The Limits of Tactical Analysis: When Data Is Empty and VAR Measures in Frames

At 11:40 pm on 12 March, I opened a file on my computer in Milan. It had nine fields. Title: empty. Source: empty. One-sentence summary: empty. List of information points: an empty array. The entities field instructed me to identify entities from the information points above, while nothing above existed.

I stared at the screen for ten minutes. The reflex of a man who once published a 6,000-word analysis of Atalanta is to fill the gap. A story assembled itself in my head: a match, a club, a transfer, a manager about to be sacked. Every detail was plausible. None of it was true.

In the summer of 2026 in Moscow, I came close to doing exactly that. I had enough data for a piece on the France-Belgium semi-final, but I was missing a different layer, and I filled it with a guess. The article failed precisely where I was most confident. It took me three months to realise I had been reading the position wrong.

That March night, I did not fill it. I closed the file and rewrote the process.

Gaps have shapes

Football analysis changed fast in a decade. GPS vests became standard in Serie A from the mid-2010s, logging thousands of data points per player per match. Expected Goals moved from academic concept to everyday commentary language. PPDA — passes allowed per defensive action, where lower means more aggressive pressing — became the standard measure of intensity. Heat maps, pressure charts, passing networks: all drawable in minutes.

Content volume multiplied. Verification infrastructure did not move.

The result is a paradox: the more data exists, the fewer people check where it came from. Gaps do not disappear. They change shape and become harder to spot.

I have classified five recurring types.

The first is an empty input. The second is a sample too thin — three matches are not enough to judge a system, but they are enough to write a headline. The third is a resolution error: correct data at a level that does not match the question. This is the most dangerous type, because it looks entirely credible. You cannot argue with a number. You can only argue about which question the number answers. The fourth is ignored negative data: nobody writes that a player has no injury record across four consecutive seasons. The fifth is a named-but-anonymous source — not a source at all, but a gap wrapped carefully and pushed into the information stream wearing the clothes of a fact.

4,500 situations and one detail outside the chart

In the summer of 2026, football stopped. I stayed indoors, re-watched 4,500 wide-attacking situations from Serie A between 2026-16 and 2026-19, and hand-drew 38 pressure maps. I wrote nothing for six months. The answer only arrived in June 2026, when the Euros began and I turned forty.

Inside that archive I found a pattern: Nicolò Barella and Marco Verratti were generating an average of 14.7 passes into dangerous zones per match through triangular movement. The model had never appeared in my dataset before. Then I found a hole inside my own conclusion.

The 14.7 measured passes into dangerous zones. It did not measure the thing that made those passes dangerous: tempo. Barella accelerates after receiving 8 to 12 metres out. Verratti decelerates to hold 6 to 9 metres, pulling defenders out of position before releasing. No metric in my set encoded that difference, because I had never thought to record it.

The number does not lie, but it does not tell the whole story either. A heat map shows position; an intent map shows thought. I had the first for three years and believed I had the second.

Frames are not millimetres

On 2 November 2026 at Villa Park, a Roberto Firmino goal for Liverpool against Aston Villa was disallowed, with the offside point measured at the shoulder. Liverpool still won 2-1 with late goals, but the image of a red line crossing a player's shoulder entered the league's history of controversy.

European broadcast feeds commonly run at 50 frames per second. A sprinting player covers roughly 8 metres per second. That is about 16 centimetres between frames. A decision that comes down to 2 centimetres sits entirely inside the measurement error of the instrument. The system does not lie. It answers a different question from the one the crowd is asking.

At the 2026 World Cup in Qatar, semi-automated offside technology combined 12 tracking cameras monitoring 29 body points with an inertial sensor in the ball transmitting 500 times per second. FIFA reported that average offside determination time fell from around 70 seconds to around 25 seconds. That is a real gain in speed. Faster checks do not automatically mean a more precise decision boundary. Calibration error, the exact moment of ball contact, and the definition of which body part may score all remain human.

I have watched enough to see the tactical consequence. When the penalty for setting off ten centimetres early becomes near-automatic, strikers wait. They attack the box less often before the pass. One Serie A coach told me he trains his players to hold half a beat, because half a beat is cheaper than a disallowed goal. Attacking instinct is being taxed by data.

The player with no injury file

When a club assesses a transfer, it holds three data groups: minutes, technical metrics, medical record. The third is the least present in public analysis and the heaviest in the final decision.

I once sat in an internal meeting at a mid-tier Serie A club. Two midfielders were on the table. Profile A was better on nearly every technical metric and priced about 40 per cent higher. Profile B had a fact Profile A lacked: four consecutive seasons above 2,800 minutes, with no muscle injury costing more than two weeks.

No column in my spreadsheet recorded the absence of injury. Ask what the system has hidden before judging a defender. Profile B signed and played steadily for two seasons. Profile A played 1,100 minutes in his first season elsewhere. Nobody can call that outcome inevitable. What can be said is that the probability was visible and ignored, because it was not in the chart.

Prices made out of nothing

The transfer market is where data gaps pay best. Rumours come in three tiers: club confirmation; two independent outlets with named negotiators and dates; and an unnamed source close to the situation. Most social-media transfer content is tier three, circulated in the prose of tier one. A fee of 60 million euros without a payment structure is a meaningless number — paid up front or spread over five years with add-ons, the cash and accounting values diverge sharply.

Financial rules make the gap more expensive. In November 2026 Everton were docked 10 points for breaching the Premier League's profitability and sustainability rules; in February 2026 the deduction was cut to 6 on appeal. In March 2026 Nottingham Forest were docked 4 points. Prices did not adjust. Prices are set not by data but by belief about data, and belief needs no verification to spread.

The underdog run is not a system proof

Greece won Euro 2026 with 7 goals in 6 matches, winning all three knockout games 1-0 under Otto Rehhagel. Croatia reached the 2026 World Cup final after three consecutive 120-minute knockout ties, two decided on penalties. Morocco became the first African semi-finalists in 2026 under Walid Regragui, conceding one goal from open play up to the semi-final.

All three are told as proof of system. Examined closely, they are proof of format. A knockout match is a sample of one. Emotion is not data noise; it is undecoded data. The right question about a small team going far is not how good their system is, but how much of that success is repeatable.

The blind spot is resolution

The biggest error in football analysis is not a wrong conclusion. It is a correct conclusion at the wrong resolution. My mistakes share a pattern: I rarely took bad data. I used good data to answer a different question from the one readers were asking, then bridged the gap with my own authority.

The industry rewards plausible certainty and offers no reward for writing three lines about why an empty file cannot be analysed. So the largest gap of all is inside the writer. Three tests now precede anything I write. Does the source have a name and a date. Is the sample bigger than one. Which question does this metric answer. They have filtered out roughly a third of my planned pieces — the readable ones, with beautiful numbers, wrong at the foundation.

Next match, every time a red line appears and the crowd howls about two centimetres, I will ask which frame the system is drawing on. The answer may annoy me. It may also be the first time in years I see what I am actually looking at.