This episode unpacks major upcoming changes to the Bittensor protocol, revealing how sub-subnets, subnet deregistration, and evolving consensus mechanisms will reshape the network's economic and competitive landscape for investors and developers.
Introduction to Sub-Subnets: A New Layer of Complexity
- Const, a core contributor, reintroduces the concept of sub-subnets, a feature that allows a single subnet to run multiple, distinct incentive mechanisms under one token. Initially shelved due to feedback from power-users who could replicate the functionality mathematically, the feature was revived to improve ease of use and transparency for all developers.
- Sub-subnets create a multi-dimensional weight matrix for each subnet, enabling separate incentive, trust, and consensus terms for different mechanisms.
- This design turns subnets into something akin to a Mixture of Experts (MoE) model in machine learning, where different components specialize in different tasks, fostering greater sophistication.
- Const explains, "It's a transparency issue... This will allow people to actually, you know, call for the weights for mechanism one and two and three and see how they're performing."
- Initially, each subnet will be able to support up to eight sub-subnets. This is a soft limit chosen to manage chain bloat while still providing significant flexibility.
Strategic Implication: For investors, this change could lead to a consolidation of projects. Teams running multiple subnets (e.g., Templar) might merge them into a single, more complex subnet. This simplifies investment analysis by focusing value accrual on a single token rather than spreading it across several.
The Economics of Sub-Subnets and Chain Bloat
- The conversation addresses the trade-offs of this new architecture. While sub-subnets will increase the amount of data (weights, bonds) stored on-chain, potentially increasing bloat, they may also reduce the demand for new subnet slots.
- By allowing teams to run R&D or open-source initiatives alongside monetized, closed-source mechanisms on the same subnet, it creates a more sustainable economic model.
- This structure allows a profitable mechanism to subsidize a more experimental or public-good-oriented one under the same token, mirroring R&D departments in traditional companies.
- This is the first version of the feature; future iterations may include distinct UID sets for each sub-subnet, though the initial implementation uses a shared set of 256 UIDs.
Subnet Deregistration: Introducing Competitive Pressure
- A major protocol change discussed is the introduction of subnet deregistration, a mechanism designed to remove underperforming or abandoned subnets from the network. This functions as a "garbage collection" process to maintain network health and quality.
- Deregistration Trigger: Subnets will be deregistered based on the lowest price, calculated using a moving average. This is analogous to how the lowest-emission miners are removed from a subnet.
- Process: When a new subnet registers, it takes the slot of the lowest-priced existing subnet. The TAO held in the deregistered subnet's liquidity pool is returned to its token holders.
- Const frames this as a necessary adaptation: "We're just adapting to the way in which the market has evolved and seeing that... fewer slots and having more attention on those fewer slots is going to be better for the network as a whole."
Strategic Implication: Deregistration transforms subnet ownership from a permanent asset into a competitive, performance-based position. This introduces a new layer of risk for investors in lower-ranked subnets but also creates upward pressure on quality and performance across the entire network.
Deregistration Mechanics and Fairness
- The community raised concerns about the fairness and implementation details of deregistration, particularly regarding incumbent subnets and the lock cost.
- Immunity Period: New subnets will have a four-month immunity period before they are subject to deregistration, giving them time to establish themselves.
- Lock Cost: The initial TAO locked to register a subnet will be returned to the owner upon deregistration, minus any owner emissions they have already received. This is a "soft landing" approach to avoid penalizing teams who registered under the old "immutable slot" assumption. In the future, this lock may be fully burned.
- Registration Cost: The cost to register a new subnet will start high (around 2,000 TAO) and decay over a half-life of approximately 3.5 days, preventing a rapid, low-cost churn of new registrations.
The TAO Halving and Its Market Impact
- With the first TAO halving approaching in December, Agrippa asks about its expected impact on the network and subnet prices.
- The halving reduces the amount of TAO emitted as a subsidy into the subnet liquidity pools.
- Because the emission mechanism adds liquidity rather than directly manipulating price, Const states the halving is not expected to directly affect subnet prices.
- However, the reduced liquidity may lead to increased price volatility (in both positive and negative directions), especially for newer subnets with smaller pools.
- There are no plans to synchronize the TAO and Alpha token halving schedules.
The Path to Decentralization: Governance and Proof-of-Stake
- Const outlines a two-pronged approach to increasing Bittensor's decentralization, addressing both the underlying consensus and the governance structure.
- Chain-Level Decentralization: A Proof-of-Stake (PoS) devnet is already operational. The team is now building a full mainnet clone to stress-test the PoS mechanism before a potential launch next year.
- Governance Decentralization: The plan is to move from the current Triumvirate model to a more distributed system. This includes giving veto power to subnet owners and a "Senate" (composed of top stakers). This allows for rapid upgrades when there is social consensus but provides checks and balances to block contentious changes.
Yuma Consensus: The Evolution to Fight Weight Copying
- The discussion shifts to Yuma Consensus, the algorithm that aligns validators. Community expert Ref joins to explain its evolution.
- Yuma 1: A basic consensus mechanism that rewarded validators for agreeing but inadvertently penalized the first validator to identify a new, better miner.
- Yuma 2: Introduced bonds and dividends to reward validators for diverging from consensus to support a new, high-performing miner, with the expectation of a larger payout once consensus shifts. However, implementation issues limited its effectiveness.
- Yuma 3: Described by Ref as a superior implementation of the Yuma 2 concept. It fixes the earlier issues and introduces a "bond penalty" parameter that more effectively punishes late validators, directly combating weight-copying strategies.
- Ref states, "Yuma 3 is better than Yuma 2 in all dimensions... The Yuma 2 implementation had some implementation problems. It was not working as well as it was intended to work."
Value Accrual: Tying Real-World Revenue to Subnet Tokens
- A recurring theme is how to ensure external revenue generated by a subnet flows back to its token holders. There is no on-chain enforcement mechanism for this.
- Const acknowledges this is a challenge being solved organically by different teams.
- Subnet owner Fish discusses two primary methods:
- Buybacks: Using external revenue to buy the subnet's own token on the open market, creating direct price pressure.
- Distributions: Paying out revenue directly to token holders, similar to stock dividends.
- The conversation highlights that this currently relies heavily on trust in the subnet owner's alignment and transparency. Automation via smart contracts or exchange APIs is possible but complex and not yet a primary focus for most teams.
Conclusion
This episode reveals that Bittensor is entering a phase of competitive maturation. The introduction of deregistration and sub-subnets forces a focus on sustainable economic models and performance, while protocol upgrades like Yuma 3 provide tools to ensure fair competition. Investors must now evaluate not just a subnet's technology but also its economic design and governance.
