The 'No VAR Errors' Premier League Table and the Model Error Nobody Audited

Minute 74 at the City Ground. Neco Williams put the ball over the line, the assistant's flag stayed down, the home end erupted. VAR intervened. The goal was disallowed for handball. Ten days later a different table spread across social platforms: Nottingham Forest up five places and two points, Brentford lifted to third on a penalty that should have been awarded against Leeds, and Tottenham bottom with nothing from three matches. No governing body published that table. It was assembled from supporter votes.

I spent two evenings tracing how it was built. The first task was separating the hard data from the opinion, because the two are wrapped in the same visual form.

VAR has operated in the Premier League since 2026/20. The resentment cycle around it has never fully faded, and figures of the stature of Roy Keane and Wayne Rooney have now publicly called for abolition. What is new in 2026/27 is the format: contested decisions are put to a public vote, the vote rewrites the result, and an alternative table is generated from there. Formally, it has columns, numbers and standings. Substantively, the input is a poll.

That is where I want to stop longest, because everything downstream depends on it.

Data does not judge anyone, it only exposes.

Start with the verifiable layer. Arsenal and Manchester City won all three matches, and neither was party to a disputed decision. That is the most robust result in the entire dataset, precisely because it rests on scorelines rather than ballots. At the top, the real table and the corrected table coincide. That overlap serves as a control group: where there is no controversy, the voting method introduces no distortion.

Everything else is different.

Manchester United sit in the bottom half of the real table but reach the top ten in the corrected one. Notably, they gain no points at all. Their position improves because others are docked. In a league table, that kind of movement only happens when the gaps between clubs are compressed to an extreme degree.

This is where a spatial model is needed. Picture the table after three matchdays as a series of nodes hanging on a single vertical wire, each node a club, the distance between adjacent nodes one or two points. Pull one node up by two points and the node alone does not move: the whole lower segment bunches, and five nodes can swap places at once. Nottingham Forest climbing five places off one disallowed goal is not evidence that they are systematically undervalued. It is the geometric consequence of a table that is still far too short. The same distortion at matchday 30, once the division has stratified into clear tiers, would move a club at most one place.

At the other end of the wire sits Tottenham. They spent over £300m in a single window, and in their first three matches they did not score. On outcome evidence alone, that is the signature of a new attacking unit that has not yet established shared movement patterns. Roberto De Zerbi is understood to be in charge, and his tactical profile belongs to the possession-control family, which demands positional discipline and rehearsed combinations. Systems of that type typically carry an above-average integration cost early in a project.

But I have to stop here, and I want to be explicit about why.

There is not a single expected-goals figure anywhere in the dataset. That means we cannot distinguish a side that creates nothing from a side that creates and does not convert. Those two problems have opposite remedies: the first is a structural failure in build-up, the second is a finishing and variance issue. Mislabeling one as the other produces the wrong coaching decision. A tactical system only survives until it meets a larger system — and here the larger system is the absence of data.

Three matches is a statistically negligible sample for judging a transfer portfolio. I log it as a watch item, not a verdict.

Transfers do not buy players, they buy the probability of success. When a club concentrates capital into a single window, it compresses all integration risk into a few months. Every early negative result amplifies media pressure, and on the accounting side, the book value of long contracts begins to detach from the value implied by performance. The risk is real, but it only materialises if the negative run persists, not after three matches.

Now the counterintuitive part, which I consider the most important.

The voting method only weighs incidents that were already disputed. Correct decisions that nobody questioned are excluded from the sample by design. This is selection bias in the strict technical sense: the dataset does not measure officiating accuracy, it measures supporter anger. On top of that, fans of the disadvantaged club tend to vote in greater numbers and with more intensity than neutrals. The entire bottom section of the corrected table rests on a single decision, and that decision is a dispute about the intervention threshold.

This is the decisive distinction the table format erases. A factual error, such as a miscalibrated offside line or a misidentified player, is checkable with tools. A dispute over whether VAR should have intervened is a judgement call, and supporter ballots are structurally unsuited to adjudicating that layer. If most of the corrected outcomes belong to the second category, the new table is not a correction. It is a second opinion, weighted by crowd sentiment.

What frightens me is not error, it is a wrong model. Error sits inside a margin and can be measured. A wrong model produces something that looks right, with the full set of columns and numbers, and then gets screenshotted and spreads faster than any methodological explanation.

I also have to record one thing. Several details in this dataset sit outside the competitive landscape I can verify: Coventry City and Hull City appearing in the top flight, De Zerbi at Tottenham. They may be a forward-dated simulation, or an editorial construct for entertainment purposes. I flag them as data requiring verification, and I build no conclusion on top of them.

On an empty ground, I hear the breathing of defenders and the cracking of a tactical plan. Here too, what I hear is not the shattering of a league table, but the cracking of a method.

The next matchday offers a far cheaper and fairer test. Instead of asking supporters which team deserves points, build a log of every disputed decision across three columns: the evidence on the footage, the intervention threshold cited, and the outcome that followed. After ten matchdays there will be enough depth to speak about VAR's effect on the table. As for Tottenham, I will wait for their expected-goals numbers before saying anything about the attack.