Reputation-Based Consensus

We intend to test an evolution of the proof-of-stake consensus algorithm that we call Proof of Reputation, which combines several factors: stake, activity in the network, specific rating aspects (particularly benefit rating), length of time elapsed with activity and rating levels above specific thresholds, and others. Machine learning can be used to optimize the combination of factors.

There is a significant overlap between what we intend with Proof of Reputation and the NEM blockchain’s “Proof of Importance” framework, so our Proof of Reputation will borrow NEM’s 16 ideas and perhaps some of their algorithms. Some component of proof-of-work may also be desirable, but we would rather solve some beneficial machine learning problem than burn cycles on cryptographic puzzles. The computational cost of these machine learning tasks varies much more than for most crypto puzzles, so this idea needs refinement over the next few years. It seems most likely that, at the end of a period of refinement and experimentation, we will end up with a Proof of Reputation framework incorporating some NEM-like aspects with a machine learning–based proof-of-work component.