What a scouting number can actually answer

The question comes before the metric

Most scouting reports start from a leaderboard, which is the wrong end of the problem, because a leaderboard answers a question nobody in the room has asked.

The useful order is the opposite one: the staff says what it needs, a wing who can defend two positions and shoot off the catch, and only then does the data go looking for candidates.

That single inversion removes most of the noise, since a metric picked to fit a need is far easier to argue about than one picked because it was already on the screen.

Context is not a footnote

The same shooting percentage means two different things in two different leagues, and the gap is not a rounding error: it is often the whole reason a signing works or fails.

Pace, physicality, referee tendencies and the quality of the average defender all move a player line, so a number carried across a border without an adjustment is a number that has quietly changed meaning.

This is why TeamKeys attaches an adaptation estimate to every cross league comparison, and why that estimate is shown next to the raw figure rather than replacing it.

Similarity is a shortcut, not a verdict

Similarity models are good at one job, which is shrinking a list of thousands of players into a list of twenty worth watching on video.

They are bad at a different job, deciding who to sign, because two players can look alike on every axis a model measures and still fit a locker room in opposite ways.

Used as a filter, similarity saves a scouting department weeks of work; used as a verdict, it produces confident mistakes that nobody in the room can explain afterwards.

What a model owes the person reading it

A number that arrives without its working cannot be argued with, and anything that cannot be argued with does not belong in a decision that costs a season.

So every answer should carry three things: where the data came from, which players it was compared against, and what would have to change for the answer to flip.

TeamKeys AI is built on that constraint, because it answers only from the TeamKeys database through forced tool calls, and it says which query produced the figure it just showed you.

The backbone under the answer

None of this works without a single spine of data, which is why TeamKeys keeps scouting, performance and cognitive profiles on the same 200 leagues rather than on three separate products.

The database holds more than 65,000 players and around 116,000 matches, and it refreshes every four hours, so a report written on Monday morning is not reading Friday night.

The platform speaks seven languages, and the same figure carries the same definition in each of them, which matters when a Milan office and a Tel Aviv office argue about the same player.

Questions people ask