Forty Empty Data Cells in the Transfer Market: An Evidence Filter Built by a Man with a Spreadsheet

Forty Empty Cells in the Middle of the Rumour Market

In the second week of the transfer window, a colleague sent me a nine-part report on a transfer target. Every section had a proper heading: league context, form by matchday, injury history, contract structure, pressing metrics, a heat map by minutes played. And every cell carried the same line: insufficient information to conclude.

He attached an apology, saying he had not finished the job. I read it twice and filed it in a folder labelled "gold standard". In a transfer window where hundreds of lines of news appear every hour about deals that never existed, what I need is not another prediction. What I need is someone willing to write into the conclusion column that there is nothing to conclude yet.

Numbers never lie. They simply wait for someone calm enough to listen.

Three kinds of noise in one window

A transfer market runs on three overlapping streams of noise, and all three share one property: their volume is inversely proportional to the real evidence behind them.

The first stream comes from agents. This is a price-driving profession, and I do not treat that as a sin. A good agent is supposed to create competition for his client. The problem is that journalists often quote that noise as if it were primary data. When an agent says "three European clubs are watching", that is not data. It is a statement with a purpose, and the purpose sits with the speaker, not the listener.

The second stream comes from clubs themselves. Teams understand the media value of a rumour. A name linked to your club for two weeks can lift engagement, ticket sales, even your valuation in a sponsor's eyes. I have seen deliberate leaks, and in several cases the leaker held the signing pen.

The third stream comes from fans and social platforms. It is the loudest and the easiest to filter, because it creates no new information. It only amplifies old information. An account spreading a rumour does not make the rumour truer; it only makes it harder to ignore.

Together those three streams create very specific pressure on the writer: have an opinion, commit, "predict the deal". That is the moment the profession starts losing control.

A five-tier evidence ladder

I built a classification ladder to protect myself from my own impatience. It has five tiers, from printable to ignore.

Tier one is a sealed document: a published contract, a federation transfer certificate, an official club announcement. No commentary needed, just a note with an absolute date.

Tier two is indirect administrative paperwork: a registration, a submitted matchday squad, a signing photo without a press release. It requires cross-checking against at least one independent source.

Tier three is an attributed quote: a named sporting director, a coach at a press conference. Valuable, but it must be read against the circumstances of delivery. A coach speaking before a derby is not the same as that coach speaking after being sacked.

Tier four is an unattributed indirect source: "someone inside", "a source close to". Use it as a lead, never as a conclusion.

Tier five is circulating rumour. Tier five is not data. It is the raw material of data, like a name without a shirt number.

Applying the ladder to a specific deal, I always start by listing what can be measured: actual minutes across the last two seasons, sprints per ninety, duel success rate, expected goals and expected goals against per ninety, remaining contract years, release clause structure, and where the player sits on the age curve for his role.

Based on my experience watching matches, one rule has held for twenty years: any metric that cannot be sourced is not allowed into a conclusion. I worship data, but I pray through verification.

The 2026 turning point and the data box

In 2026 I was thirty-two, the only data reporter at a newsroom in Nha Trang. After matchday twenty of the V.League, I published a series using expected goals against to show that Thanh Hoa's defence, then praised as the best in the league, was actually conceding more than expected: 1.9 xGA per match, with a goalkeeper save rate of only 64 per cent.

The coaching staff called me a man sitting in a cold room. On 7 February 2026, Thanh Hoa lost 0-3 to Ulsan Hyundai in the AFC Champions League play-off, exactly the script the numbers had shown a month earlier.

The lesson was not that I was right. The lesson was that had I published five days later, or omitted the save rate, or failed to state my sample of twenty matches, a single sentence could have demolished my conclusion. From then on I pushed the newsroom to standardise a data box at the end of every match report: source, sample size, update date, and the name of the person who checked it. It became mandatory for every reporter.

That rule followed me into transfer work. A deal without a data box is not a deal. It is a story.

When pressing metrics ran ahead of public opinion

In 2026 I was sent to Russia for the World Cup, aged thirty-three. The Asian press corps was praising Germany's defence. I sat with my spreadsheet and found a number that did not fit: Germany's PPDA, the passes allowed per defensive action, had risen from 7.3 in the 2026 cycle to 12.8 in 2026 qualifying. A rising figure means the team had lost high pressing. Pressure was no longer being applied in the opponent's half.

I wrote that Germany would go out in the group stage. Colleagues laughed. On 27 June 2026, Germany lost 0-2 to South Korea in Kazan and finished bottom of Group F.

PPDA did not take me to Russia. It only opened a door, and I walked through it myself. That is what I tell young reporters: a metric does not turn its reader into an expert. A metric opens a door. Walking through is the analyst's job, and that job includes accepting that some doors must stay shut.

Empty stadiums and the value of a control sample

In March 2026 the pandemic emptied every stadium. At thirty-five I saw a natural laboratory nobody wanted to exploit: fourteen home matches for Binh Duong with fans, at 1.85 expected goals per match, against ten matches without fans, at 1.31. A gap of twenty-nine per cent.

