Football Data and the Gap Nobody Verifies

The 63rd minute of a Premier League match on a Saturday afternoon. I was sitting in a newsroom in Manchester with two screens side by side: one carrying the broadcast, one running the live data board from the provider. The board stopped. The minute counter froze at 61. The on-air graphics kept ticking along: xG 1.84, 412 passes, 58 percent possession. Nobody in the room noticed the source had gone quiet nearly three minutes earlier.

When I called the programme's producer, she was blunt: "We have no protocol for when the feed stops." The system was built to run when everything works, and was never built for the moment it stops working. A gap opened up in the pipeline, and that gap was filled with the previous value, with an interpolation, or with a zero.

The context of an invisible infrastructure

Football's data industry is now infrastructure rather than an accessory. Stats Perform owns Opta. Sportradar supplies data to bookmakers and federations. Genius Sports has held the Premier League's official betting data rights since 2026. Hawk-Eye Innovations, owned by Sony, has run goal-line technology in the English top flight since 2026-14 and VAR since 2026-20. The goal-line system is usually described as 14 optical cameras around the pitch recording up to 500 frames per second, enough to establish whether the ball crossed the line in less time than a breath.

At the other end of the pipeline, data flows into in-play odds boards, into broadcast analysis desks, into club recruitment spreadsheets, into our copy. Investment in data infrastructure rises with every rights contract, and so does something else: the number of decisions made on values nobody checks again.

I entered the trade in 2026 at the Newark Advertiser with a paper notebook and a pencil. Data was something I counted myself. Thirty-seven years later I sit in a newsroom with three competing data sources on screen, and most of my working day answers one old question: where did this value come from.

Four layers, three failure modes

Football's data supply chain has four layers. Collection is optical cameras and human coders inside the ground. Transmission is APIs, satellites and fibre, timed in milliseconds. Interpretation is models, algorithms, graphics, presenters. Consumption is audiences, bookmakers, coaches, reporters like me.

What matters is that failures rarely happen at collection. Cameras do not invent. A ball over the line is recorded as over the line. The faults sit in the three layers above, and they take very specific shapes.

The most common fault lies in how a missing value is handled. When a metric is not updated, the system displays it as zero rather than leaving it blank. To a forecasting model, zero is information, and it is false information. A team that made no passes for seven minutes because the feed jammed is entirely different from a team that made no passes for seven minutes because it was pinned back. The same empty cell, two opposing meanings, and the model reads only one.

Another fault appears after two hops. A metric travels from provider to news wire, from news wire to commentary piece, from commentary piece to social media. At the final link, people cite it without knowing what method produced it, how many matches the sample covered, whether penalties were excluded. The provider never said what the metric measures; the consumer assigns the meaning. Once a metric loses its source, it stops being data and becomes a belief with a numeric format.

One more: small samples promoted to trends. Three matches become a pattern. Five become a style. Ten become an identity. Based on my experience watching matches both inside stadiums and through screens, this is where my profession and analytics collide, and both are equally weak there.

In Moscow in 2026 I followed a source out of an anti-doping laboratory. One midfielder's file showed abnormal red-cell markers across three consecutive samples. I did not publish. Over the following four months I built an analysis framework for Russian football from 2026 to 2026, cross-referencing 212 public samples against 47 competitive matches. The 4,000-word draft came back rejected for lacking direct evidence. I filed it in my own archive. Records do not lie, but people build records to speak lies for them, and I had no intention of lending a hand.

In April 2026 the stadiums shut. An accountant at Derby County handed me a bundle. I applied the Moscow framework to 18 player-linked loans between 2026 and 2026. Seven million pounds had moved through a shell company in the British Virgin Islands, matching the purchase of winger Tom Lawrence. The piece, published in June 2026, showed the club had used COVID-19 relief funds to service the personal loan interest of three directors. The EFL opened an independent investigation. Derby were docked nine points in the 2026-22 season. Football does not go bankrupt on its own. Someone stands behind the collapse, collecting the pieces.

In Qatar in 2026 I reviewed World Cup stadium construction contracts. The 3.2 billion dollar deal for Lusail Stadium carried a registered address matching a middleman company from the 2026 Russian doping case. I worked for two months, checked 86 bank transactions, then brought in a Swiss data analyst to verify the payment chain. The investigation traced 1.1 billion dollars of it to opaque investment funds in the Middle East. FIFA asked me to supply evidence. Nothing followed. From a Moscow laboratory to the turf in Doha, money does not need a passport.

Those three cases taught me one working rule: every claim carries either a money-flow diagram or a value confirmed by two independent sources. With only one source, the draft sits for 72 hours. Not because I like moving slowly, but because I have come close to being wrong.

Where football data meets money

The darkest side effect of sport's digitisation is that in-stadium live data is sold to betting companies, and it is not discussed enough. The same pipeline that feeds on-air graphics feeds in-play odds boards. Latency becomes a tradable asset. Whoever receives the data half a second earlier can take a position before the market adjusts. The industry calls it latency arbitrage; it is really a race whose winner does not need to understand football, only fibre optics.

In that race, a missing value is processed instantly. Nobody waits for data to return. The model must run. And when it must run, it takes the last value, interpolates, or defaults to zero.

There is another spot in the data few read closely. The five-substitution rule restructured matches: deeper squads gained an edge, and the final twenty minutes became a war of attrition. Yet those same twenty minutes are where models are weakest, because every substitution breaks the comparison sample. A side that throws on three substitutes inside ten minutes creates a configuration that has never existed in historical data. The model still returns a value, because it is programmed to always return a value.

The same happens with a player returning from injury. Demanding that someone just back from rehabilitation prove themselves in their first match is cruel, and it raises the risk of re-injury. Data does not see the cruelty. Data sees minutes played, and the minutes of a returning player are placed beside the minutes of a fit one as though the two were comparable.

The other side of the argument

Fairness requires this. The major data providers are usually far more rigorous than their reputation suggests. They run verification processes, quality-control layers, trained coders, incident logs when systems fail, and they publish missing-data cases rather than inventing a value. Technically, when a field returns empty, stating that it is empty is the correct behaviour, and in many cases it is the most honest conclusion available.

The problem sits at the consumption layer. A graphics board cannot sell a blank. A betting model cannot price a blank. A commentary piece is not compelling if its conclusion is "we do not have enough data to conclude". The whole system is under pressure to fill, and that pressure runs against the professional ethics of the layer below. The coder in the ground does the job right, and the right value is then distorted at the third layer by someone who needs something to broadcast.

I have been on the other side of that pressure. In 2026, when I could not publish the Moscow file, I knew I had done the right thing, and I also knew I had failed by newsroom standards. Both were true at once. An empty dataset can be a finding. But only if readers are allowed to see it.

What should happen next

The standard I want to see across football's data industry is a null certificate: when there is no verified value, the system displays exactly that, and the editor carries the responsibility rather than the provider. Clean is not the same as transparent. One is a scent, the other is double-entry bookkeeping. In a major tournament season, when emotion is compressed into each knockout round and every eye lands on a single fixture, an honest blank is worth more than a filled-in value.