Blockchain Data Provenance: When a Durga Puja Gown Picked Up the 'cricket_asia' Label

At 11:47 last night my ingestion pipeline told a lie. A new record arrived, domain label—cricket_asia. I opened the table. No 64-match spreadsheet, no PPDA, no xG, no set-piece xG, no distance covered. Instead: a maroon gown, a jewellery list, layers of makeup, and photo indices from 1/10 to 7/10. Actress Mandira Chakraborty's pre-Durga Puja look. No innings, no wicket, not one ball. Yet the pipeline admitted it into the cricket dataset, silently, without a warning.

In thirteen years of data work I have seen many lies—wrong xG, wrong formations, wrong football scores, goals placed in the wrong minute. But calling a festival dress "cricket" is a new kind of lie. This time the lie is not on the scoreboard; it is in the label. And the label is where data pretends to speak the truth.

I am Tamim Khan, a data journalist from Chattogram. In 2026, as a twenty-one-year-old statistics student at Chattogram University, I launched the page xG Chattogram. I built xG Chattogram because the league table was lying in plain sight. After Chattogram Abahani beat Sheikh Jamal Dhanmondi 2-1, I logged all fourteen shots by hand and assigned xG values. The result: Abahani scored two goals from 1.3 xG; Sheikh Jamal generated 1.9 xG from eleven shots. The post earned 5,200 shares. That day I understood that new media rewards verifiable numbers over hot takes.

At the 2026 Russia World Cup I built a 64-match spreadsheet—PPDA, xG, set-piece xG, distance. That 64-match spreadsheet was not a prediction; it was a confession—a reckoning of what I could not stop counting. Croatia conceded 1.4 xG per match yet won two penalty shootouts; France allowed only 0.8 xG. The "World Cup by Numbers" thread brought 18,000 followers.

In 2026, furloughed, I scraped 306 matches—Bundesliga, Premier League, La Liga, Serie A, Ligue 1—before and after the empty-stadium restart. Home win rate fell from 45.2% to 40.1%; home goals per game dropped from 1.53 to 1.26. When the stadiums emptied, the numbers did not go quiet; they changed their accent. That is when I learned to keep a control variable in every article, and to stop treating "home advantage" as a fixed cliché.

Across these thirteen years I have done one thing without pause—audit data. Every label, every source, every assumption. Because I know data does not lie on its own; the pipeline lies. And when a pipeline forgets how to recognise cricket, it starts to see any Bengali word as cricket. The word "Puja," a Bengali name, a festival date—put the three together and our classifier gets confused. That is exactly what is in front of me.

I audited the record across eight dimensions. The result is brutally clear, and every dimension stopped at the same place.

Format and match: no format exists. No Test, ODI, T20, or The Hundred. No powerplay, no middle overs, no death overs. No session, no venue, no pitch, no dew, no DLS. If a cricket article does not contain a single match, it is not a cricket article. That simple, yet the pipeline could not catch it.

Player data: Mandira Chakraborty is an actress, not a cricketer. No batting average, no strike rate, no bowling economy, no situational splits, no age curve. The only "data" present are photo indices—1/10, 4/10, 7/10—and styling attributes: gown cut, jewellery, makeup. These are not cricket metrics, and translating them into cricket metrics would be pure fabrication. I decline.

Team landscape: no national team, no franchise, no ICC ranking, no squad, no bench depth, no age structure. The geographic anchor—"Asia," "Bengal"—is cultural, not competitive.

League and commercial ecosystem: no IPL, no BPL, no PSL, no SA20, no MLC. No broadcast-rights value, no franchise valuation, no auction, no purse, no RTM, no transfer fee.

Rules and governance: no ICC, no BCCI, no ECB, no CA, no NOC, no anti-corruption unit, no eligibility dispute.

Risk: sporting risk, personnel risk, commercial risk—all zero. But the one real risk is data-pipeline contamination. A non-cricket item has entered a cricket dataset, and this is a high-likelihood, medium-impact event.

Public narrative: this is a festival-timed celebrity fashion story, tied to the Durga Puja calendar. The caption—"I don't follow the spotlight, I become it"—is a personal-brand statement, not a competitive declaration.

Industry transmission: broadcast, sponsorship, talent supply chain, betting, derivatives—no cricket channel is engaged. The only "industry" transmission is inside entertainment/celebrity media.

Every one of the eight dimensions returns the same answer: insufficient information. There is no data here, only a wrong label. And a wrong label is the most dangerous contamination, because it passes itself off as truth.

Now the real question: how do we stop this lie? This is where blockchain becomes relevant—but carefully, because blockchain by itself solves nothing.

I see a three-layer response. Layer one—a domain gate. Before content enters the cricket dataset it must present at least one recognised cricket entity: a player name, a match ID, a venue, or a tournament ID. If every decision of this gate is written to an immutable ledger, then who admitted a piece of content as cricket, when, and under which rule, can never later be deleted or denied. Today our problem is not only the error; the problem is that the error has no audit trail. A record goes in, nobody knows why, nobody can prove when.

Layer two—separate entity registries. When an actress and a cricketer sit on the same list, the entity graph is polluted. A smart contract can enforce this: entertainment entities and sports entities live in two separate on-chain registries, and cross-writing is forbidden. Then no celebrity name can ever enter the cricket entity index.

Layer three—misclassification rate as a pipeline KPI. If such wrong labels grow in volume, entity-frequency and sentiment signals will be distorted, and downstream cricket models will inherit a false positive. Blockchain transparency helps here: the monthly share of mislabelled records can be published on-chain, and anyone can verify it.

But I will stay honest. Blockchain is not magic. It does not fix a broken classifier. It only ensures the error cannot be hidden or denied. For data integrity, that is exactly what is needed—an immutable record of the lie.

There is an uncomfortable possibility I cannot dodge. Perhaps this error says less about the article and more about our model. Because an error is harmless only while it is isolated. But if hundreds of such wrong labels arrive from Bengali entertainment feeds, the problem is not the content—it is the classifier.

One more caution, part of my professional habit: correlation is not causation. The word "Puja" and cricket co-occur, but they do not cause each other. Our pipeline confused the two because it never learned to tell them apart. I learned this in 2026, when I stopped assuming "home advantage" and began measuring crowd effects in numbers. By the same logic, a geographic tag cannot be read as a topical tag. "Asia" is a geographic anchor; "cricket" is a subject. Two different layers, two different decisions.

Yet I also admit this: the blockchain fix is itself a trap. The Data Monk does not worship numbers; he interrogates them until they confess context. If I start writing every small error to the chain, my ledger will swell like spreadsheet sprawl and become unreadable. The discipline is—lead with one indictment metric, keep the rest in an appendix. In this article that metric is the misclassification rate.

I removed this record from cricket analysis. My professional verdict is clear: no match, no player, no team, no league, no governance, no commercial cricket. The right action—return it to Stage-1, reclassify it as entertainment/fashion, and treat the error as a pipeline-quality incident, not a cricket data point.

The three signals I will watch in the next round: domain-label accuracy, entity-registry separation, and clusters of festival-keyword false positives. If all three start rising together, I will know the problem is not in the content—it is in our eyes.

And one question keeps turning in my head: if a Puja gown can slip into a cricket dataset, how much of that dataset should I still trust? Numbers do not lie—but labels do. And if the label is not on-chain, who can prove which one was true?