Thanks for the heads up. The researcher deanonymized centralized mixers, due to them using arbitrary amounts. I previously wrote about it here (Issue 3: Mixing Unequal Inputs (The Wrong Way)): https://medium.com/@nopara73/traditional-bitcoin-mixers-6a092e59d8c2
That is right. If BC analysis can recognize the wallet then it can recognize that this transaction has 2 interpretations with the same likelihood, which of course would be equivalent to a 2 anonymity set mix.
That’s a quite interesting issue. Of course, we could think round based coinjoins are benefiting from Schnorr a lot more than JoinMarket transactions, because there you aggregate a lot more signatures. And of course it is true, but Schnorr will be an implementation nightmare, because it is interactive.