On Consensus

The Bittensor Network owes much of its efficacy to its carefully designed consensus mechanism, which has the following aims:

  1. To deter malicious activity amongst Servers in the network. The way it does this is by rewarding Servers in the network who are “trusted” with earning magnitude greater than low-trust Servers, who are weakened to the point of de-registration.
  2. To encourage Validators to distribute honest assessments. The way this is done is by placing a limit on the voting power of Validators, while also incentivizing them through a bonding mechanism to interact preferentially with high-performing Servers.
  3. To promote the interaction of high-value peers in the network, such that valuable nodes receive significantly higher earnings, and are more likely to stay registered. This promotes the creation of value, while also giving external users of the network access to the top-performing Servers, and the ability to “tune” their activities according to their informational needs.

This is how it works:

Servers receive two separate scores that determine the amount of TAO (τ) they earn.

Consensus (C): This score is determined by the number of approval votes given by Validators (trust). This score is not linear: Validators with more TAO can distribute a greater number of trust votes, and the score is regularized by a sigmoid function such that “trusted” Servers - those who receive 51% or more positive votes - are rewarded orders of magnitude higher.

Rank (R): These are individual scores given by Validators based on the value a Server has provided. This metric is also non-linear and is a combination of the score (S) given and the TAO holdings of the individual Validator.

Validators are subject to a set of conditions that determine the amount of TAO they earn:

The bonding mechanism: When Validators rank a Server, they purchase “bonds” in that node, and thus receive more TAO when they interact with more valuable, high-stake Servers.

Limited score supply: Validators have a capped amount of scores they can distribute, and so must strategically allocate their “investment” of scores.

The ultimate result of all these factors is maximum network utility, and the promotion of high-value informational value in the system. It is also important to note that our consensus mechanism is subject to change, as it is the nature of consensus mechanisms that the same strategy will not work the same way all of the time. As the network scales, our strategies will evolve alongside it.

Subscribe to Bittensor

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.