How do they know there are Cashapp or Robinhood?

Most often this is done through input correlation and known cold address interaction. The heuristic is that:

• any transaction that has inputs from address A and B is from the same entity
• any transaction that deposits into a known cold address belongs to the entity that owns the cold address

coinjoins break the former, which is why many KYC exchanges will reject your tx when you coinjoin, because their analytics software doesn't work on it.

I assume you would like to know the link between this attributed address and the original attributed address? I don't think OXT had that, since there can be hundreds of transactions between those. We would like to eventually build something like that though.

But like @optimism wrote, we cluster addresses together based on common spend heuristics just like Chainalysis does. This way if we add an attribution to 1 address and that address is spend together with another address the second address gets the attribution as well. And if that second address is spend with a third address the third address also gets the same attribution and so on...

This way you can group millions of addresses together.