Similarity and affinity are different questions

One asks who else plays like this, the other asks who plays well next to him. Most rosters that do not work were built by asking only the first.

Two questions that sound like one

When a scouting department says it wants a player like the one leaving, it is asking about similarity: same production, same shot profile, same role in the offence, ideally at a lower cost.

When a coaching staff says a signing has to work next to the point guard it already has, it is asking about affinity, which is a different calculation entirely and often points at a different player.

The two questions get mixed up because they use the same verb. A model that answers one and gets presented as the answer to the other is the quiet start of a roster that does not fit together.

What similarity is good for

Similarity shrinks a list of thousands into twenty worth watching, which is exactly the work a scouting department cannot do by hand across 200 leagues.

It works best on a replacement problem: a wing is leaving, the system stays the same, and you need someone who does roughly what he did without paying what he cost.

Its limit is that it is blind to the rest of the roster. Two players can sit at 0.91 similarity and still be the wrong signing, because the team already has three of them.

What affinity is good for

Affinity asks a harder question: given these four players on the floor, which fifth one raises the group instead of duplicating it.

It reads what the existing group already produces and where it gives something away, then looks for the profiles that cover the second part without breaking the first.

This is the model that stops a team from buying its fourth good shooter and its first real problem in the paint, which is a mistake that costs a season and looks reasonable on the day it is made.

The cognitive layer, where it gets interesting

TeamKeys runs the same two questions on cognitive profiles as well, and there the difference between them becomes very visible.

Cognitive similarity measures whether a player thinks the way your system requires, while cognitive affinity measures whether he strengthens the domains your squad is weakest in.

A player can score high on one and low on the other, and knowing which one is which changes what you do with him: one fits the system today, the other fixes something the system has been hiding.

Neither one signs a player

Both models produce a shortlist, and a shortlist is where scouting starts rather than where it ends.

The value is in the weeks they give back: instead of watching two hundred games to find twenty candidates, a department watches twenty games on candidates the data already argued for.

What the models cannot see is the locker room, the medical history and the conversation with the agent, which is why the last call still belongs to people.

Questions people ask

Can I get similar players across different leagues?

Yes, because every league sits on the same data structure, and each cross league result carries its own adaptation estimate instead of pretending the numbers travel unchanged.

Does high similarity mean the same player?

No. It means the same measurable profile, and two players with the same profile can still differ in everything a model does not measure.

How is affinity calculated?

From what the existing group produces and where it concedes, so the result depends on your roster and changes the moment your roster does.

Which one should we start from?

From the question you actually have: similarity if someone is leaving and the system stays, affinity if the system is the thing you are trying to fix.