The Wrong Label and the Empty Dictionary: How an EV Filing Landed Inside a Tennis Dataset

On Friday a notice was filed with the Pakistan Stock Exchange. Sazgar Engineering Works Limited announced it is bringing BAIC Group's electric-vehicle brand ARCFOX to Pakistan. The document carries the company's incorporation in 2026, its listing in 2026, BAIC's entry into the country in 2026, and the HAVAL brand and hybrid rollout of 2026 — every milestone dated. It also notes technology partnerships with Magna and Huawei.

There is no tennis anywhere in that paper. No player, no ranking, no draw, no court, no serve-return data.

Yet the document entered an analytics pipeline carrying a tennis domain label. Seven analytical slots — technique, form, tournament system, tour landscape, governance, team management, risk — each returned insufficient information. The reason is not mysterious. Every entity in the paper — Sazgar, BAIC, ARCFOX, Magna, Huawei, HAVAL, PSX — falls outside the tennis lexicon.

For nine years I have been trying to build a dictionary for a sport whose dictionary was never built. A rotator-cuff injury taught me that pain is unstructured data waiting for a schema.

Context: the document with time, the sport without it

The PSX filing is itself a clean record. A corporate announcement contains what a corporate announcement should — the year of incorporation, the year of listing, the year of brand entry, the year of production rollout. Who did what in which year, laid onto an append-only timeline.

Now look at tennis. How long is the documented timeline of the sport in Bangladesh? The National Championship launched in 2026. The Davis Cup Asia/Oceania semi-final came in 2026. Then roughly three dormant decades. In 2026, Zarif Abrar's ITF junior J30 title — the country's first ITF junior crown. Beyond about six verifiable names, the dataset is effectively closed: Khaled Salahuddin, Sree-Amol Roy, Shibu Lal, Ranjan Ram, Jonathan Mridha, Zarif Abrar.

I logged 32 matches of the 2026 National Championship at the Ramna complex by hand — serve percentage, unforced errors, break-point conversion. That data showed the champion won just 54 percent of baseline rallies but 78 percent of net approaches. I built my first database because memory alone could not carry the weight of a season.

The word schema may sound excessive here. In data governance it simply means a written definition of what belongs to a dataset and what does not. A domain-consistency gate does exactly that — it intersects the entities extracted at Stage 1 against a known tennis dictionary. Zero overlap means the item is held back.

It was not held back.

Core: the test nobody failed, because there was no test

Start with a membership check. How large is the intersection between the entity set pulled from the filing and the known set of tennis entities? Zero. A match rate of zero. The classifier still called it tennis.

There are two ways to read that. One: the classifier is broken. Two: the classifier is working, but the reference list placed in front of it is so thin that it cannot stop a false positive.

The analysis itself raises two possible origins — a keyword collision, or a copy-paste error at the labelling stage. Confidence is medium. I will not stand that inference up as proof. What I can stand up is the consequence.

Schema design obeys a simple rule: the denser the reference dictionary, the shorter the lifespan of a false positive. Imagine a tennis dictionary with four hundred ranked players, each with timestamped match records, each tournament tier documented. An EV press release bounces at the first check. But when the dictionary holds six names and the timeline holds a three-decade hole, the contest is not one of admission but of rejection — and there is no evidence with which to reject.

The sample size needs stating. The verifiable domestic player pool is roughly six names. No trend can be inferred from that. What can be done is description rather than inference, ranges rather than point estimates. An analysis that does not print its n is itself a false positive.

Source asymmetry: a regulator on one side, memory on the other

Source quality points the same way. Information points 5 through 13 mostly leave the source field blank. Yet the document that acquired a tennis label was produced under legal obligation, timestamped, and verifiable by anyone.

Here is the real contradiction. A new EV brand entering a national market gets recorded on a stock exchange, because capital is involved, shareholders are involved, a regulator is involved. Bangladesh's biggest tennis achievement — a first ITF junior title — gets recorded in a news report, not in a registry. Six years later, if someone asks on what date, at which event, for how many ranking points, the answer has to be retrieved from memory.

The asymmetry is not only documentary. It is attentional. Sponsors follow television; television ignores tennis. The result is no durable data infrastructure for the sport. BKSP's girls keep sweeping domestic events — that streak does not land in a table; it lands in a report. A news article and a dataset are not the same object.

Seven analytical slots came back empty. Some would call that a pipeline failure. I call it an honest result, because trying to fill the empty cells would have produced fabricated information. But beside those empty slots sits a systemic risk the analysis itself rates high: contamination. If the document is aggregated into any tennis-industry index, the index distorts. And the distortion will not be caught, because there is no reconciliation rule there either.

The core idea of a blockchain applies here without the chain: a record that is append-only, timestamped, and visible to everyone. Tennis needs that record, not the technological sheen.

Cross-sport lesson: football built the dictionary before it lifted the lantern

In 2026 I logged xG and PPDA for all 64 matches of the Russia World Cup, and before the final I wrote that France's real strength was not Mbappe's speed but their 0.7 xGA per match. France won 4-2. In 2026, reading Morocco's PPDA of 8.3, I wrote they would reach the semi-final. They did.

People read that as predictive power. It is not prediction; it is a by-product of a dictionary. Expected goals are not prophecy; they are a lantern held against a dark stadium. Before lifting the lantern, football had built its dictionary — who plays where, how many shots per match, all of it documented. In tennis we want the lantern without building the dictionary.

The contrarian read: state the plain hypothesis first

Let me put the ordinary hypothesis on the table: the classifier made a mistake, so fixing the classifier ends the problem. That is probably true, at least in part. Change the label, remove the document, done.

But the real question arrives after that. In a pipeline that ingests many items daily, why did a single false positive fail to surface on its own?

The answer is not inside the classifier. It is inside the reference data. Strong pipelines fail safe — when in doubt, they hold — because their power to reject is high. Tennis's pipeline does not have that power. A wall built from six names and a three-decade gap lets a stray press release walk straight through.

The analysis recommends installing a domain-validation gate. Reasonable. The question that remains is ownership: which institution maintains the tennis entity dictionary, who takes responsibility for it, who reviews rejected items? Without an answer to that, the gate is just another file.

Forward signal

There is nothing to be gained by staring at the quarantined document. News from Pakistan's EV market belongs to an automotive analyst, not to tennis.

The thing to watch is different, and the window is short. Suppose Zarif Abrar, or one of the BKSP girls, posts another junior result in the next six months, or plays another J30 draw. If it is not recorded with a timestamp and a source, the same failure will recur next year — and it will no longer be a classifier's failure.

It will be another instalment of the same shortfall that has cut across this sport for three decades. Changing a label is easy. Building a dictionary is hard.