Not One Ball in Seventeen Points: Label Failure in Sports Data Pipelines and the Unfinished Ledger of On-Chain Provenance

I counted the pages. Seventeen information points, numbered one through seventeen. In Moscow I had twenty-one pages and ninety-eight sample IDs; there, the numbers counted themselves. Here I had a single spreadsheet row with a domain label reading “football.” I read all seventeen points. No ball. No club. No coach, no transfer, no match, no governing body.

What was there: Tinder's new feature, Group Hangouts — a tool for matching with a group of friends. A dating-app product announcement that had walked into a sports data pipeline and never knocked on the door on the way in.

I am not alleging that anyone wrote a falsehood. I am saying that nobody read the item before assigning the label. And that is the only genuinely football-related fact in the entire file.

Context: how labels disappear inside a festival

A tournament cycle is a strange machine. When national-team fervour peaks, sports verticals must swallow thousands of items a day to feed the appetite. Humans do not do the swallowing. Feeds, tags, automated classifiers and a taxonomy everyone has agreed to call “sports” do the swallowing.

The problem is not the taxonomy. The problem is that every item gets a label stapled to it on the way in, and nobody ever looks at that label again.

From years of watching matches and counting documents, I have learned one thing: a wrong label is not the damage. The damage is having no mechanism to catch a wrong label.

I learned that in 2026 while working the Bramley-Moore Dock files. Fourteen Freedom of Information requests, forty-seven pages of contracts, 3.2 gigabytes of planning emails, nine payments to a consultancy linked to a club director, six hours of council meeting tapes, and a twenty-two-vote timeline. At the end of it: the council had waived £8.2m in survey fees. A 6,000-word longform went out.

That work left me two habits. One: hash every file, log every page count. Two: stop relying on club press officers — because a press officer does not hand you information, he hands you the shadow of information.

In Moscow in 2026 those habits paid. Twenty-one pages of WADA correspondence, ninety-eight sample IDs, FIFA medical staff lists — and a finding: seven IDs had been cleared by a single doctor. I built a ninety-eight-row spreadsheet and checked each ID against three databases. It ran before the final. In Moscow the paper trail was short; the silence was long.

In 2026, during the global hiatus, I audited twelve Premier League clubs' pandemic accounts. More than £87m in related-party loans from eleven offshore lenders, spread across sixty-three line items. Matchday revenue drops: Everton £24m, Arsenal £39m, Manchester United £61m. I stopped using club-supplied figures and cited only filed accounts.

This time I ran the same checklist — not on a transfer, but on a content item. The result is uncomfortable.

Core: nine rows, nine identical verdicts

My audit sheet has nine rows. Tactical analysis, club finance and the transfer market, results and the public-opinion cycle, league landscape and team positioning, rules and governance, management and dressing room, risk profile, media narrative, and industry transmission.

Every row returned the same verdict: insufficient information.

On tactics I look for formation, xG, PPDA, possession patterns. None present — because tactical analysis requires at least one team. On finance: no transfer, no wage bill, no net debt, no FFP or PSR figure. The only commercial fact is that the feature is free to use.

On results and public opinion, the sample is zero. Zero matches. No standings, no recent form, no fixture pressure. On league landscape: no league, no club, no tier.

The governance row gives the clearest answer. It contained an eighteen-plus age gate and platform community guidelines. Those are terms of service. They are not FIFA, UEFA, IFAB or league regulation. The dressing-room row had no owner, no sporting director, no coach, no squad — so there was no ecosystem to assess.

One row out of nine stayed alive: media narrative.

It stayed alive because the item is nothing but narrative. According to Tinder's own announcement, Group Hangouts is in a limited testing phase in specific markets. Participation requires a group of at least three, invitations are sent manually, and meetups can run to fifteen people. The feature is free, but it does not work alongside premium features. Terms may change later — the announcement says so itself.

Here is my first real finding. Every one of the seventeen information points traces to a single source: Tinder itself. A press release is not a document set. A press release is a claim. And I never give a claim the same weight as a dataset.

