Garbage Data in the Scouting Room: The Cost of a Wrong Label

In the summer of 2026, at the Trigoria training centre, I opened a data file at six in the morning. AS Roma's internal system had dropped it into the folder labelled football. Inside was a transcript of an argument between guests on a reality television programme: a host, a handful of contestants, and a remark considered tasteless about a presenter on a rival network. No line-up. No metrics. Not one player. The label had been attached by an algorithm. It met the words gala, elimination, rival, host and competition, and concluded this was sport. Had I not opened the file to check, that wrong label would have sat quietly in the database, waiting for some recruitment meeting to cite it as a piece of evidence. What keeps me thinking about that morning lies elsewhere. A wrong label on a news bulletin is harmless. A wrong label in a recruitment room can change a player's fate. The database grows; the number of readers does not Based on my experience watching matches and working with academies in Italy, modern youth setups absorb volumes of data nobody could have imagined a decade ago. Every week, a mid-sized Serie A academy takes in thousands of files from third-party providers: scouting reports, match-metric packages, session logs. All of it flows into one pipe, and all of it must be labelled before anyone has time to read it. The number of readers does not rise with it. The analytics department at most European clubs still numbers a few people. They filter by label first and read with their eyes second. When the label is the only handle for choosing which file is worth opening, the quality of the label becomes the quality of the whole recruitment process. In Vietnam, centres such as PVF or the HAGL JMG academy are also building their own data stores. The scale is far smaller, but the problem is identical: importing data is easy, keeping it clean is hard. Polysemy is the doorway for rubbish The mechanism is almost absurdly simple. Classification models learn surface vocabulary. In English and Spanish, gala appears both at a Ballon d'Or ceremony and on a television show. Host is both the home team and the presenter. Elimination is both a knockout round in European competition and an eviction night. Those four keywords are enough for an entertainment file to slip through the gate. Scouting runs into exactly the same trap, only in another language. A player described as a regista in Serie B may in truth be a deep-lying ball-winner. A machine-translated report from Portuguese into Italian can turn an attacking full-back into a defensive one. The club buys him for the wrong role, coaches him to the wrong template, and two years later concludes he was never good enough. The first discipline I learned from an old head of player development: one file, one topic. If the source covers several things, split it or drop it. An honest blank beats an invented judgement In my own reports I write the sentence insufficient information to assess so often that younger colleagues tease me as the man of empty boxes. They dislike it, because it makes a report look unfinished. But between an empty box and a fabricated judgement, the distance is the credibility of an entire department. An empty box invites someone to come back and check. A fabricated judgement closes the file. That is why I treat saying I do not know as a professional act. That lesson cost me something concrete. In 2026, while working as a data analysis assistant at the AS Roma academy, I stayed behind after the Coppa Italia tie against Virtus Entella to watch the footage again. I found that the young midfielder Nicolò Zaniolo, then seventeen, landed on his left foot at a 62% deviation, loading his right knee asymmetrically after every turn. I wrote the report and sent it to the medical staff. No reply came. Not long afterwards Zaniolo went down in a turn, his anterior cruciate ligament torn. A warning with no owner is like a wrong label with no checker. Both sit quietly in the store, and both surface only when it is too late. Sediment does not lie; it only deceives those without the patience to dig. Conversely, some data reaches only the person willing to come back. In the summer of 2026, with stadiums shut by the pandemic, development was cut in half. I designed a small programme for fifteen Roma trainees: fifteen minutes of skipping, resistance bands, and a WhatsApp group to track each of them. Living rooms became gyms, because talent does not wait for someone to make its bed. On that list was Edoardo Bove, then eighteen, a name nobody noticed. I logged every session, every breathing rhythm. When the season resumed, Bove had added 12% to his maximal endurance and was promoted to the first team for the Europa League fixture against Young Boys. That number came from no data provider. It came from one person reading every line another person had written. In 2026, at the European Under-21 Championship, I travelled with Italy as an analysis assistant. In the quarter-final against Portugal, Italy lost 3-5 on penalties. The score was not what I brought home. I spent two hours logging eighteen line-breaking passes from Sandro Tonali, who kept dropping deep to drag opposing centre-backs out of position. The result was a forty-page report on the space between the lines, and the youth coach put it into the syllabus. Three stories pointing one way. What creates value in recruitment is the number of times a human being sits down to check a line of data the system already believes. Three gates before data reaches the meeting room The first gate is entity checking. Every name in the file must belong to football: players, clubs, competitions, coaches, referees. A file labelled football whose entity list is all presenters and contestants is rejected at the door. The second gate is one file, one topic. Mixed files are split or discarded. It is the most time-consuming rule and the one that saves the most meetings. The third gate is source and absolute date. Every fact carries a source name and a specific date. Phrases like last week or recently are deleted. These gates need no expensive software. They need someone willing to read. The counter-intuitive angle: noise breeds faster than signal The industry's common belief is that more data produces better recruitment. Operational reality shows the opposite at one important point: noise breeds faster than signal. An academy imports ten thousand files a season. If only 2% are mislabelled, the store already holds two hundred pieces of rubbish. Rubbish rises to the surface first because it is written in more seductive language: arguments, conflict, familiar names. Real signal is quiet: a 62% deviation, a 12% endurance gain, eighteen passes nobody counted. The incentive structure of the profession pushes rubbish in faster than any algorithmic fault. The analyst who writes insufficient information to assess is read as incompetent. The one who invents a plausible comparison is read as having personality. A data store does not generate mistakes on its own; it preserves the ones people were too lazy to check. Reflection The next competitive edge for academies may lie in the willingness to throw data away. A club that dares to say this file does not belong to us will save more than a club that buys another provider. That morning at Trigoria, I could have skimmed past the file labelled football and trusted the system. Had I done so, what would have become of the names that genuinely needed to be read?