How league adaptation actually works

Every summer a signing arrives with a line that looked excellent somewhere else. What that line becomes here is a question with a real answer, and it is not the same number.

What actually changes across a border

Four things move a player line when the competition changes: how fast the game is played, how good the average defender is, what referees allow, and what role the new team will actually give him.

Pace is the easiest to correct, because totals per game can be rebased on possessions, and once that is done a lot of apparent scoring differences disappear.

Defensive quality is harder, because it does not sit in the box score of the player who faced it, and it is the single biggest reason a shooting percentage stops travelling.

Why the raw number stays visible

A model that silently rewrites a percentage is a model nobody in the room can argue with, and a figure nobody can argue with does not belong in a signing decision.

TeamKeys shows the original line and the adapted estimate side by side, so a scouting director can see both the claim and the correction being applied to it.

That also makes the estimate falsifiable: when the player arrives and the season starts, everybody can check whether the correction was right, which is how a model gets better instead of getting defended.

Where the estimate is strong, and where it is not

The estimate is strongest between leagues that already exchange a lot of players, because there the model has hundreds of real transitions to learn from rather than a theory.

It gets weaker on a small sample, on a player moving between competitions that rarely trade, and on a role change that the numbers cannot see coming.

This is why the platform reports the estimate with the context it deserves, and why a staff should treat a wide interval as a reason to watch more video rather than as a green light.

The part a model will never do

Adaptation says how the production might travel. It says nothing about whether a player will settle in a new city, learn a new language, or accept coming off the bench.

Those things decide plenty of signings, and they are the reason the model is a filter for the shortlist instead of an answer for the meeting.

Used that way it earns its place: it removes the candidates whose numbers were an artefact of their league, and leaves the ones worth the flight.

Questions people ask

Does the platform adjust the numbers automatically?

It shows the adapted estimate next to the original figure rather than replacing it, so you always see both the raw line and the correction.

How much can a shooting percentage move?

It depends on the pair of leagues and on the volume behind the percentage, which is why the estimate is given with context instead of as a single confident number.

Does it work for a player moving up a division?

Yes, and that is where it is most useful, because a division change is exactly the case where a raw line is most misleading.

Can we see the history the model learned from?

The platform holds up to 25 years of boxscore history across 200 leagues, which is the material the transitions are built on.