Delegators 2.0 - Topic Specific Thread

Research & Protocol Improvements
1

This is a thematic topic thread related to the proposed Livepeer 2.0 Protocol Updates. There will be many such threads in the coming days and weeks to dive into each topic in detail.

This thread covers the role of delegator, and impacts on their economics in Livepeer 2.0.

The role of delegation in Livepeer has always been to participate in the quality assurance and security of the network. They stake token in order to route work towards effective nodes, and in return they earn a share of inflationary token rewards and fees. In many ways, these concepts hold in 2.0 - though the underlying mechanics, and potentially the economics, change a bit.

In Livepeer 2.0, delegators directly stake towards validator candidates. The top validator candidates by stake are elected into the active set. The validators earn inflationary LPT incentives for participating in a protocol that determines to what extent worker nodes are reward eligible. Since validators play such a big role in incentivizing and rewarding a highly performant, diverse, scalable network, it’s critical that an honest and high performing set of validators are prioritized - and hence delegator incentives exist to play the key role of securing this set via stake commitments.

As a starting candidate, the exact same bonding, unbonding, 7-day unbonding period, reward cut mechanics can be used to elect the validator set, with delegators playing the same exact role.

How might delegator economics change?

The big change is in how LPT rewards are split. They are currently split to orchestrators according to stake (who share a portion to their delegators), and the treasury. In 2.0, they’ll be split to node operators based on fees, the validators (who share with their delegators according to stake), and the treasury. The validator/node/treasury split are parameters to be determined, but there seems to be consensus that we want the majority of rewards flowing to node operators to incentivize the actual high quality of service for the utility the network provides and as part of the BME model.

If a big portion of rewards, let’s say 75% as an example, get redirected to node operators, doesn’t this dramatically reduce the delegators potential yield? The answer is potentially yes and potentially no. First let’s look at a reframing of the Livepeer Token (LPT)…

Imagine two extremes for token holders who are considering being delegators.

  1. On one extreme, there is no connection beween LPT value capture and rising network usage. Inflationary LPT rewards delegators with a high perceived yield, but are also just diluting the $/LPT because so many more LPT are being issued, and even rising fees don’t flow value direct to LPT in a strong way. Delegators earn a lot more LPT, but price of LPT may constantly go down.

  2. On the other extreme, there’s a strong connection between LPT value capture and rising fees. But there’s no inflationary LPT issued in yield for delegation. Here delegators earn almost no additional LPT, but they do expect the price/LPT to rise as the network is used.

Of course the current protocol and new 2.0 proposed protocol updates sit between these two extremes. Currently the protocol has slipped closer to #1, and a constantly declining LPT value doesn’t serve anyone, even with high yield. The proposed changes for 2.0 bring it closer to #2. Of course there’s still room for inflationary LPT incentives/yield, as there is a key role that delegators are playing. But if it’s lower, under the belief that value capture is achieved through appreciating LPT, is that a better outcome? I would argue yes, though this shifts the perception of delegation from “I stake this to earn a high LPT-denominated yield” to “I hold this token because it will appreciate as fees rise, and if I delegate it, I am earning incremental yield rather than zero.” It’s a cherry on top. What might that incremental yield be?

It depends on participation rate. In the example of 75% of fees flowing away from existing the existing stake-based allocation, one might expect delegator rewards to drop by 75%. However its likely that the participation rate may drop. One reason is that every node that wants to operate and compete for work will need to post a fixed-bond. This means LPT flows away from stake and towards the node operator set. This is, after all, where the majority of rewards will be directed. On the other hand, the reduced yield may mean some delegators decide to merely hold LPT rather than stake.

If the participation rate in the validator set drops by 50%, then the per delegator yield actually doubles.

Given the above (completely arbitrary) examples of 75% of rewards flowing to fee-based incentives and the validator participation rate dropping by 50%, delegator LPT rewards would actually be cut in half. Though if LPT value appreciates with rising usage rather than drops, then the impact may be minimized or even net positive. Is this desirable or not? Let’s get some feedback. Many have advocated that delegators are being significantly overpaid with inflationary LPT by the protocol in the current scheme.

As mentioned, it all comes down to many market determined factors - the participation rate, LPT value, reward-cut competition to attract delegation in the validator set, and more. Ultimately, if you look at delegators as token holders, then they get the benefits of both appreciation with increased network usage AND yield → and that should be viewed as favorable, relative to the status quo.

