Data Voids in the Transfer Window: The Discipline of Saying 'Insufficient Evidence'

In January, in the middle of the transfer window, a fourteen-page dossier landed on my desk. Fourteen pages, and almost every data cell was empty. Minutes played: empty. Chances created per 90: empty. Average distance covered: empty. What remained were three handwritten lines from a scout and a four-minute video link cut by an agent, containing only successful touches.

The sender wanted a price, a signature, a decision within ten days. I sent back a single line: insufficient evidence to reach a conclusion. Twenty minutes later the reply arrived: "We hired you for answers, not for a list of questions."

That exchange is the entire transfer-window problem compressed into one email: the machinery is built so that nobody ever has to say the words "I do not know". Agents are not paid to hesitate. Clubs are not paid to wait. Journalists are not paid to write that there is nothing to write yet. And analysts are paid to turn that emptiness into a number confident enough to sign.

It took me years to understand that the hardest part of working with football data is not finding a number. It is determining when the number does not yet exist.

Why the cells are empty

In Vietnam's top division, a team plays roughly 26 matches a season, about 2,300 minutes. That sounds like a sufficient sample until you look inside it: most of those minutes come from matches where event data depends on hand-coding, with different error margins between coding crews. Young players typically have 300 to 600 top-flight minutes, with the rest spent in the second tier or in friendlies nobody codes at all.

Vietnamese football moved through 2026 to 2026 with a generation evaluated more by storytelling than by numbers. We know the U23 side reached the 2026 Asian final in Changzhou, that the senior team won the 2026 AFF Cup, that Vietnam reached the third round of World Cup qualifying for the first time. But ask for the average distance covered across those four knockout matches, or the successful pressures per 90 in midfield, and reliable answers are scarce. The same applies to the Quang Hai, Cong Phuong and Van Hau generation: their continental journeys survive mostly in newspaper pages, not in event data.

National memory is rich; the data warehouse is poor. That does not make Vietnamese football worse than anyone else. It makes every conclusion about Vietnamese football more expensive, because it must be paid for in observation time rather than in a single file download.

Three kinds of void

A data void is not one thing. I separate three, because each demands a different fix.

A coverage void appears when nobody records anything. A second-tier match has no coded events, no shot coordinates, nothing beyond the scoreline. The only remedy is live observation, and live observation has a ceiling: the human eye captures roughly twelve to fifteen meaningful actions per match before memory begins to edit itself around the result.

A sample void appears when data exists but is far too thin. A 19-year-old midfielder has 320 minutes, 180 of them as a substitute from the 75th minute in matches already decided. Every metric built on that base carries a confidence interval too wide to price anything. The remedy is conditional pooling: merge only minutes that are comparable in role, opponent quality and match state.

A structural void is the most dangerous, because it does not look like a void. A player who spent two seasons in a 4-2-3-1 moves to a club playing 3-5-2, carrying a beautiful set of numbers that no longer mean anything. He did not get worse. He is answering a different question from the one his new club is asking.

Every number is a testimony; only the patient listener hears the whole trial. And testimony lifted from another case is unusable, however credible the witness who signed it.

The minimum viable sample

I do not believe in luck - I believe in a sufficiently large sample. But "sufficiently large" must be defined before collection begins, otherwise it stretches to fit whatever conclusion someone wants.

For Vietnamese football, my working threshold is 900 top-flight minutes, at least 400 of them against above-average opposition. Below that threshold, the file changes mode: it is no longer a valuation report but an observation report, and its conclusions must be expressed as ranges.

Thirty minutes of agent-cut video is not a sample. It is an advertisement, designed by someone with a direct financial interest in my believing it. Nobody sends me four minutes of midfield turnovers.

The fix is not to refuse watching. The fix is to build my own cut: every turnover in the first 30 minutes of the first half, every dribble conceded down the right channel, every situation where he is the second-closest player in the box during an opposition counter. A self-built cut costs four times as long, and that is exactly why it is worth it: time is the only thing that forces an analyst to select hypotheses instead of hoarding metrics.

The lesson of a rejected deal

In January 2026 I was asked to assess a 21-year-old midfielder in Argentina. His profile sat in a grey zone: an xG chain of 0.45 per match placed him in the top five percent of the league, but his average distance covered was 9.8 km, below the 11.2 km threshold many Asian clubs use as a filter.

The club's sporting director looked at the second line only. He rejected the file within ten minutes and signed a domestic midfielder with better running numbers. The Argentine player was Enzo Fernandez. Seven months later he joined Benfica, five months after that he won the 2026 World Cup and the tournament's best young player award, and in January 2026 he moved to Chelsea for what was then a British record fee, around 106.8 million pounds.

I retell this not to congratulate myself. I retell it because it contains two symmetrical errors. The director's error was turning a single metric into a knockout criterion. My error was placing two metrics side by side without explaining that they measure different things: xG chain measures value inside chance-creation sequences, while distance covered measures off-ball workload. A playmaking midfielder in Argentina running less does not mean he is lazy; he operates in a possession-and-distribution system where space is created by passing, not sprinting.