Home advantage was inflated. I published the study, stated the sample of twenty-four matches, stated that this was one club in one unusual period, and stated that the finding could not be extrapolated to the whole league. In August 2026 I signed a full-time data consultancy with Binh Duong and left the newsroom.

What I carried out of journalism was not a spreadsheet. It was the habit of stating my own limits. An analysis without a limitations section is an unfinished analysis.

The two-hat rule and its price

In 2026, aged thirty-seven, I was both a long-serving writer and a consultant for Khanh Hoa. My model showed Morocco's defence as the most undervalued at the World Cup: a 71 per cent success rate on the offside trap, and goalkeeper Yassine Bounou beating post-shot expected goals by plus 3.2. My series drew attention, and I was first to report a loan move between two Portuguese clubs based on fitness data.

Then Khanh Hoa were fighting near the bottom in 2026. My old newsroom wanted me to disclose internal data to keep my press credentials. I chose to protect the club. The relationship ended.

Since then I keep a two-hat rule: never mix exclusive club data into public writing, use only official platform data. It cost me sources. It also means every number I publish can be checked by someone else, which is the larger return.

The five-question filter

When a transfer story lands on my screen, I run five questions in a fixed order.

One: which tier of the evidence ladder is this. Tier four or five means a note, nothing more.

Two: how many independent sources confirm it. Three outlets repeating one original article is still one source.

Three: what problem does the deal solve on the pitch. Buying a striker to fix a goal shortage is logic. Buying a striker when you already have three in that role signals a decision that did not come from the analysis room.

Four: what does the financial structure look like. The fee is the tip. Below the waterline sit wages, signing fees, agent commissions, bonuses and sell-on clauses. Forty million paid outright is not forty million paid in twelve instalments with performance add-ons.

Five: is the sample big enough. If a player has under one thousand top-flight minutes, every advanced metric carries a warning label.

Those questions filter most noise. They do not filter the person asking them. That is the real limit, and the part I must write before writing any conclusion.

The blind spot of the spreadsheet man

Correlation is not causation, and this is where I have been wrong more often than I care to admit.

In 2026 I found a beautiful correlation between sprints per ninety and points across ten matches. I nearly wrote that teams who sprint more win more. One morning I reopened the positional data and found the real driver: winning teams trail less often, and teams that rarely trail rarely need to sprint in recovery. My variable did not cause the outcome. It was a consequence of it.

Had I published it, I would have added another junk metric to the industry.

The second problem is small samples. Ten matches is small. Twenty is small if those matches fall in an abnormal period. One win against the leaders proves nothing about a team's real strength. I want at least three consecutive seasons with a roughly stable squad structure before I allow myself a sentence that judges a level.

The third problem is mental state, which I once tended to avoid because it is hard to measure. My current approach treats emotion as qualitative data: quote the words verbatim, record when and where they were said, and label them as qualitative data rather than metrics. A striker returning after three hundred days out with a cruciate injury can hit every fitness benchmark and still move half a beat late. Fitness metrics cannot measure that half beat. Video can.

The fourth problem is overconfidence in a clean dataset. A dataset can be perfectly accurate on every touch and still produce a wrong conclusion if the analyst asked the wrong question. Clean data does not rescue a dirty question.

Luck is the residual

I always keep an empty final column in my spreadsheets, labelled residual. It holds what the model cannot explain: a shot against the post, a red card in the eightieth minute, a downpour in the seventieth.

Luck is the residual the model cannot explain, and I never round it to zero.

Keeping that column has one practical effect: it stops me writing absolute prophecies. A data consultant who says "certainly" is a consultant about to lose his job. The one who says "the probability leans this way, with this margin of error" lasts much longer.

Why an empty report still has value

Back to the nine-part document with forty empty cells.

Its value is that it blocks a deal being judged by feeling. When the analysis says "not enough data", the recruitment room must choose: go and collect more data, or admit the decision is intuition and own that intuition. Both options are far better than dressing an intuitive decision in the language of numbers so it looks scientific.

In this industry the most expensive thing is not data. It is honesty about which data you are missing.

This is where I part company with most online practice. Decisiveness is rewarded. An article claiming "this player will succeed" travels further than one saying "not enough data to conclude". But if the whole industry rewards only decisiveness, the end state is a market full of verdicts that are never audited.

I once told a training session that if your spreadsheet has not returned "insufficient information" in a month, your spreadsheet is serving someone else rather than the truth.

Signals for the next window

Next window I will track four measurable signals instead of four rumours.

First, release clause structure: the number lives in the contract, not in the article. Second, the wage bill after the deal completes, because a club can pay a low fee and very high wages, or the reverse, and those two paths lead to two different dressing rooms. Third, actual minutes in the first three months, the one metric that cannot be faked. Fourth, how often the player is used against top-half opponents, where advanced metrics carry the most meaning.

A season should be read as a sequence of probabilities, not a sequence of events. So should a transfer window.

Forty empty cells this week may become forty filled cells next month, once the data arrives. My job is to keep the order intact: data first, conclusion second, and one residual column always left blank.

Before I trust a reputation, I need to see the data behind it. And when the data has not arrived, I choose to write that it has not arrived.