Livepeer 2.0 Protocol Evolution

In July an exciting vision update for Livepeer was shared publicly: Livepeer 2.0 is an open video agent platform, that combines a new media planning and execution agent harness, in Livepeer Agent, with a best-in-class network infrastructure for media centric compute, in the Livepeer Network. Agents are taking over video production, and the Livepeer Agent makes the network reachable from the tools people already use.

It also surfaced a series of candidate protocol updates to better align the network incentives and behavior to support this new world of multi-faceted media compute on the GPUs on the Livepeer Network, however it stopped short of communicating specific commitments about the mechanisms that would enable this. The brand new Livepeer 2.0 Litepaper, recently released, now commits to one concrete design with specific mechanisms and starting parameters, for updating the Livepeer Protocol for this new world. Here are the main pieces of this key protocol update.

Rewards follow real work

Today, LPT rewards are paid in proportion to stake. Participants earn whether or not the network is doing anything useful. That struggles to scale to hundreds of different AI media job types, where stake alone can’t tell real usage from a node paying itself.

In Livepeer 2.0, a node’s LPT reward is the smaller of two numbers: its share of network fees, or its share of network stake. This is called Min-Fee-Stake (MFS). A node that has 10% of network stake, but earns only 2% of network fees in a round, has its rewards capped at 2%. Fees themselves are never capped, so doing more work always pays more to the node operator, but this encourages stake to move from underperforming nodes to those who are putting real points on the board and performing meaningful work in service of the network.

In this new model, anyone can register as a node operator, with no fixed bond requirement and no 100-node cap. Delegation also becomes an active job. Moving stake to a busy, but under-staked node helps raise its reward ceiling, and is far more profitable for the delegator. Leaving delegation sitting on an idle node, will earn zero rewards.

Validators check the work

Livepeer 2.0 adds a role that the protocol has never had before: validators, who judge whether work was done honestly. The top 33 nodes by total stake, including delegated stake, can register as validators. Validators score each node from 0.0 to 1.0, and the median score decides how much a the node’s eligible rewards it receives. The score is meant to catch dishonestly such as self-dealing, fabricating fees, or maliciously returning incorrect results, rather than act as a quality-of-service grade.

Rewards also take 7 rounds to become claimable so that scoring can prevent an attacker from cashing out prior to being noticed. The unbonding period is extended to 21 rounds, to apply a real cost-of-capital penalty to those who are backing nodes that maliciously attack the network.

Fees are paid in USDC, and half go to burn LPT

Network payments move from ETH to USDC, so node operators can now price work in dollars. At launch, 50% of network fees go straight to the node, and 50% are used to buy back and burn LPT. The more that the network is used, the more LPT leaves the supply. Node operators do not miss out on this compensation however - they are rewarded with newly minted LPT in proportion to the MFS algorithm described above, which is aimed to eventually meet equilibrium with the amount of burned LPT.

Emissions follow a fixed schedule

Today, inflation in Livepeer adjusts based on how much LPT is staked. In 2.0 emissions follow a fixed schedule. They start at today’s level of issuance, and they halve every two years. This means that the maximum LPT supply is known up front. In reality, any rewards not eligible to be claimed under MFS, are never minted and the supply is even lower.

As fees grow, burn catches up with new issuance. The litepaper models this burn mint equilibrium, and in modest growth scenarios it arrives somewhere in years 5 to 8. It can take longer if the token valuation runs far ahead of network revenue, or will occur faster if fees rise more quickly.

While the emissions schedule is set in stone in Livepeer 2.0, many governance parameters are exposed to manage the shift from largely rewards based incentives for network participants, to fee based incentives for participants. The parameters of the BME, such as how many fees are used to buy and burn LPT, what % is burned, what % flows to node operators vs the treasury, are all tunable via governance to meet the network’s needs. This is important, because the rewards and incentives available depend largely on the amount of fees, and the market cap of the network.

Read it and weigh in

While the Litepaper does lay out a set of concrete proposals to begin implementation, simulation, and testing, nothing is set in stone. The implementation and simulation constraints will inform the final designs, and nothing passes without governance and community approval. Read the Litepaper, share your comments in the forum, and be part of bringing Livepeer 2.0 into existence.