Marseille, March 2026. Twenty minutes after the final whistle at the Stade Vélodrome, I sat in the fourth row of the press conference room holding an A3 sheet on which I had hand-drawn the movement paths of twenty-two players during a 1-3 defeat to Paris Saint-Germain. When my turn came, I asked the coach about the gap between the midfield line and the left-back. A male reporter to my right laughed: "Do women watch football with their emotions?"

I did not answer him. I unfolded the sheet and pointed to seven red marks: seven times in ninety minutes, Marseille's left channel was left open in exactly the same position. Bixente Lizarazu received the ball there, turned, and found nobody within ten metres. The room went silent. A week later I was invited to write a tactical column for La Provence. The nickname "Madame Tactique" was born from one A3 sheet and one pencil.

I tell that story not to talk about recognition. I tell it because it marks the dividing line between two eras of my working life. In 2026 in Marseille, data was something I counted by hand. In 2026, data is something a technical pipeline pushes onto my screen at seven in the morning, before I have finished my first coffee. And across those twenty-five years of transition, I have watched a kind of accident that nobody wants to name: the report that looks so complete that nobody checks whether it is empty.

The danger of empty data lies in the fact that it takes the shape of full data.

Context: from a notebook to a three-tier pipeline

Across the first two decades of this century, professional football analysis went through a quiet industrialisation. Clubs in Ligue 1, the Premier League or La Liga no longer hire a lone analyst with video-editing software. They run a pipeline with three tiers.

The first tier is ingestion. Optical tracking cameras, chips inside the ball, sensors in shirts, event data marked manually. This tier returns raw material: coordinates, timestamps, player identities, action types.

The second tier is deconstruction. Here raw data is broken down into discrete, citable information points: who passed to whom, where, when, under pressure from how many opponents, and with what outcome. This is the stage that operational documents call "stage-one deconstruction".

The third tier is deep analysis — where everything is placed on the table and dissected along nine dimensions: tactics, club finance, results cycles, league landscape, rules and governance, dressing room, risk profile, media narrative, and industry transmission chains.

The problem is this: the third tier is worth exactly as much as the volume of information the second tier returns.

During the past season, a top-flight French club handed me a thirty-two-page scouting dossier on a central midfielder playing in the Belgian top division. The cover had a logo, an author's name, a date, a digital signature. Inside were fourteen tables: heat maps, pass-distribution charts, ball-progression indices by zone, aerial duel win rates, converted defensive impact metrics. So beautiful that it took me forty minutes to notice the anomaly: half the cells were empty, and the software had not raised a single error. It simply left them blank and tidied the table.

That player was rated poorly for long passing because the system had not received long-pass data from the Belgian league's provider for six matchdays. He was not bad at long passing. Nobody had simply measured it. And in the report, that silence turned into a zero.

An empty cell and a zero are two entirely different truths, but on the coach's desk they look identical.

I call this phenomenon an integrity-gate accident. In every serious data pipeline in finance or healthcare, there is a mandatory blocking step: if the input fails minimum completeness thresholds, the whole process must stop and return the status "insufficient information to assess". Football barely has that step. We tend to fill the gaps with intuition, then call intuition experience, then call experience a conclusion.

Core analysis: three variables that decide whether a report can be trusted

Variable one — the measurement layer, where trust begins to crack

Every modern metric has a technical limit that ordinary users never see. Expected goals — usually written xG, the probability that a given shot becomes a goal — is the clearest example. The same shot from the left edge of the box can produce three different values from three different providers: 0.04, 0.07, 0.11. Nobody is lying. They use different models, different training datasets, different definitions of the goalkeeper's position.

That does not make xG useless. It makes xG a tool that must be read with a label attached. When a report cites xG without naming the source, the reader is receiving a number with no provenance — and in the spirit of a proper data pipeline, a number without provenance must be flagged as unverifiable.

I have spent years cross-checking my own tracking data against commercial event data in Ligue 1 and Champions League matches. The result made me calmer rather than more pessimistic. The discrepancy between two sources usually sits within an acceptable range when a match takes place under normal conditions. But the error spikes in three situations: heavy rain that makes optical systems lose the ball, matches with a high density of corners that obscure the penalty area, and competitions with under-invested measurement infrastructure.

