The Royal–Thomian Water Polo File Was Labelled Football: A Classification Defect and Its Chain of Custody

The Royal–Thomian Water Polo File Was Labelled Football: A Classification Defect and Its Chain of Custody

I found the first contradiction in a document no one had requested. No federation sent it to me, no club leaked it, no whistleblower pulled it from a locked drawer. It was the output of a content pipeline—a system that ingests sports writing, tags it, and files it into a database. The domain field said one word: football.

The subject was something else entirely. The Royal–Thomian Annual Water Polo Encounter 2026, Under-15, First Leg. Royal College, Colombo against S. Thomas' College, Mount Lavinia. Water polo: the sport of World Aquatics, not FIFA, not IFAB. Seven players a side, six-on-five man-up play, a code in which the phrase "pass completion" does not exist.

I watched the highlights twice. From years of sitting through match footage, I can tell you this is not reporting—it is promotion. There is no scoreline overlay, no exclusion count, no man-up conversion rate, no save percentage. Only goals, saves and "key plays." But I stopped for a different reason. That two-word gap—"football" and "water polo"—told me that everything travelling with the file into that database would travel in the wrong direction: every number, every tag, every downstream decision.

Context. The Royal–Thomian sporting relationship is one of the oldest institutional rivalries in Sri Lankan school sport; in cricket it is called the Battle of the Blues, but it runs through football, rugby, athletics and aquatics too. The water polo fixture carries a multi-year continuity. The Under-15 band means players aged roughly thirteen to fifteen. Sri Lankan school aquatics is amateur: money comes institutionally, not from broadcast rights and certainly not from a transfer market. It is subsidised by schools, parents and old boys' associations. That context matters because it tells you which questions are meaningful and which are theatre.

The second thing to understand is what a pipeline does. It reads, identifies a subject, and assigns a label—football, cricket, basketball, tennis. That label then decides where a piece routes, which advertising sits beside it, which audience segment is served, and which analytical dataset absorbs the count. The label is not decoration. It is a routing decision and an accounting entry.

A wrong label never fails loudly. It fails silently. Nobody catches it, because nobody cross-checks a school water polo highlights reel.

Where the label came from. The article carried football-shaped tokens: match, goal, save, first leg, key plays, highlights. To a keyword classifier these read as a football fingerprint. "Goal" appears in football and in water polo, hockey, handball and ice hockey. "Save" implies a goalkeeper, and water polo has one. "First leg" evokes a two-legged knockout tie. And "Royal–Thomian" in the Sri Lankan context attaches to both cricket and football—never to water polo. That is where I stopped. The model did not err because it was stupid. It erred because people like me have not written about water polo. The system returns what it is fed.

Thirteen information points, zero sources. Every one of the thirteen information points lists its source as: none. Not one. We talk constantly about the provenance of allegations and leaks. Here the inversion is total: every claim in a promotional article has no paper behind it. When was the match? Where? What was the score? On leg-one evidence, who leads? Nobody knows, because the article does not know. The article knows only that a video exists and that you should watch it. An unsourced claim is not information—it is an assertion, worth exactly as much as its chain of custody proves.

Adjectives becoming data. The piece returns repeatedly to "prestigious," "exciting," "competitive," "young and competitive." These are the author's opinions, not facts, and they are bound into the same bundle as the label and the absent sources. Once inside a dataset, a sentiment-scoring model may record "exciting" as 0.7. That 0.7 joins thousands of other articles to build a "viewer engagement index." The index is manufactured, nobody questions it, and three years later it is treated as true. When promotional adjectives enter a database, they stop being promotional and become statistics.

U15, safeguarding and silence. Not a single player, coach or official is named. For minor athletes this is correct practice—no profiling of children, coverage kept at event level. But that silence has a second edge. To a database, a piece with no names is a football match with no players, no teams and no stars: worthless to sporting analysis, valuable to ad targeting. The weight is false and is used as though true.

