to anyone who wants to work in sports analytics at least once in their life: I rewatched all 263 Bundesliga matches of the 2026-20 season.

What I found was not in the league table. The home win rate fell from 46 per cent to 29 per cent when matches were played without spectators. But that fall was not evenly distributed. Some teams lost almost nothing, and some teams lost almost everything. The club that lost the most was Union Berlin, famous for its supporter wall known as "Mauer-Kultur", the wall culture. With the stands empty, Union Berlin surrendered 61 per cent of the points it had been taking with a crowd.

I built a metric to measure that vulnerability and named it the Decay Coefficient. The Decay Coefficient does not measure which team is stronger. It measures how much of a team's performance disappears when its surrounding environment changes: no crowd, no key player, no game version, no head coach. I wrote it up as a forty-page report and sent it out like an open letter.

A transfer consultancy in Berlin bought the rights to the report outright and hired me as a transfer market administrator. That was the pivot that turned me from a pure writer into someone who prices players. But the deeper lesson lay elsewhere: a data gap, specifically the gap left by absent spectators, was the very thing that generated a new model with commercial value. The absence of one variable created another variable entirely.

Since then, I have been known in my team for one catchphrase: "Data never lies, but I have to ask it again three times." My Berlin colleagues treat it as a joke. It is not a joke.

EURO 2026, the 2026 World Cup, and the rule of converting emotion into metrics

At EURO 2026, when Christian Eriksen collapsed on the pitch, I wrote not a single word about emotion. Not because I had none, but because I believe a writer's emotion, inserted raw into a piece, only corrupts the reader's data.

Instead, I followed Denmark's next four matches. Their PPDA dropped from 11.2 to 9.8, meaning they pressed faster, earlier, harder. High-speed running distance rose 7 per cent. Those are two measurable numbers for something journalism usually calls "fighting spirit". I called it by its more precise name: cohesion after psychological trauma, measured in data.

In 2026 I carried that same lens into the World Cup and analysed the match many called the tournament's biggest shock: Saudi Arabia beating Argentina 2-1. In my piece there was no room for the word "miracle". There were three things: an offside trap that stripped Argentina of four goals, a high press that crushed the opposing midfield, and a defence willing to sit deep in the second half in exchange for counter-attacking space. That article later became an internal scouting document for a Bundesliga club.

From then on I added two fixed sections to every tactical analysis: "pressing trigger" and "sprint distance". These two sections work as a professional ethical fence. I am never allowed to use the phrase "fighting spirit" without sprint data. I am never allowed to praise a playing style as "courageous" without a PPDA figure. My readers began calling me the "data monk". I accept the nickname, because it describes a job that is mostly about re-checking your own scriptures.

EURO 2026, 1,400 data points, and turning down a star

This year I turned thirty and became head of the analysis team. A Bundesliga club asked me to value three transfer targets in the same window. Target one was a breakout star of EURO 2026, a player who appeared in only six matches but scored in three and became the most-searched name on social platforms. Target two was a Ligue 1 striker averaging 0.52 xG per match across three consecutive seasons, with no explosive season and no collapse. Target three was a defender just returning from a long-term injury.

I refused the glamour of a short tournament. I built a regression model on 1,400 data points spanning three seasons and two different tactical systems, and I chose the Ligue 1 striker. The club's scouting department judged the pick "boring". Three months later, the EURO star tore a ligament, the defender lost form and missed two months, and the striker we chose scored 14 goals over the rest of the season.

I published my most-read piece of that year: "How we turned down a World Cup star with 1,400 data points". But the point I wanted to make was not the outcome. It was this: the EURO star was a six-match sample, and the Ligue 1 striker was a ninety-match sample. In statistics those belong to two different universes, and in the transfer market they carry two prices about four times apart.

Since then I have shifted to a style I call "inverse valuation": starting from the question of why not to buy a name, rather than why to buy one. A transfer is not the purchase of a person, it is the purchase of a probability distribution. And every transfer piece of mine must contain three scenarios: optimistic, base, pessimistic. I forbid myself from using words like "blockbuster" or "super project" unless a data model has proved them.

