A 'Tennis' Label Stuck on a $40 Billion File: A Fact-Checking Lesson

Last Tuesday a file of 32 data points landed in my queue under a single classification label: Tennis. I opened it the way you open a serve-statistics sheet — first-serve percentage, second-serve points won, break-point conversion. There was not a single player inside. Not a set. Not a surface. Not a Grand Slam final, not an ATP or WTA ranking, not one tennis governing body named anywhere. What was inside was the SIFC — Pakistan's Special Investment Facilitation Council — a USD 40 billion investment pipeline, the ML-1 railway project with its financing and design still open, the K-IV water supply scheme for Karachi, and the oversight record of the National Assembly Standing Committee on Economic Affairs. I have read sports data for nearly thirty years. Files arrive late — I am used to that. Files arrive with columns missing — I am used to that. Files arrive in the wrong category altogether — that is rare. But when it happens, it always deserves a stop. Numbers never lie, but they can stay silent: silent about the very label someone stuck on them. I LEARNED VERIFICATION BEFORE I LEARNED ANALYSIS I joined Sports Illustrated in 2026 at the lowest rung of the chain: fact-checking. My job was not to write. My job was to disprove. Every number a reporter typed into a draft needed a source. Every name, every date, every scoreline needed to be checked at least twice. Six years at the Daily Mail after that taught me one more thing: written discipline begins with observation, and observation begins with knowing what you are actually looking at. That is why, twenty-eight years later, I still keep a habit many younger colleagues consider a waste of time: before analysing any dataset, I read the metadata first and the content second. Tuesday's file made that habit worth its cost again. Technically, this is a Pakistan political-economy file. Its entity roster includes the SIFC; the National Assembly Standing Committee on Economic Affairs Division; the two lawmakers Jamil Qureshi and Mirza Ikhtiar Baig; the Prime Minister's Office; the Asian Development Bank; the Asian Infrastructure Investment Bank; the World Bank; the European Investment Bank; the Islamic Development Bank; JICA; the Ministry of Planning, Development and Special Initiatives; the Ministry of Finance and Revenue; the Sindh Planning and Development Board; the Sindh Finance Department; WAPDA; and the Karachi Water and Sewerage Corporation. The sectors named are oil and gas, railways, telecommunications, and agriculture. None of those sectors is sport. None of those entities is a sporting body. I call this a label-layer failure, and it differs from a data error. A data error is an empty cell or a malformed field. A label-layer failure is a file that is correctly formatted, correctly structured, syntactically clean — and handed to a competent person for the wrong job. For a modern sports desk this is not a trivial matter. Newsrooms today ingest wire content through automated pipelines carrying domain tags. A wrong tag upstream travels downstream unchecked. The mild outcome is a tennis analyst burning two hours on a file about Sindh's water supply. The severe outcome is a sports item published under a headline about the ML-1 railway, with nobody in the newsroom noticing in time. COUNTING WHAT IS ABSENT My method for a doubtful file is not to skim for keywords. I run an entity inventory first. The inventory returns plainly: zero players. Zero tournaments. Zero surfaces. Zero matches. Zero tennis governing bodies. Serve data, return data, break-point data, ranking data: all zero. That is the most important hidden number in the entire file. Not the USD 40 billion figure. The zero. A young analyst is usually pulled toward the largest number on the page. Forty billion US dollars sounds imposing. But in my language, a large number without a denominator is decoration. Forty billion over how many years? Against what baseline? How much is committed, how much remains intent? The source does not say. And if it cannot say, the figure cannot support a conclusion — only a quotation. The financing architecture is notable in a different way. The lender list runs from the ADB, the AIIB, the World Bank, the EIB and the IsDB through to JICA. Five institutions, five procurement rulebooks, five approval timelines, all sitting on one project. In sport I have seen the consequence of that arrangement: an athlete with five coaches at once has no coach accountable for the result. The ML-1 project sits at a design and financing stage that remains unresolved. The K-IV scheme pulls in both WAPDA and the Karachi Water and Sewerage Corporation, which is a federal-provincial overlap layer. And the National Assembly Standing Committee on Economic Affairs produces opinions, not decisions. Opinions are post-match commentary. Decisions are the scoreline. This is the lesson I once paid the most expensive tuition to learn. In 2026, working as an analyst for Fox Sports Australia, I built a private dataset from 380 matches to examine Aaron Mooy, then at Huddersfield Town. Based on my match-by-match observation and note-taking across that season, his running volume reached 12.7 kilometres per match. But the decisive number sat elsewhere: 87 percent of