I stopped drawing conclusions from any single indicator, and began every report with a section titled "evidence against me". If, across the final three matches of a season, the metric I intend to lean on collapses against a high-pressing opponent, that belongs at the top of the report, not in an appendix. The decision-maker has a right to know where my argument is weakest. Without that section I am simply selling confidence, and confidence cannot be priced in euros.

The empty-stadium laboratory

In 2026, when European leagues returned behind closed doors, I was handed an experimental condition modern football had never offered anyone: same team, near-identical opponents and schedule, one variable changed, the presence or absence of noise in the stands. Recalculating home teams' PPDA across five seasons, the pre-pandemic average was 9.6; in empty stadiums it fell to 8.9. Home teams pressed less when nobody was cheering for them.

The empty stadium is the largest laboratory modern football has ever had. It showed that much of what we attribute to home advantage - the pitch, the weather, travel distance - actually sits in the ear. And it taught something more useful for my current work: any metric measuring collective will must be read alongside the social state of the match.

What does that mean for Vietnamese football? A meaningful share of home strength in full stadiums such as Hang Day or Thien Truong lies in factors no data table records: when the crowd starts to roar, how referees react to volume, how a young player handles a misplaced pass while an entire stand exhales. When I assess a player moving from a packed ground to a quiet one, I deduct three to five percent from his attacking output depending on position. That margin sits in no public model, and it is one of the few genuine competitive edges available to an analyst working in Southeast Asia.

What xG cannot do

xG is not the truth - it is a compass, and a compass never points to a shortcut. Expected goals describes the quality of a shot under the average conditions of a large sample. It does not know the player is in his fourth match in ten days. It does not know his team just lost its first-choice centre-back. It does not know the pitch is flooded.

The danger runs the other way too: when xG and the scoreline diverge, people assume the scoreline is lying. Not necessarily. A team conceding twice from shots worth 0.07 xG may simply have a systemic positional error repeated in both situations, and the repetition is the signal, not the xG figure.

For years I have reviewed matches against exactly three fixed questions and no more: where did the chance originate, who broke the last line of defence, and did the shape recover within seven seconds of losing the ball. Every metric afterwards exists only to count how often those three answers occur.

The price of manufactured certainty

The transfer market rewards certainty, not accuracy. A report saying a player is fifty percent likely to start, thirty percent likely to improve, twenty percent likely to fail will be judged indecisive, even when that is the most honest description the data supports. Meanwhile a report that is completely wrong but emphatic is remembered, and occasionally paid for twice.

That is why transfer numbers drift so far from football value. In the transfer market, an 80 million euro figure can be... a joke. It can be an auction between two Saudi clubs that need a face for a national tourism campaign more than they need a striker, or the product of financial rules forcing a European club to sell before buying, producing a deal priced by a deadline rather than by ability. Nothing in that market is meaningless; most of it simply is not about football.

Croatia 2026 taught me that a 12 percent probability is still a number worth betting on. What that lesson truly teaches is that the tail of a distribution only matters when conditions converge. A team reaching a final with a low model probability needs three things at once: a goalkeeper at peak form, a defensive system that does not depend on individuals, and a bracket containing no more than two favourites. Croatia 2026 had all three. With only one, a 12 percent probability is a polite way of saying "impossible".

There is another invisible referee shaping league outcomes that rarely enters player data: rule changes. More substitutions, revised added-time calculations, tightened video review thresholds. Each rule change revalues one profile and devalues another, not because players become better or worse but because the game changes shape. Champions in the first two years after a rule change are praised for character; in reality they adapted faster, and that adaptation gets mistaken for ability.

Signals for the next window

The next transfer window will carry more headlines and less verifiable data, because money in Southeast Asian leagues is growing faster than the supply of analytical staff. I will watch four signals.

First, the ratio of signings to full-time scouting positions. A club signing eight players while nobody tracks opponents weekly is buying lottery tickets with bank money.

Second, contract structure rather than transfer fee: release clauses, sell-on percentages, actual term versus announced term. In Vietnam money flows mainly through shirt sponsorship and local commercial deals; when a global sponsor arrives, what it buys is exposure metrics, not a relationship with the stands, and that gradually erodes the bond between a club and the community that produced it.

Third, how clubs publish their own uncertainty. The club willing to write "insufficient data" into an internal assessment is the club building a real process.

Fourth, the share of minutes given to under-21 players in matches where the result still matters, not in fixtures decided before kick-off.

Numbers never lie - only the way we read them is wrong. In a window where every meeting demands an answer within ten days, keeping one cell blank on the spreadsheet and daring to write "unknown" in it may be the single best decision a club makes all season. Analysts are not paid to be certain. They are paid to be the one person in the room who, after everyone already knows the answer, goes back and checks whether the question was right.