Note §
- This is a weekly update from the part of the team that didn’t attend the All-Hands Event in Athens.
Mixnet (Network Privacy) §
Research §
- Regarding the current situation with the mixnet and staking, we have decided to focus on the simplest approach, that is only single-staking without stake delegation as suggested for the first iteration of the staking. Due to the simplicity of this approach, we can model it more precisely and learn more of its properties.
- We have received a couple of comments regarding the mixnet from an external reviewer. The comments and our analysis can be seen in this document.
- Reviewed an interesting critique of mixnets and suggestions on their improvements. One of them is the Poisson mixing design choice that is part of Nym and is claimed to be wrong. This led to a discussion on how to design a better mixing mechanism. We don’t know exactly what is broken in the Poisson mix; there are some claims and some proposals on how to solve the problem suggested in the presentation. We have asked for more paperwork supporting their claims and the authors of the presentation are working on it. In the meantime, a proposal has been prepared based on suggestions from the presentation.
- Looked at the Bittensor proposal (especially the idea of treating a node as a learning machine and combining it with a rewarding function), read through all of their papers. If we would like to use a similar approach then we must prove that the outcome of the mechanism cannot be biased - which might be very hard.
- Optimization of mixnets: written the initial specification and some functions for the algorithm which uses the (asymptotic) lower bound on the probability that the fraction of compromised paths belongs to some interval to compute the optimal number of layers. The details are provided in this document.
- Also, tried to tighten the bound on this probability using the AM–GM and Markov’s inequalities, but trying this approach has not produced any improvements so far. The details are provided in this document.