When a Sanitation Report Gets Tagged as Football: Bad Data and the Deals That Are 'Almost Done'

At 6:12 in the morning, Marseille time, I opened my feed. Between lines about a striker negotiating an extension and a full-back returning from a hamstring injury, one item carried the tag "football". Its content: "Work has begun to establish a Suthra Punjab-style sanitation system in the civilian areas of Rawalpindi and Chaklala." I read it three times. No club. No player. No competition. Not a scoreline, not a minute of play, not a formation diagram. The only named individual was a member of the National Assembly. The body referenced was a cantonment board. The unit of value in the story was a grant that had not yet been approved.

What made me stop was not quite the absurdity of the label. What made me stop was that I had read it exactly the way I read a transfer story.

Context

Let me be precise about the report. MNA Malik Abrar Ahmed, of PML-N, sent proposals to the chief secretary of Punjab. The cantonment administrations of Rawalpindi and Chaklala were said to be taking necessary measures. A special grant would be sought from the provincial government. The stated window: four months. The amount: absent. The budget line: absent. The approval status: absent. The sourcing: one legislator, meaning the direct beneficiary of the announcement itself, plus a few unidentified "sources" indistinguishable from one another.

The central verb of the whole report is "will seek". The satellite verbs are "have begun", "is expected", "reportedly".

This is the grammar of intent, not the grammar of outcome. And that grammar is football's mother tongue.

I have spent most of my career reading press conferences. Fate is not decided in the press conference — but it begins to be written there. The same logic applies to a legislator in Rawalpindi and a sporting director in London: both speak before there is anything to say, because speech is the only instrument they are allowed to use for free.

So where does the gap actually sit?

Analysis

The lethal gap in every sports report sits between "in progress" and "done". Both are written in the present tense, but they belong to two different data universes.

Based on my experience watching matches and transfer windows, I divide the state of a deal into four layers, and I do not let myself blend one into another.

Layer one, interest. A club calls an agent. No legal consequence, no financial consequence, no binding element.

Layer two, negotiation. A figure has been quoted; no figure has been accepted.

Layer three, agreement. An offer has been accepted by the selling side on paper. This is the most dangerous layer, because it looks so much like layer four that newsrooms routinely skip the distinction.

Layer four, registration. The player has passed a medical, the contract has been signed, the paperwork has been filed with the governing body. Only this layer is binary. One, or not one.

In August 2026, Liverpool and Brighton agreed a fee of one hundred and eleven million pounds for Moisés Caicedo. Leading outlets reported it as though the deal had closed. Within less than two days, Chelsea agreed one hundred and fifteen million pounds and the player signed for Chelsea. Those "done deal" reports were wrong at precisely layer three, where the event had no binary quality at all.

Six months earlier, in January 2026, Caicedo himself posted a message on social media asking to leave Brighton. Arsenal submitted a bid; it was rejected. The player stayed. A public declaration from the subject himself — the strongest evidence a supporter can imagine — did not translate into a transfer event.

In that same window, Enzo Fernández left Benfica for Chelsea at one hundred and six point eight million pounds. The deal only completed on the final day of the window, after a release clause was triggered. Before that came weeks of negotiation, weeks of speculation, and a volume of coverage in perfect inverse proportion to the probability of success at any given moment.

That is layer three and layer four: one signature apart, one truth apart.

Back to Rawalpindi. The sanitation report has the identical structure of a layer-two transfer story. Proposals have been sent. Measures have been pledged. Money will be sought. A four-month window is stated with no costed plan standing behind it. Only when the grant is notified in writing and the tender is opened does the matter jump from layer two to layer four.

A phase with no deadline is a phase with few events but a great many words. A deal with no closing date produces one article. A deal with a closing date produces thirty. Coverage volume does not measure probability; it measures time pressure. Transfer literature swells in the final ten days of January and the final day of August, not because football has more events, but because the clock has more ticking.

Sourcing behaves the same way. The Rawalpindi report rests on two types: a legislator who is a beneficiary, and unnamed "sources". In football, the equivalents are agents and clubs — both with direct valuation interests. When the only named source is also the beneficiary, I downgrade credibility by two notches, regardless of how many times the information is repeated across outlets.

Unnamed "sources" do something else: they manufacture the illusion of multiple sourcing for a single origin. One party releases a piece of information, four outlets copy it, and the reader sees four confirmation points on the credibility map. In reality there is one point of origin.

I want to place a counter-example beside this, so the standard is visible.

In 2026, I rewatched the Champions League final between Porto and Monaco eleven times in three days. Porto held forty-three percent of the ball and created five goalscoring chances; Monaco created one. The match ended three nil. When people look at Porto 2026 and see a miracle, I see an equation waiting for its solution. Miracle is the name we give to the part of the data we have not yet read.

The difference between a Porto report and a layer-two transfer report is this: Porto had numbers. A proposal sent to a provincial chief secretary does not.

I want to add a methodological note, because it bears directly on how sports data is commercially mined. Live data feeds supplied to betting companies are the darkest side effect of the digitisation of sport. A report with no amount, no approval date, and no implementation evidence, but with enough verbs to generate odds — that is exactly the kind of content any system should block at the gate.

The counter-intuitive angle

The first reaction will be to blame the classifier. A sanitation article slipped into the football feed. Fix the tag, done.

I think that misreads the centre of gravity. The classifier learns from the data we poured into it. If millions of sports sentences are written in the grammar of intent — negotiating, close to extending, will submit an offer, reportedly interested — then a machine-learning system will learn exactly one thing: intent is content. The sanitation report in Rawalpindi entered the football feed because it was written like football.

The execution blind spot is not in the model. It is in the fact that we taught audiences that "in progress" is news. We did that for twenty years, and we did it consistently enough that a small machine with no football knowledge could imitate it successfully.

Put another way, this error is a mirror, not a technical fault.

That leads to a professional consequence I know will not be welcome. Our editorial standards are currently lower than our classification standards. We demand that an algorithm be accurate down to the individual entity, while we ourselves publish articles in which the central entity — the contract — does not yet exist. When the machine errs, we call it a system failure. When we commit the same error, we call it breaking news.

Takeaway

So what needs fixing?

A gate rule: if a report contains no football entity at all — no club, player, competition or match — it does not belong in the football feed. This rule is cheap, automatic, and would have blocked both the Rawalpindi report and hundreds like it every month.

A second rule, more expensive: every forward-looking claim must carry its tense. "Will seek" is layer two. "Has been approved" is layer four. Readers have a right to know which layer they are reading, just as they have a right to know whether an odds line was computed from real data or from a headline.

A third rule, impossible to automate: when the only named source is the beneficiary, write it that way. Not "according to sources", but "according to the legislator who proposed the project". Not "reportedly", but "the person supplying the information is the person who will claim the credit if the project succeeds".

How will I verify this in the next match? In exactly the way I verify a transfer. I will track the Punjab provincial grant notification, and I will wait for the first tender in Rawalpindi or Chaklala. If after four months there is no notification and no tender, I will know that report was layer two from beginning to end — and I will know that the "football" tag above it was, in truth, more accurate than it appeared.

Collapse is not the end of the tunnel. It is the largest dataset life provides. A misclassified system is the same. It is a free measurement of the quality of the language we produce every day.

The transfer market is a market of hope, and hope rarely follows valuation. Sanitation reports are no different. The only difference is that in the transfer market, we have had twenty years of practice in recognising it.