In Moscow I worked to a rule: any consequential claim needed three databases behind it. Here there was one announcing party.

So how did the label land in that row? I see three plausible paths. First, keyword collision — “match” belongs to the vocabulary of both sport and dating apps. Second, feed-tag collision, where an aggregator's lifestyle bucket spills into the sports bucket. Third, plain taxonomy laziness: under live-content pressure, nobody read it twice.

None of the three is a conspiracy. All three are process failures.

What would a proper chain of custody look like? My template has six fields. The item hash. The ingestion timestamp. The source tier — first-party, second-party, or independent journalism. The label attestation. The reviewer identity. The version number.

Five of those six fields sit empty in almost every sports pipeline today.

This is where the blockchain question arrives.

I do not treat blockchain as an expensive religion. I treat it as a timestamp machine. The problem this incident exposes is simple: who assigned the label, when, and did anyone quietly change it afterwards?

An on-chain label ledger can answer exactly that. Hash every item at ingestion. Bind a batch of hashes into a Merkle root. When a label changes, the change is not erased but appended — with a full record of who changed it and when. With a shared ledger between two organisations, one party can no longer silently retag.

This is not speculation. In content provenance, C2PA — the Coalition for Content Provenance and Authenticity — launched in 2026 with Adobe, Arm, BBC, Intel, Microsoft and Truepic, aiming at precisely this kind of origin signature. The EU AI Act, Regulation (EU) 2026/1689, entered into force on 1 August 2026, with transparency obligations phasing in. The industry is already walking in one direction: systems that make decisions may be required to keep accounts.

Why should sports data pipelines stay behind?

One number is worth putting down here, and I will be explicit — this is arithmetic illustration, not measurement. Take a football vertical ingesting five thousand items a day. If the mislabel rate is 0.1 percent, five items a day enter the wrong room. More than 1,800 a year. Those items are not merely read; they enter retrieval systems, summarisation models, training sets. A wrong label is created once, but it repeats a thousand times.

The spreadsheet did not accuse anyone. It only refused to forget.

Contrarian: what the critics miss

The first reflex is always the same: it is one tag, one error, fix it and move on.

That reflex skips the actual problem.

The actual problem is not the wrong label. It is that nothing inside the pipeline could say, on its own: there is no ball in this item.

And the more uncomfortable truth is that the error was caught by a human who read all seventeen points from start to finish. The very manual labour that scale was supposed to eliminate remains the only safeguard. If a system cannot detect its own silence, it is not an intelligence system — it is a filing cabinet with a marketing budget.

The second thing blockchain enthusiasts miss matters more.

An on-chain ledger can prove who wrote “football,” at which second, and whether anyone altered it later. It cannot prove the entry was correct. An immutable ledger of wrong labels is nothing more than a permanently wrong ledger. Garbage on-chain is notarised garbage.

There is another gap. Since all seventeen points trace to Tinder itself, a first-party source can simply attest its own claims on-chain. Provenance then becomes a laundry: clean on paper, unrelated to truth.

Third, immutability has a legal edge that sports data teams routinely ignore. Data-protection rules give people a right to erasure; blockchain gives a ledger that refuses to forget. How do you apply a right to be forgotten to a book that never forgets? That question is still open.

And the most dangerous items will never be caught. Not this Tinder entry. The ones that are ninety percent football and ten percent something else — they never lose their label, because they fit well enough. There is no way to catch them unless you read every row yourself.

Which returns me to an old habit.

Takeaway: a demand for accountability

Three demands. First, publish the source tier alongside every item — first-party, second-party, or independently verified. Second, require a second reader's signature before any domain label is changed. Third, stop treating label assignment as a default; treat it as a signed act, with a time, a name, and a liability attached.

Blockchain can make the first two technically easier. The third requires an older journalistic virtue that no chain can supply: if someone writes a claim, at least one human being reads it.

The dock files were not hidden. They were just never read.

The question now sits exactly where it sits for sports data: if a pipeline cannot tell a dating app from a football match, who audits the rest of the file?