Back to the empty PDF: four kinds of silence, four different actions

Now we can return to the story at the top. The forty-page report full of N/A that I opened at 3:12 a.m. is a perfect example of what I call "the homogenisation of silence". All nine analytical dimensions in our system have hard prerequisites. The patch and meta dimension needs a specific game title and a specific version. The tournament format dimension needs a named tournament. The roster and player dimension needs names. Without those three things, all nine dimensions return the same word: insufficient information.

But "insufficient information" gets read by the recipient as "no risk". That is a semantic transformation that happens silently inside a decision-maker's head, and it is dangerous precisely because nobody has to take responsibility for it. No model is wrong. No number is distorted. Only a blank space is read with its sign reversed.

Data never lies, only the reader's heart turns it into a lie. An empty list, placed inside a handsomely formatted document with headed tables, will be read as a confirmation. This is why I require my operations team to attach an explicit status to any file whose extraction stage failed. That status is not "no findings". That status is "blocked due to insufficient input". Those two sentences differ legally, operationally, and ethically.

There is one more subtlety worth spending words on. When an extraction file comes back empty, the only information we truly have is information about the extraction process itself. We know that the source might be a JavaScript-rendered page; that it might be video; that it might sit behind a paywall; or that it might be a genuinely empty article. These four possibilities have different prior probabilities, and distinguishing them is cheap: open the original source, check the format, check the HTTP status code, check whether description tags exist. That is ten minutes of work. Yet in many organisations it is not done, because the extraction layer is treated as a black box and no input quality gate exists.

The clean-bill-of-health paradox

In finance there is an unwritten rule anyone who has done audit work knows: the absence of evidence is not evidence of absence. In esports and professional football, that rule is violated every day.

Imagine an investment fund receiving our empty report. In it, the "unpaid wages" section has no data, the "competitive integrity" section has no data, the "contract disputes" section has no data, the "protection of minor players" section has no data. A hurried reader sees four empty boxes and concludes there is no problem in any of the four. The truth is that we never opened a single document to check.

There is a line I use in every internal training session: every crisis is unlabelled data. A wage crisis does not automatically appear as a line in a financial statement. It appears as a player changing his profile picture, a livestream cancelled without notice, a bank transfer three days late. Those signals carry no label. We have to label them, and to label them we first have to see them.

The Decay Coefficient applied to data itself

There is an application of the Decay Coefficient I have never published: applying it to the datasets themselves.

Data decays too. An xG model trained on 2026 data steadily loses accuracy as semi-automated offside technology is introduced, as teams reorganise their defending, as pitch quality and match balls change. A player rating metric built on one game version becomes biased after two major updates. I once watched an analytics department use an old dataset for fourteen months and make a transfer decision that I judged fundamentally wrong.

My approach is to assign every dataset its own decay coefficient, calculated in months since collection and weighted by the number of version changes in that period. The older a dataset relative to the update cadence of its discipline, the lower its weight. This sounds obvious, yet very few organisations do it, because doing it means admitting that most of the data they are proud of has expired.

A data store that is not refreshed periodically is not an asset — it is technical debt denominated in trust.

The contrarian angle: correlation is not causation, and glamour is not ability

Here I have to say plainly what my colleagues do not want to hear. In esports analytics today, most of the conclusions sold to clients are correlations presented as causal relationships. A player has high metrics and his team wins a lot. A team changes head coach and wins in a row. A national team changes formation and goes deep. All of these are correlations, and all can be reversed by a hidden variable.

The most common hidden variable in esports is the schedule. A team with beautiful metrics early in the season is often simply a team with a soft schedule. The second hidden variable is regional strength: a player dominating a regional qualifier may simply be competing in a region that weakened that season. The third is the game version: a roster built for one specific meta collapses within two weeks of the meta changing.

