July in Boston, and the second monitor glows with an eleven-page transfer dossier. Eleven pages, but only three fields are filled in: name, date of birth, position. Minutes played, left blank. Progressive passing metrics, left blank. Current wage, release clause, image-rights split — all of it sits behind a single line of text: "insufficient information."

The sender is a technical director at a top-flight club. The accompanying question is one line long: "Is he worth that fee?"

In 2026 I would have answered with a feeling. I used to do that, and I used to pay for it. My job now is interrogating data, and the first principle of that job is simple: when the testimony is absent, you do not deliver a verdict — you adjourn the hearing.

The result is a lie that time has memorised; xG is the confession. In the transfer window, that confession goes by other names: contract structure, actual minutes, wage-to-revenue ratio. Not the six a.m. headline.

The transfer window is the one moment in the year when silence carries more economic value than speech. An empty dossier is not a bad dossier. It is an unverified dossier, and those two things differ by enough to shift a player's valuation by fifteen percent because a single dash was filled in the wrong box.

Context: structured noise, unstructured signal

Summer 2026 runs on a mechanism I have watched for eighteen years: for every contract actually signed, somewhere between forty and sixty rumours are pushed out ahead of it. That ratio is not an accident. It is a product. Agents need pricing pressure, clubs need negotiating pressure, platforms need page views, and supporters need the feeling that their club is doing something.

The problem is that noise is structured while signal is not. A rumour can be manufactured to formula: one player, one big club, one plausible transfer fee, and a window of "the next forty-eight hours." A real signal — a release clause triggered, an instalment schedule spread across fiscal years, a wage pushing against the salary ceiling — appears quietly, usually as a single line in a half-year financial report.

My department built a four-tier evidence ladder and applies it to every dossier that crosses the desk.

Tier A is what has been signed: official statements, registered contracts, match data from licensed providers. It is the only tier that can be used for a financial decision.

Tier B is what has been cross-confirmed: a release clause cited at the same figure by two independent sources, a loan with an obligation to buy, medical documents leaked from two different sides.

Tier C is what is attributed to authority: agent statements, articles citing "sources close to the situation," reporters with a seven-out-of-ten hit rate.

Tier D is everything else: social media, anonymous accounts, airport photographs.

Of those four tiers, only A and B carry weight in a valuation model. Tier C is for monitoring, not for calculating. Tier D is entertainment, and it never enters a client report.

The eleven-page dossier sits somewhere between Tier C and Tier D. The three filled fields come from an article citing no source. That is why I answered with a request rather than a ruling: send me event data from the last two seasons, and a copy of the current contract if it can be accessed lawfully.

Core: four old trials as precedent

To explain why I do not rush, I usually retell four cases I was personally involved in — four occasions when data overturned a conclusion the crowd had already signed off on.

June 2026, at Foxborough, New England Revolution hosted Toronto FC. Toronto held 72 percent of possession, fired 21 shots, and finished with 2.3 expected goals. The scoreline: a 0-1 defeat, the only goal from Diego Fagundez. I was an intern writing match reports, and my editor asked me to celebrate the home side's "inspired night." I pulled the data from StatsBomb and wrote the opposite case: Toronto deserved to win 3-0, and the scoreboard was lying. The piece reached fifty thousand reads in twenty-four hours, and the newsroom had to publish a correction.

The lesson was not in the read count. The lesson was that the same match, the same dataset, produced two different truths depending on how it was narrated — and only one of those truths could be used to forecast the next match. After that, I set my own rule: when the numbers and the story disagree, I trust the numbers, and I publish the disagreement instead of hiding it.

The second case was Croatia at the 2026 World Cup. I was commissioned to build a PPDA table for all thirty-two teams before the quarter-finals. Croatia posted 8.9 — meaning each defensive sequence allowed the opponent an average of just 8.9 passes before being disrupted, the lowest of the eight remaining teams. Croatia's 2026 PPDA board did not measure pressure; it measured pride. Marcelo Brozović covered 13.8 kilometres and made nine ball recoveries against Argentina. I wrote that Croatia did not have luck, Croatia had a system. When they reached the final, I became a name cited in data meetings, and a Championship club hired me as a part-time data consultant.

What matters here is not the correct prediction. What matters is the method: I did not measure courage, I measured the number of passes the opponent was permitted to complete. When a team is pinned back repeatedly and still holds the scoreline, people call it character. When that metric repeats across seven matches, people have to call it a system — and systems can be bought, while character cannot be priced in the transfer market.

The third case was the season without crowds in 2026. The Boston consultancy where I worked cut forty percent of its headcount. I did not ask for an exemption. I wrote a report titled "The Stand Effect: Evidence from 372 Bundesliga Matches Before and During COVID." The findings: home win rate fell from 45 percent to 31 percent, and penalties awarded fell by 28 percent. Empty stadiums in 2026 were a natural experiment: football did not need a crowd to reveal its nature. Huddersfield Town hired me for the final eight rounds of the Championship. I proposed a rotation model built on sprint distance above the 6 m/s threshold; any player who fell below eighty percent of that threshold in two consecutive matches had to start on the bench. They took fourteen points from twenty-four and survived relegation by exactly one point.

