This is part of our series on a year of Bittensor experience, leading up to our anniversary at the 13th of July. After discussing the Dirac incentive landscape, we now look at its opposite: the flat incentive landscape. Please let us know what you think in our Discord channel!

Subnets with a fully flat incentive landscape, where all miners earn the same for their contribution of effort, will fall prey to blockchain-savvy miners that know how to get preference when registering UIDs, or that know tricks to prevent deregistration. If the registration fee is dynamic (which is the case for most subnets) the blockchain-savvy miner will be able to optimize the cost/reward ratio and morph the subnet into a monopoly or oligopoly.

Subnets with a slightly skewed, but still mostly flat incentive curve (and dynamic registration fee) will inevitably morph into a monopoly or oligopoly, where top miners will spend significant parts of their earnings on registration fees. No special blockchain skill is needed; just a small edge to become a top miner (honestly or otherwise) will do, the dynamic registration fee will do the rest. The top miners do this either to keep their UIDs after incidental deregistration (some subnets are a bit noisy in their scoring), or they buy up UIDs periodically just to keep the registration fee so high, that new miners cannot enter the competition without taking a significant (and often unacceptable) financial risk.

The top miner will actually earn something. Those in second place or lower will spend most of their earnings on registration fees – the top miner simply forces the price up to (or beyond) that point.

The dynamic registration fee seems to be a cornerstone of Bittensor. In practice, the downside of abusive monopolies can exceed the perceived upside of the price forming mechanism. Many successful subnets have (had) a near-zero registration fee, sometimes even constant, and therefore a low barrier to entry. This has value for the ecosystem, because it attracts new talent – something that miners don’t, but subnet owners should want.

In the next article we analyze what happens when combining the properties of Dirac and flat incentive landscapes.


0 Comments

Leave a Reply

Avatar placeholder

Your email address will not be published. Required fields are marked *