Bittensor
Search…
Consensus, Incentive, and Bonding
The network blockchain uses a consensus algorithm to reach a common agreement over who the most valuable participants in the system are. At each block step, Tao is distributed throughout the system to reward these participants.
Servers are rewarded based on their performance in individual instances of knowledge production, and receive two separate assessments from Validators that are combined to determine their Tao inflation:
  • The first assessment is the consensus score and is determined by the number of approval votes given by validators (trust). This score is not earned in a linear fashion, as Validators with more stake (Tao) are able to distribute a greater number of votes. The score is also regularized by a sigmoid function: if less than 51% of the votes a Server receives are positive, the score is reduced exponentially, if more than 51% are positive, it is likewise inflated.
  • This consensus score is then combined with the rank - numerical determinant of value - achieved to determine the server's Tao inflation. Again, this score (rank) is skewed: it is combination of weights set by the validator, and the stake of that validator.
Validators, on the other hand, are incentivized to collaborate with highly valuable Servers in the system who are also trusted by the majority through a pair of conditions:
  • The first of these is the “bonding mechanism”: when Validators rank a given Server, they simultaneously purchase equity in that node, and receive inflation proportional to this. Because of the consensus mechanism previously described, Validators benefit from distributing their highest scores to Servers who are likely to be scored highly by other Validators as well - “trusted” Servers.
  • In addition, Validators are limited in the number and size of scores they are allowed to distribute, and so must choose wisely where their investment is allocated: they must only choose the best.
Copy link