Very much appreciate that you're taking the time to look deeper into the tech.
I very much agree with your take, the arbitrary amount coinjoin coordination protocol opens up an entirely new design space of anonymous bitcoin transactions.
We've been doing a pretty bad job of that in the past, and Wasabi 2.0 is our first attempt to make a substantial improvement, yet still, it can get even so much better!
As far as I have been able to see, the decomposition of the input tx into the set fractions is random, it does not have to go from largest to smallest, or smallest to largest.
I also love the recombination it does in later mixes of utxo already mixed with new utxo.
Starting from different utxo and separated in time, I find it very difficult if not impossible to even associate later consolidations.
I do not know very well how to describe it, I would say that all this procedure adds a lot of entropy to the system, your utxo do not remain static, they can change when a new unmixed utxo enters and a later mix is performed even without you selecting those utxo already mixed with the previous version of Wasabi.
Honestly it is an incredible job.
Correct.
The number of inputs your client registers is random, with a bias depending on the size of your existing utxo set.
The group of inputs your client registers is random, depending on the anonscore calculated for each utxo and the value of the utxo.
The number of outputs your client registers is random, depending on how many possible decomposition there are.
The value of outputs your client registers is random, depending on the value of other users input, and the sum value of all your inputs.
There are a lot of different possible input and output groups, we remove those that are very expensive, and pick randomly from the rest.
Here is a bit more thorough explanation from a year ago, there have been numerous changes and improvements since though.
https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2022-April/020202.html