That is why the first question I put to any data provider is never "how accurate is this metric". The first question is: "What do you return when you cannot measure?"

If the answer is "we skip it", the report may still be safe. If the answer is "we fill it with the league average", the report has been contaminated at the root.

Variable two — the attribution layer, where data becomes a story

There is a beautiful paradox in the Champions League final of 26 May 2026 at the Arena AufSchalke. José Mourinho's Porto beat Monaco 3-0. I rewatched that match eleven times in three days, and I still stand by the conclusion I wrote across twelve thousand words back then.

When people look at Porto 2026 and see a miracle, I see an equation waiting to be solved.

Porto had roughly forty-three percent possession. But they created five clear chances; Monaco created one. Monaco passed more, touched the ball more, and kept passing into areas where Porto had already set their traps. That is the geometry of proactive defending: winning the ball not by chasing it, but by luring the opponent into passing where you already stand.

If an analyst reads only the short post-match statistical table, he will conclude that Monaco controlled the game and lost through bad luck. That conclusion is wrong at the attribution layer, not at the data layer. The possession figure was measured correctly. It simply does not answer the question people assign to it.

Ludovic Giuly left the pitch with a groin injury in the first half, and Monaco lost the only link capable of breaking Porto's second defensive line with pace. Ricardo Carvalho and Jorge Costa did not need to duel much. Deco and Maniche did not need to hold the ball long. Carlos Alberto scored the opener in the thirty-ninth minute from a move that lasted just four passes.

Clean data. Dirty conclusion. That is the line I still use when talking to young coaches, and I often have to remind them that the line is not a criticism of data. It is a criticism of people who read data and forget to ask what the data was generated to do.

The same mechanism repeated at Barcelona in the 2026-2026 season. In the Champions League semi-final second leg at Camp Nou, Pep Guardiola's side had over seventy percent possession, played hundreds of passes, and won 1-0. But Mourinho's Inter Milan went through 3-2 on aggregate. The statistical table described a dominant match. Reality described a match broken exactly to the away side's plan.

And in the round of sixteen at the 2026 World Cup, Spain touched the ball more than a thousand times against Russia, held nearly seventy-five percent of possession, and went out on penalties. Once again: correct data, wrong story.

Space on a pitch is wider than any great figure who ever stood on it. A statistical table does not measure space. It measures what people chose to record about space.

Variable three — the transmission layer, where an empty cell becomes a contract

If the story stopped at the first two layers, it would remain academic. The third layer is where it carries the weight of real money.

Football data today flows down three major conduits. The first flows to the coaching bench. The second flows to the scouting desk. The third flows to the betting market. The third is the most dangerous, because it feeds back into the first two within seconds, and it does not care what the data means. It only cares which way the data moves the price.

In more than a decade of working in this field, I have always held one clear position: selling live match data directly to betting companies is the darkest side effect of the digitisation of sport. An injured player in the twelfth minute is logged and repriced before he has left the pitch. A misplaced pass in the eighty-ninth minute becomes a buy or sell signal within a shorter interval than a single breath.

In the second conduit — scouting — an empty cell can be the direct cause of money spent wrongly. I have seen it with my own eyes in transfer dossiers.

Transfers are the market of hope, and hope rarely follows valuation.

In January 2026, Chelsea spent more than three hundred million pounds in a single winter window, including Enzo Fernández arriving from Benfica for around one hundred and twenty-one million euros on deadline day. In August of the same year, Moisés Caicedo arrived from Brighton for one hundred and fifteen million pounds, breaking the record for a transfer between Premier League clubs. Both deals were justified with data models: ball-progression indices, ball-recovery indices, line-breaking pass indices.

The question I ask is not whether those two players are good. My question is: which model, with which data, across how many matches, and with what assumptions about convertibility from the Benfica and Brighton environments to the Chelsea environment? A ball-progression index computed on seventy percent of a season's matches has a completely different reliability from the same index computed on ninety-five percent of matches across three seasons. And in most dossiers I have read, the convertibility assumption is not written as a separate line. It lives in the author's head.

In the outbound conduit, the distortion is even larger. The Saudi Pro League has in recent seasons attracted a series of stars past their European peak with large contracts. My reading of that phenomenon does not concern the quality of the league. The published numbers — goals, assists, contribution metrics — are all real. They simply measure something other than what the league is actually buying.