I am reserved about names that are trending online, not out of contempt, but because I know a six-match run cannot distinguish an excellent player from a lucky one. In statistics this is a sample-size problem. In media it is a question of whether it sells advertising.

"Trending" and "genuinely good" are two things that must be proved separately, with two separate datasets, over two separate time windows.

There is another contrarian angle directly tied to this piece's subject. When the data is empty, an analyst's natural instinct is to fill the gap with narrative. This is the greatest temptation and the greatest sin of the trade. Filling gaps with storytelling produces a product that is prettier, more sellable, more shareable, and more wrong. For a data monk, falsifying your own scripture is the worst possible error, because it leaves no trace to audit.

The dark side of digitisation: live data and grey zones

I have to address a subject my industry usually avoids: live data supplied to betting companies is the darkest side effect of the digitisation of sport. The same positional data stream, the same sampling frequency, the same transmission infrastructure — the thing that lets me compute PPDA and sprint distance is also the thing that lets an in-play betting market operate at millisecond resolution. I am not writing this piece to make any claim about betting; I do not have the data to do so. I merely record a structural fact: the data infrastructure I use for analysis and the data infrastructure used for grey-zone markets are the same infrastructure, and that raises a question of responsibility the analytics industry has not answered.

When I tell German partners that every crisis is unlabelled data, I usually get a polite nod. But when I add that, precisely for that reason, mislabelling a crisis is itself a crisis, the room goes quiet. A mislabelled dataset gets reused in other models, in other reports, in other decisions, and every reuse multiplies the error.

Three scenarios for the next cycle

By my own convention, every transfer analysis must end with three scenarios. On this piece's subject — the input quality of sports analytics systems — I keep the same convention.

The optimistic scenario: sports analytics organisations adopt a minimum input gate. The gate requires at least one named entity, one specific discipline, and three sourced information points before the deep analysis layer is allowed to run. When the gate fails, the system returns a hard error rather than a descriptive document. The cost is close to zero; the value is enormous: it turns a dangerous empty report into a clear operational signal.

The base scenario: the gate is not applied universally but is applied in high-value processes, such as investment due diligence and scouting. Lower-value processes continue to produce handsome but empty report files, and occasionally a bad decision still slips through.

The pessimistic scenario: no gate is applied. Empty files keep being read as clean bills of health. A few deals are signed on the belief that "there is nothing to worry about". When the problem erupts, the cause is attributed to a specific individual rather than to a collective architectural failure. And the industry learns the same old lesson the most expensive way.

Signals to track in the coming cycle

I do not believe in intuition — I believe in the decay coefficient of intuition. So instead of a conclusion, I leave a list of signals I will be tracking over the next six months.

First, the share of extraction files returned empty across the whole system. If that share rises, it is not a sign that articles have become duller, but a sign that the extractor is degrading or that sources are changing format. One simple operational metric can separate those two causes: the distribution of source formats over time.

Second, the provenance of domain labels. If the label "esports" is assigned from metadata, URL, or distribution channel rather than from body text, it is not reliable as an analytical basis. This is a cheap check and I recommend every analytics team run it periodically.

Third, the number of scouting decisions made on datasets that have expired under the decay coefficient. If that number is greater than zero, the organisation is buying risk without knowing it.

Fourth, the distinction between "no findings" and "unable to assess" inside internal documents. This is a small change in wording and a large change in culture. An organisation that can distinguish those two states is already ahead of most of the industry.

What I still ask myself

The empty-stadium summer, I heard the data falling drop by drop. In 2026 I heard it in the gap left by the crowd. This year I heard it in the gap of a forty-page PDF. Some matches end when the referee blows the whistle — and some only begin when the data speaks. But the question that keeps me awake is not when the data will speak. The question is: when the data is silent, how many people in my industry dare to write the two words "not measured" in the conclusion box, instead of filling it with a more plausible-sounding story?

I know I will not answer that question with a dataset. But I know exactly how to start testing it: count the empty cells in every report my organisation has sent out over the past twelve months, and read back what they were interpreted as.