The fourth case was Morocco at Qatar 2026. Before the tournament I published a series arguing that Morocco do not defend, they operate data. Yassine Bounou posted a goals-prevented figure 4.3 above expectation; Achraf Hakimi completed 6.8 progressive passes per match. When they beat Portugal 1-0 and reached the semi-finals, international platforms called me. In the summer of 2026, a Saudi investment fund asked me to assess Cristiano Ronaldo for a contract extension. I submitted a forty-page report: his actual xG created was 0.55, inflated to 0.82 by set-piece situations. I recommended against further spending. The fund objected. Three months later, Ronaldo's market valuation had fallen fifteen percent.

Four cases, four times the same structure. A conclusion the public accepted, a dataset it ignored, and the gap between the two — and that gap is where the value sits.

Transfer data is like a tide: you cannot read it from the surface, you have to measure the seabed. The surface is the headline. The seabed is the three things the eleven-page dossier lacked: actual minutes by role, wage structure relative to the buying club's salary ceiling, and remaining career length measured against the sprint-speed decline curve.

Based on my experience watching matches, a striker producing 0.42 xG per ninety in a possession system drops to roughly 0.27 in a counter-attacking system, because touches inside the box fall by nearly half. That is the calculation no article prints, because it has no highlight reel. Yet it is the calculation that decides whether a signing returns a profit.

For an empty dossier, I compile a five-item risk flag table, and a single red flag is enough to halt a recommendation. First, a minutes sample too small for the fee. Second, most of the attacking output coming from set pieces. Third, defensive metrics dependent on one specific teammate. Fourth, a repeat history of hamstring or groin injuries. Fifth, a wage structure exceeding thirty percent of the buying club's revenue.

In this particular case, I could neither raise nor clear a single flag. And that is precisely the answer I sent back to the technical director: the fee can only be judged once the seabed has been measured, not once the headline has been read.

The contrarian angle: absent evidence is not evidence of weakness

There is one mistake I have made, and it cost more than any error I have ever made with numbers. It is the habit of reading data's silence as a verdict.

In my first summer as a full-time consultant, I removed a left-back from our shortlist purely because our data provider did not cover his league. Two years later he moved to a leading European club for six times the figure we had estimated. The cause was not the player's ability. The cause was an empty data field that I misread as a bad data field.

Correlation is not causation, and the absence of a correlation is not the cause of anything. In statistics this is called the missing-data problem, and there are at least three valid responses: collect more, use a proxy variable, or state clearly that no conclusion is available. The only invalid response is filling in the blank with intuition and calling it analysis.

This leads to a second, more uncomfortable paradox. Passionate supporters are usually treated as objects of data — as though emotion were a form of noise to be filtered out. I do not see it that way. A crowd packing a stadium is a measurable variable, and it shapes results in ways my model can quantify: a fourteen-percentage-point difference in home win rate between playing with and without a crowd. When I call emotion noise, I am dropping an explanatory variable out of the equation. When I measure it, I gain a coefficient.

A third paradox concerns how I import tools from esports. Esports logs every millisecond, every keypress, every turning decision. Football offers only discrete events, roughly three thousand touches per match. The temptation is to lay the entire esports framework onto grass, and that temptation produces models that are elegant and wrong. Pressure is not how much you run. Pressure is the number of passes you prevent per defensive sequence — PPDA. The same phenomenon, two measures, and only one of them says something about the next match.

The same holds for the transfer window. Social media engagement is a measurable index, but it measures curiosity, not player value. A post with ten million views about a collapsed deal may tell you more about the content distribution algorithm than about the club involved.

Takeaway: the next cycle and the signals worth measuring

This transfer window will leave behind three signals I intend to track until the next one.

The first is clubs splitting remuneration into two tiers: fixed wages compressed, bonuses for minutes played and performance metrics pushed up. This structure shifts injury risk from the club to the player. When a team applies it to more than three contracts in a single window, that is a sign of financial hedging, not of ambition.

The second is release clauses expressed as a percentage of the buying club's revenue rather than as absolute figures. It is a contract form that appears only at clubs with strong legal departments, and it makes deals harder to replicate — a competitive advantage that never appears on the pitch.

The third is minutes played by under-23 players in high-stress matches, measured by the share of passes completed under pressure. A club that gives young players minutes in peak fixtures tends to gain a valuation advantage over the next two transfer windows, simply because the market pays for evidence rather than for potential.

I still keep that eleven-page dossier in the folder; I have not deleted it. Three weeks later, the supplementary data arrived. Actual minutes came to forty percent of the figure cited in the rumour, and seven of eleven goals came from set pieces. The quoted fee was thirty-eight percent above our model's valuation. I sent back a short note: no conclusion, just a risk flag table with three red boxes.

The club did not sign him. And throughout the entire process, there was not a single moment when I needed the score of any match on any scoreboard.