A Cricket Label, A Barrel of Petrol: The Quiet Rot Inside Scouting Data

It was 2:40 in the morning. The feed was open on my phone and one item carried the tag cricket_asia. I opened it because I was hunting for a scorecard from Pakistan's domestic circuit, or maybe a clip of an Under-19 spell. There was no cricket inside. There was petrol inside. Twenty-two days of petrol stock, thirty days of diesel, a September import plan of 511,000 tonnes, Ogra's daily ex-depot price notifications, and a written reply from Petroleum Minister Ali Pervaiz Malik in the National Assembly. Not one player. Not one team. Not one format. Not one venue. I found it at 2 a.m., somewhere between a buffering video and a hunch — except twelve years ago the screen held a fifteen-year-old left-hander, and tonight it held an energy policy file that an automated classifier had decided was cricket.

I went looking for a timestamp, the way I always do, pausing every frame to ask what time it is. The file said: figures as of 22 September. It did not say which year. The document was silent about its own age. An item that had been granted cricket status could not tell me how old it was.

The numbers inside were immaculately arranged. Monthly import cargoes — 395,000 tonnes, 506,000, 568,000, 511,000, 692,000, 262,000. Petrol sales up 18 per cent year-on-year in July, then down 9 per cent month-on-month in August. Price-differential claims worth Rs120.3 billion against Rs53.8 billion actually disbursed. Prices pegged to Platts' seven-working-day rolling average Arab Gulf FOB. For an energy desk this is clean, competent reporting. For a cricket desk it is zero.

The real story is how the file reached that slot at all. Feeds like this run on automated classifiers that key on geographic and topical tokens. There is Pakistan. There is Asia. There is a subcontinental news source. So the label settles: cricket_asia. Nobody asked which entities were inside. There is a minister, a regulator, a legislature, an MNA named Sharmila Faruqui asking a question. Not one of those belongs on the entity-type list of a cricket database.

The pipeline's real failure is not that it misread one file. It is that it shipped the label without a confidence score attached. A label with no number behind it stops being a claim and becomes a habit — and habits are what editors stop arguing with.

I have a cheap fix for this, one I have used in scouting reports for years: a cricket label should only attach if at least one of four entity types is present — a player, a team, a format, or a board. None of the four, no label. It is such a small rule that writing it down feels embarrassing, and so essential that a dataset without it is worthless.

The document that looks most analysable is the one that does the most damage. This file was dense with percentages, days of cover, month-on-month deltas. That is precisely the shape that pushes a language model into extracting false statistics. An 18 per cent rise and a 9 per cent fall read like a form curve to anyone skimming. They are consumption figures, not runs or wickets. Dropping them into a player's performance graph is a category error, and a plausible-sounding one.

Two words in that file should stop any analyst. The first is "abnormal sales". Ogra's field teams, working with provincial law enforcement, were investigating dealers whose retail volumes had suddenly spiked. That is fuel hoarding, plainly. But in betting-market vocabulary, "abnormal sales" points towards fixing. A careless tagger turns it into an integrity alert. The second is "audit" — the disbursed subsidy is to be audited, with payments staged. Reading that as a compliance breach is equally easy. There is also a naming coincidence that amuses me and should terrify a data team: Ogra and the ACU — a Pakistani energy regulator and the ICC's anti-corruption unit. The word shapes are nearly identical. It is a dictionary-level mis-tag hazard.

And then the point that outgrows a single file. If this record is one of many, one bad label puts the whole batch's labels in question. A single error is not a system fault; it is a fault in how much the system is trusted.

Here is where I have to argue with my own trade. The easy accusation goes at the machine — the classifier is stupid. I think the pressure arrives from the other direction. A desk needs cricket copy around the clock, the slot cannot stay empty, and when filling the slot becomes the job, a human does exactly what the classifier did: takes whatever is available and sticks a label on it. Lockdown taught me that distance is a layer, not a wall. A wrong tag works the same way — not a wall, a layer, and truth gets buried under the layers.

The most overlooked part is this. If this file occupied that slot, where did the story that belonged there go? A report from a Ranji or Quaid-e-Azam Trophy match, a note on an Under-19 spell, an update on a fast bowler back from injury. That record is missing from the batch and no alarm sounded. The most expensive things in cricket data are the ones that were never recorded at all.

A feed is not a mine, it is an excavation site — and every excavation throws up soil along with the find. The question is never whether soil came out. The question is who called it gold and locked it in the vault.

Next time an item tagged cricket_asia opens on your screen, do not begin by asking what is inside it. Begin by asking who attached the label, and whether they ever opened the file.