his passes were played under high pressure. I staked my reputation on it and pushed back against the conventional view that Mooy was merely an average midfielder. That dataset worked because I defined the sample before I defined the conclusion. I chose 380 matches, chose the high-pressure threshold, chose the definition of a completed pass — and only then looked at the results. The following year I did it in reverse order. And I paid for it. CROATIA, AND THE DAY I LEARNED TO LISTEN TO DATA In 2026, under pressure to repeat a success, I published a World Cup scoreline model before the tournament in Russia began. It ran on xG, PPDA and squad-rotation volatility. The output: Brazil champions, 78 percent probability. Croatia reached the final and burned the model to ash. How you respond matters more than the error itself. I did not write a defence of the model. I wrote a series titled 'Where did the Data Monk go wrong?', analysed Croatia's six matches, and surfaced an indicator nobody was measuring: pressing transition. I burned my own model with Croatia. That was the day I learned to listen to data. Since then I write in the language of probability rather than the language of assertion. I attach confidence intervals to every judgement. And I keep a fixed section at the end of each analysis, an error log, so readers see the reasoning process rather than only the correct outcomes. Tuesday's file is a failure one layer below Croatia. With Croatia I chose the wrong variable. With this file, someone chose the wrong box to put the variables in. And that is what makes it more dangerous: a model can survive a bad input. It cannot survive a bad schema. If the domain label says tennis, then every calculation behind it — however sophisticated — is answering a question nobody asked. THREE LABEL LAYERS SPORTS DATA DESKS TRUST WITHOUT CHECKING The first layer is the domain label: does this file belong to sport at all. The second is the surface code: was the match played on hard court, clay or grass. The third is the tournament tier: Grand Slam, Masters, or something lower. Of the three, the second is checked least and costs most. A match assigned the wrong surface code drags every downstream analysis of scoring rhythm, serve performance and movement tactics out of alignment — yet it still looks persuasive, because the arithmetic never throws an error. Bad data rarely incriminates itself. It simply fails smoothly. THE COUNTERINTUITIVE ANGLE: RE-TAGGING IS NOT FIXING The intuitive response is to re-tag the file and move on. That is precisely the trap. Re-tagging treats the symptom. The real problem sits in the incentive structure. Data-processing systems today tend to measure productivity by throughput: how many files processed per hour, how many records pushed down the pipeline. No metric rewards checking whether the label itself is correct, because label-checking produces no visible output. It only produces work that did not happen. The consequence is that an entire sports desk's credibility may rest on a layer nobody owns. There is a further blind spot, and I have to confess it before I describe it. After Croatia I audited my entire private archive. I found a material share of matches carrying the wrong surface code. I had never published that number. Now I have. An analyst earns trust when people know where he has been wrong, not when people believe he never has been. WHAT THE DATA CANNOT SAY This file cannot say whether the USD 40 billion pipeline will materialise. It cannot say whether ML-1 will close its financing gap. It cannot say whether K-IV will deliver enough water to Karachi. It can only say that the relevant parties — the SIFC, the Ministry of Planning, Development and Special Initiatives, the Ministry of Finance and Revenue, the Sindh authorities, WAPDA and the Karachi Water and Sewerage Corporation — are sitting at the same table. Sitting at the same table is one event. Reaching agreement is another. Disbursement is a third. Those three events habitually get compressed into a single headline, and that is where data starts being distorted from the reader's side rather than the source's. Every passage of play leaves a footprint. The best are not those who run the most, but those who leave their footprints in the right places. The problem with this file is that it was thrown onto the right pitch at the wrong stadium. THREE SIGNALS TO TRACK, AND ONE SELF-AUDIT On the Pakistan file, three signals deserve tracking over the coming rounds. First, whether the USD 40 billion pipeline converts from pipeline status into financial close, and what the real conversion rate is. Second, whether ML-1's financing and design components close or keep slipping past their deadlines. Third, whether the opinions of the National Assembly Standing Committee on Economic Affairs become binding, or remain internal minutes. On my own desk, there is one task: put label auditing on a regular schedule, the way surface codes get re-verified after every surface swing. A mislabelled file today is a mislabelled conclusion next month. And a mislabelled conclusion already published cannot be fixed by any model. A label does not belong to the data. But it decides how the data gets read. In a pipeline where nobody audits labels, the last person to pay is always the reader.