A thirty-four-year-old leaving Europe is no longer a tactical variable. He becomes a walking tourism ambassador. He sells tickets, shirts, regional broadcast rights, and generates content for a new market. In that model, the metric that matters is not goals scored but impressions delivered. And if an analyst uses a tactical framework to evaluate a deal designed with a marketing framework, he will always conclude wrongly, in the opposite direction.

The counterintuitive angle: dirty data is not the enemy

What I want to say against most readers' intuition is this. Clubs do not fail because of dirty data. They fail because the data is so clean that nobody bothers to check it.

Dirty data incriminates itself. When a table shows visible empty cells, when a metric returns an infinite value, when a player's name appears twice in the same list, the analyst stops. He picks up the phone. He checks the source. Messiness is a safety signal.

The danger lies in the perfectly formatted report. Thirty-two pages, fourteen tables, consistent fonts, not one misaligned cell. It looks like a conclusion. And in a meeting that starts at eight in the morning with twenty-five decisions to make before noon, a document that looks like a conclusion will be accepted as one.

This is where I have to be honest with myself. I am the kind of person who prefers ideas to people, and I know the flip side of that trait. It makes me prone to building an analytical framework so tight that the framework lives on its own, generates its own conclusions, and protects itself from new data. After every major event in my career, I force myself to ask: what if my hypothesis is wrong, and which data would prove it? If I cannot answer, my framework has become an empty cell with good formatting.

Collapse is the largest data source. Throughout my career I have learned more from defeats than from wins, because a defeat forces every variable into the open. But only when I accepted that my own report could be empty did I start checking it before handing it to anyone else.

The biggest execution blind spot in football analysis today sits right here. We invest millions of euros in cameras, algorithms, and staff with data-science degrees. We invest almost nothing in the integrity gate. Nobody pays a salary to the person whose only job is to answer one question: does this report contain enough raw material to support the conclusion it is drawing?

And even when a report does contain enough material, there is one more failure layer. Fate is not decided in the press conference room — but it begins to be written there. Every answer a coach gives in a press conference is an encrypted tactical blueprint. The evasions, the silences, the platitudes repeated word for word week after week — all of them are signals. They do not reveal what the coach thinks. They reveal what he is trying to hide.

I usually handle this layer by hand. I transcribe every answer, count how often a keyword appears, and cross-check against the actual shape on the pitch in the next match. In many cases, the gaps in the answers line up with the gaps on the pitch. That is the most trustworthy kind of empty cell I have ever encountered, because it is not a pipeline error. It is intentional.

What to verify next matchday

If you want to check whether a club runs an integrity gate, do not read the metrics they publish. Read the footnotes at the end of the report.

If the footnotes name the data provider, the latest update date, the number of matches with complete data out of the total, and which metrics are missing — that club is running a serious pipeline. If the footnotes contain only a software name and a copyright line, every conclusion above it is standing on sand.

Next matchday I will be watching three things. First, how big clubs publish post-match data when tracking systems fail — whether they dare to write "insufficient data" or fill the gap with an estimate. Second, how coaches answer when asked about the gap between their lines, because the answer to that question is always a more precise signal than any published post-match table. Third, how many leaked scouting dossiers come with a provenance footnote attached.

Esports is at the stage football never had the chance to return to: being written correctly from the start. If a young industry wants to avoid football's thirty years of mistakes, the first thing to do is not to buy more data. The first thing is to write the integrity gate before writing the algorithm.

Marseille, twenty-five years after that A3 sheet. I still keep the habit of hand-counting one match a week, even though my analysis room has enough automated data that I never have to. The habit is not nostalgia. It is a test: if my hand count differs from the pipeline, I need to inspect the pipeline. If my hand count matches, I inspect again anyway, because a number that is right twice may be an empty cell learning to lie.

Technical note: the metrics xG, xGA (expected goals against) and PPDA (passes allowed per defensive action) are referenced here as examples of model limitations, not as concluding evidence. The factual anchors concerning the Champions League final of 26 May 2026, the Barcelona — Inter semi-final of the 2026-2026 season, the 2026 World Cup round of sixteen between Spain and Russia, Chelsea's January 2026 transfer window, and the Moisés Caicedo transfer of August 2026 are drawn from public match records and transfer records.