A Mislabel and the Limits of Single-Source Data

That morning, the file opened with a familiar label: "Tennis." Just below it was the headline: "Over 71 terrorists killed in Balochistan operations over 96 hours: ISPR." Not a single player. Not a single court. Not a single game. Only casualty figures and statements issued by the media wing of the Pakistani military.

I sat for a long time in front of the screen, not to analyze a match — there simply was no match to analyze. The only thing I could honestly do was trace the error: how did a security and South Asian geopolitical report slip into an analytical pipeline meant only for tennis?

This is the story of a mislabel, and of the habit I believe sits at the root of most errors in the measuring trade: we trust the label more than we trust the content beneath it.

What the report is actually about

The text originates as a security report. Its content revolves around counter-terrorism operations conducted by Pakistani security forces in Balochistan, lasting roughly 96 hours, with time markers falling in late September, around September 21 and September 25. Every data point — the figure 71, along with the groups 33, 23, 11 and 6 — is a casualty or incident count, not a match metric. There is no serve, no first-serve points-won percentage, no break-point conversion rate.

The entities appearing in the report include ISPR — the media wing of the Pakistani military — along with President Zardari, Prime Minister Shehbaz Sharif and Interior Minister Naqvi. The campaign framework invoked is named Azm-e-Istehkam. The militant groups targeted are named as the state names them, including Fitna Al-Khawarij and Fitna Al-Hindustan. Not one figure belongs to the tennis world. No tournament is mentioned. There is no ranking, no seed, no draw.

In other words, this is a text belonging to the domains of military affairs, counter-terrorism and South Asian geopolitics. Its value for any tennis purpose is nil. And I will not delude myself by translating a military report into the language of the court and calling it analysis.

The flaw lies in the labelling step

I found the flaw not in the player's body but in how we measure it. This time, the player's body did not even exist — yet that sentence holds exactly as it is. The flaw lies in the automated topic-classification step: a machine tagged the text "Tennis" while the content had nothing to do with tennis.

The consequence does not stop at one off-topic article. If this label travels into later steps, the entire downstream analytical chain is contaminated. A model built on wrong input data will produce wrong conclusions — and worse, those conclusions carry the appearance of a serious calculation, convincing enough to fool both writer and reader.

Data never lies; only the way we read it is wrong. Here, the misreading begins when we label before we verify. I once thought this kind of error happened only with medical data in sport. But classification errors are everywhere, differing only in how much damage they do later.

The single-source problem

There is one point I cannot overlook, even though the topic sits outside my expertise. Every fact in the report traces to a single source: ISPR, together with state officials who share the same viewpoint. Not one independent monitoring body is cited. Not one witness is named. Not one dissenting voice appears anywhere in the text.

In my trade, a data table coming from a single unverified source is a data table that cannot yet be used. I have repeatedly compared teams' tracking data and found discrepancies of up to dozens of percent, even when both sides claimed to provide official figures. With only one source, we must call it testimony, not fact.

The attributions about foreign sponsorship of the conflict fall into this group too. They are allegations by one party to a conflict, inherently disputed, and not an adjudicated truth. Presenting an allegation as a complete fact is a familiar distortion, and it repeats exactly how some sports reporting turns a performance metric into proof of effort.

What this has to do with sport

Before anyone calls this topic alien to sport, let me recount something from my own trade. Based on my experience watching matches, I have seen metrics such as distance covered and sprint counts packaged as a measure of a player's effort. But futile running also produces pretty numbers. A defensive midfielder standing in the wrong position, chasing the ball without purpose, can finish a match with a more impressive distance than the best player on the pitch. The metric is not wrong — the way we attach it to the concept of effort is.

As for how I feel when watching matches, drawn-out VAR reviews are shredding the rhythm of the game. A two-minute wait is enough to cool a goal that has just exploded, enough for emotion in the stands to settle before the referee points to the spot. But this is a judgment stemming from my own inclination. Others are entitled to treat accuracy as the greater priority over emotion. What I object to is not the review itself, but the way we package that time into a neutral number — easy to count, hard to argue with.

An injury is a story — but that story begins long before the player collapses. And a security report labelled as tennis also begins long before it is exported: at a labelling step no one re-checked.

When the label replaces the content

What troubles me most is not the wrongness of a single input file, but the possibility that it passes through the system without being stopped. One automated classification step assigning a wrong label was enough to drag along an entire chain of processing that should have been reserved for a completely different topic.

In my work, I have built risk-assessment models. I know the feeling of trusting a result simply because it is presented neatly. A risk model saves no one; it only tells you where to look. If the input is mislabelled, the model points you to the wrong place — very persuasively.

Prevention is not elaborate. Every classification step needs a checkpoint that cross-checks label against content before passing data onward. If the label says "Tennis" while the text is about a military operation, the checkpoint must block it. Skipping this seemingly trivial step is precisely where a data disaster begins — silently, and without warning.

What I take away

Over my years in the trade, I have been wrong many times, each time leaving a scar deep enough to make me rewrite my own method. This time, what made me uneasy was not a wrong diagnosis of a player's body, but a wrong diagnosis of an entire text's topic.

I believe in numbers that have been verified; I do not believe in labels. A label has value only when the content beneath it confirms it. When content and label say two different things, the only correct action is to stop and reclassify — not to strain to translate a military report into the language of the court and then fool oneself with an empty heap of analysis.

So, instead of a tennis commentary on a topic with no tennis, the most honest answer I can give is a warning about process. This is a security report. It should be routed to the analytical track that fits it. And the missing checkpoint at the labelling stage should be reinstalled before one more file goes down the wrong road.

I do not believe in luck; I believe in numbers that have been verified. A mislabel, unchecked by anyone, will quietly enter every conclusion that follows. And the moment we stop cross-checking label against content is the moment we lose the right to say we are analysing.