An Electric-Vehicle Story Wearing a Tennis Label: A Log of Classification Failure in the Analysis Pipeline

The filing took me less than a minute to finish, because it had nothing to do with tennis. The notice lodged with the Pakistan Stock Exchange announced that Sazgar Engineering Works Limited was bringing BAIC Group's electric-vehicle brand ARCFOX to the Pakistani market. Company name, product name, filing day — all explicit. Yet the analysis record placed in front of me carried a single domain label: tennis. My first reaction was not irritation. It was unease. How an electric-car corporate disclosure walked into a tennis analytical frame — and why catching that error sent me back to a hand-built spreadsheet nobody ever asked me to make — is the actual story here. In 2026, when I moved from a print desk to a digital-first role, I built a file nobody requested. Colleagues called it the Split-Times sheet. It held National Tennis Championship winners from 2026 onward, every Davis Cup tie since Bangladesh's 2026 debut, and a match-by-match map of the 2026 Asia/Oceania semi-final run. In the same file I logged Shirin Akter's 100m splits from Rio 2026, timing her start off broadcast video frame by frame. Nobody asked for any of it. But the sheet later became my defence, because I would not file a claim I could not match in two independent places. Now imagine that logic installed inside an automated pipeline — with the domain name wrong. Inside content labelled tennis there is not a single tennis player, tournament, ranking point, or serve-and-return figure. No court dimensions, no surface reference, no scoreline. What exists is a securities-market disclosure and an automotive brand's market expansion. The Sazgar paper, verified on its own terms, yields clean corporate chronology: incorporated in 2026, publicly listed in 2026. The BAIC relationship becomes visible in 2026, and the HAVAL and hybrid rollout begins in 2026. The notice also cites technology collaboration with Magna and Huawei. ARCFOX is the newest layer stacked onto that ladder. These numbers are corporate chronology, not competitive form curves. Running them through a "form" or "ranking-defence pressure" slot is a category error. You cannot score a ball that landed outside the baseline, and you cannot map an EV market's investment timeline onto a tennis season. I know what a real tennis story looks like. Zarif Abrar's 2026 junior title is historic by local standards and small by global standards — precisely as this filing is useful to an automotive desk and entirely irrelevant to a tennis desk. In Bangladesh we talk about Davis Cup Group V progress, ITF J30 titles, and the BKSP women's pathway; an electric car launching in Pakistan features in none of that. This may sound like a pointless detour, because in March 2026 the National Tennis Complex in Ramna fell silent and we grew used to keeping ledgers of absence. The National Championship, the Victory Day and Independence Day tournaments, the divisional meets — all cancelled. The Tokyo postponement stripped Shirin Akter and Jahir Rayhan of a qualifying window. Out of that habit came one rule: date every prediction. In June 2026, working with a Rajshahi-based stringer, I wrote that revival would come from ITF J30 junior events and school courts, not talent hunts, and gave it a five-year horizon. By 2026 it could be checked, because the date was on it. A data pipeline needs exactly that rule — and not the rule alone, but the check that goes with it. Twenty-four days in Russia in 2026 taught me one thing. VAR does not stop play; it redraws the geometry of play. Defensive lines drop deeper, the effective playing area shrinks by roughly five metres. I published that before the knockouts, and the quarter-finals largely confirmed it. A classification label behaves the same way. It does not halt the pipeline; it redraws the report's geometry. Once "tennis" is attached, every downstream stage switches on by itself — form analysis, ranking-point structure, tournament positioning, the media-narrative heat cycle. Even with no tennis in the input, the output takes the shape of tennis. That is where the danger sits, because an empty cell — "metric: not applicable" — is an honest answer. A wrong label is a false answer wearing the appearance of honesty. Suppose this item enters a large sports dataset. Coverage of a Pakistani EV brand launch joins a tennis-industry index. The picture of South Asian tennis investment distorts, exactly as Bangladesh begins thinking past the 2026 collapse toward J30 events and school courts. When I tried to match this story against the six segments of the tennis industry, I got one result. Prize-money ecosystem, Grand Slam business, agency and endorsements, capital and event investment, equipment technology, derivatives and mass market — every cell empty. There is no transmission pathway here. Those blanks are not a failure. They are the only honest answer available. An analysis afraid of writing an empty cell starts writing guesses, and guesses become indices. The reflexive response is blame: the classifier is bad, the AI got it wrong. I don't buy it. Errors of this kind rarely happen on bright boundaries; they happen where domains look adjacent. An equipment company's financial results filed under a "sports" label, or an athletics story landing on a cricket desk, could have been caught two stages earlier if anyone had wanted to. The fault is not a person's, it is a stage's. Between Stage-1 and Stage-2 there is a verification gap. Most information points list no source at all, and the date is only "Friday". If the source fields are that empty, why should the domain-label field be filled by inference? The third reason is the most uncomfortable. Sports journalists have worked off weak sourcing for a long time. If my own habit of refusing single-source stories about "TV rights" or "league launches" protects the index, the question points back at me. The fix is unglamorous. First, a domain-consistency gate: before Stage-2, check whether extracted entities intersect a known tennis-entity dictionary — players, tournaments, governing bodies. If none of the three is matched, quarantine the item. Second, a source-completeness threshold: if missing-source points exceed 20 percent in Stage-1, analysis proceeds only at low confidence and draws no conclusion. Third, a ledger. Every verified item timestamps into an immutable record — which day, which source, whose approval. The logic is the blockchain's: once written, it cannot be quietly altered; altering it forces the whole chain into view. In sports journalism I call that dated accountability. I learned it in my first year on a Dhaka desk: never file a claim without a date on it. Over years it put me in a position where old work can be checked. I will admit my own field of vision has limits. My career ran through Dhaka desks and the Ramna, Gulshan, and Officers Club circuit. Before generalising, I have to bring in BKSP girls, Rajshahi coaches, divisional meets. This error carries the same warning: step outside your own bubble and verify. So here is a dated prediction. By December 31, 2026, at least one major sports-data platform will add a mandatory domain-consistency gate between Stage-1 and Stage-2, matching entities against a tennis dictionary. My confidence is 70 percent. I could be wrong: if platforms instead try to "solve" this by simply expanding classifier training data, the prediction fails. More training data lowers the overall error rate but does not change the kind of error — boundary-adjacent mistakes survive. Whether a story about an electric car sits in a tennis file is no longer a tennis question. It is this: can the system that selects our news catch its own mistake? If it can, the game does not change — the geometry of the game does.