Where the financial framework collapses. No financial or transfer framework can be applied here, and anyone who applies one is writing fiction. There is no transfer, so no transfer fee; no broadcast rights, so no broadcast revenue; no wage bill, so no wage-to-revenue ratio; no FFP or PSR exposure, because Sri Lankan school aquatics is amateur and institutionally funded. I stress this because it is the lesson of the case. A framework does not fit everywhere, and the analyst who forces it pushes the system's original error one step further down the line. "Not applicable" is the hardest sentence in journalism, and the one reporters must write most often.

Downstream damage. The real cost lands where nobody looks. A mis-classified pipeline returns bad keyword search results. Entity matching stalls or mis-matches "Royal College" against a club database. Audience targeting shows a water polo clip to football-interested viewers; click-through falls; the system concludes the audience disliked the sport, when in fact the audience was simply handed the wrong product. Then comes the worst of it: aggregation. Twenty water polo highlights filed as football distort monthly football content volume, average goals, and regional spread. Nobody forges anything. One wrong tag is applied, and the silence around that tag does the damage.

What a correct record would look like. Remove the fourth label and ask what should be attached. First, a timestamp proving the record's existence: when it was published. This case has none. Second, source attestation: who supplied each fact, on the basis of which document. Here all thirteen say none. Third, sport verification: the file is water polo, so the governing bodies are World Aquatics and the Sri Lankan national aquatic federation—not FIFA. Fourth, in some cases, verified evidence of age-eligibility documentation, especially in youth bands. Fifth, a correction log: when an error is caught, who fixed it, and how long that took. Now imagine all five written as hashes into a tamper-evident ledger—this record was written at this time, from this source, corrected then and thus. Anyone could open it and see whether it had been altered. Nobody would need to open a drawer or delete anything. That is the modest benefit of custody.

Three old files, one pattern. In 2026 in Chattogram I found an approximately fifty-thousand-dollar gap between three players' declared transfer fees and the actual bank transfers at a local club. The papers were clean. The numbers were clean. The discrepancy lived in custody. A year later, in leaked Russian doping records, a test date had been moved to avoid a positive result; the report itself looked immaculate. The lab data was clean. The chain of custody was not. In 2026 I read a La Liga club's pandemic-era accounts backward and found fifteen million dollars of inflated revenue from the sale of an intangible asset to a related party. Every clean transfer has a second set of books somewhere: a date, a source name, a correction note. Here that small book reads, in every line, source: none.

What a ledger does and does not prove. A tamper-evident ledger can prove a record has not been changed since it was written. It cannot prove the record was true when written. Hash a false claim into a ledger and the ledger will immortalise it, not vindicate it. A ledger protects custody, not imagination—and the assumption that custody equals truth is ours, not evidence's. Lose that distinction and the future of sports data becomes an immutable archive of falsehoods that nobody can delete and nobody can find.

The contrarian angle. The obvious reaction is: it is an Under-15 school water polo highlights video; who cares about a tag? That is the wrong frame. The story is not the tag. The story is that the error stayed invisible precisely because the subject is small and unobserved. Invisibility scales. A mistake in a major fixture invites argument, rivals, scrutiny. A mistake in a minor event has no interested party, no commercial loss, no editor of known identity—so it survives untreated and dissolves into the totals. There is also a second reading, more uncomfortable than machine error: if the classification was deliberate—keeping a football audience engaged, or inflating a football market—then it is not an accident but a routing strategy. Accidents can be repaired. Strategies cannot, because strategies have beneficiaries. And the oldest caution applies: this article is about no school, no coach and no child. Open the file and the accused are the classifier, the publisher, and the system that swallowed the record whole.

Takeaway. Three things will show next. Whether the label is corrected—and whether that correction is logged or done quietly. Whether the leg-two highlights arrive with anything written in the source field, or give birth to thirteen more "none." And whether any other aquatic article in the same pipeline has been labelled football; that is the real test. The distance between a football match and a water polo match is two words. A system that cannot tell those two words apart should not be handed the arithmetic of the sport. I stopped counting the denials when the bank records arrived. There are no bank records here, no federation minutes, no visa letters—only a video and a wrong label. That does not mean nothing happened. It means nobody has looked yet.