ive been following this discussion , and i most say, the level of control you need to have over the network to use it for a practical sybil attack is unfeasable, my worst case scenario tells me max 10% could be controlled, now if you do run it through say 10 times, the chance of being unmasked is 10% and we are talking about several factors that have to happen..
you need the 10% to be on EVERY mixing stage - Very unlikely im not even going to try and calculate the chance
Just trying to keep up with the concepts here...
If a tx is compromised on stage 1, for example, is it compromised through all of the following stages as well (assuming it doesn't pass through any more sybil nodes)?
You may be confusing 2 different things.
Each Darksend has 3 stages:
Collect inputs
Collect blind signed outputs
Collect signed inputs
The denial-of-service issue that would cause Darksend to repeat over and over and be blocked would be if an adversary refused to complete their part of the stages. Evan solved this by attaching a collateral penalty payment to stage 1 and this is charged only to those to don't complete all stages.
In doing that, Evan had to expose each Darksender's IP, input, and output to the randomly chosen MasterNode. So the MasterNode knows your identity, the anonymity is broken.
But if the MasterNode is not an adversary, it will not use that information against you (it won't retain nor share it with others).
If the MasterNode is an adversary, then that Darksend is not anonymous for you. There are other ways that a Darksend can end up not anonymous for you, see
my calculation example upthread, e.g. Tor is also not 100% anonymous every time you use it.
So let's say that 20% of MasterNodes are adversaries (e.g. China's NSA buys up 20% of MasterNodes), then every time you do a Darksend, you have at least 20% chance of losing anonymity for that Darksend. (and more than 20% because Tor isn't 100%, etc)
So you do multiple Darksends and each one has a 20% (or more) chance of being not anonymous. But if any one of them is anonymous, then you are anonymous.
So take that percentage (including Tor factor, etc) and raise it to the exponent of the number of Darksends you will do on the same funds. So let's say 0.30^10 = 0.0000059049. That is 6 in a million. So very strong anonymity.
But please see
my upthread calculation, because there are many factors and your anonymity will not be that strong in reality.
The adversary can also Sybil attack the inputs, meaning they flood the Darksend with inputs and thus assuming that Evan needs to set some limit on the number of inputs in a Darksend, then this attack can lower the anonymity set as I explained in my calculation.
Can the final input, output and IP be compromised at each mixing stage? If so, how do multiple stages help at all?
Once a sender is compromised, how are they tracked in the future? IP? Wallet address? Some other way?
Your IP, input and output are compromised at the probability (frequency)
of the calculation.