Open Questions

In my mind, the biggest open question is: should delegators be able to post the fixed-bond on behalf of willing node operators, and therefore get access to the LPT derived from their fees?

There are several pros and cons to this. You would see a LOT more nodes, and a more dynamic fee share marketplace develop for node operators to attract delegator staked-bonds. However there would be less of a reason for node operators to BUY LPT in order to participate in the competition to earn fee-derived rewards in the BME, since they could just attract existing stake from the market. And they would have less incentive to do honest work, as they’d be putting someone else’s stake at risk of a 90-day freeze with no rewards, rather than their own. Node-based bonds would compete with validator stake, which is needed to secure the validator set from attack. The less stake on validators, the lower the cost to buy the majority of validator slots and affect reward allocation.

Do you leave this outside of the protocol, and let a market develop for node operators borrowing LPT from holders and paying them yield in order to run more nodes and attract more work and rewards? Or do you put it directly in the protocol via letting delegators post the bond and earn a share of the rewards through a fee cut? Let’s get some feedback.

Another big open question is what happens to existing stake during the migration to 2.0?

Migration is a big topic to be picked up in another thread. I’d lean towards the existing stake and orchestrator set inheriting the initial validator slots as a starting point, so that rewards continue and no major delegator migration/action is needed - since that stakeholder group is too passive. However this can be discussed.

1 Like
Node Operator 2.0 - Topic Specific Thread
2

Here are a few delegator questions from the Discord Chat from Breaker. Let me address them here for reference and follow on discussion.

  1. Do delegators participate in the fee-proportional stream? The litepaper states node-operator rewards are “delivered in a liquid and unbonded state in proportion to fees that the node earns.” Is a share of that stream enforced on-chain for the node’s delegators (something like today’s rewardCut/feeCut), left to each node as a purely commercial choice, or not expected at all — with delegator compensation limited to the stake-allocated validator portion (“minimum viable compensation”)?

This is a big open question as identified in the original post above. There are pros and cons of allowing delegators to post the fixed bond for node operators and access a share of the fee-derived rewards. I have opinions, but I’d like to hear some community reactions and feedback first, before I share.

  1. Order of magnitude. If delegators only earn from the validator portion: what rough split of total issuance is envisioned between the validator set and node operators (“the majority”)? Even a range (10/90? 30/70?) would let delegators model expected yields.

Further modeling is necessary in order to determine the right split here. The big question is how much do we need to pay validator/delegator set in order to incentivize high quality performance in the role, and have enough security such that the cost of validator set takeover is high enough? The treasury also gets a cut.

I used an example of 70/25/5 before across these 3 entities, with validator set getting 25% of inflationary rewards. But if the participation rate drops within that set, and LPT value rises in correlation with fees, then this counteracts the yield drop on the rewards themselves, and could lead to comparable yield earned by delegators as today. It’s ultimately market determined by many factors.

  1. What remains of the incentive to attract stake? Today orchestrators share generously because delegated stake routes work to them. In 2.0, stake elects validators but no longer routes jobs. Beyond securing a validator slot, what economic reason remains for a node to attract — and pay for — delegated stake? Is there a risk cuts drift upward once stake no longer drives fee income?

First let’s acknowledge that stake-proportional work routing was a good theoretical idea in early transcoding centric 1.0, but it got weakened in practice by the high QoS requirements of live transcoding, and even more by diverse AI job types. So more stake == more work has significantly loosened and almost doesn’t exist today. That’s a problem.

In this 2.0 model where there’s a fixed bond/node, more nodes == more work opportunity. AKA node operators buying more LPT means they can attract more work and earn more rewards. (Theoretically - there are arguments that a single node can perform as much work as two nodes, though I’d say that the agentic selection algorithm will have randomization and favor redundancy, so more nodes earn more work).

But as it relates to delegation, why would a validator want to attract more stake? First of all, they get to keep a cut of their delegators rewards through the reward cut. The more stake, the more take home rewards for a validator, and the more they can invest in their validator operation - testing harnesses, agents, transparency tools, etc. Its delegators job to balance the long term health of the network (and what’s good for LPT), with short term yield through reward cut optimization. So it’s in their favor to elect a diverse set of high performing validators and pay them in reward commission. Secondly, more stake entrenches the validator’s slot in the scarce validator set.

High performing validators who have a lot of stake on their own may set their reward cut to 100%, and not attract external stake. That’s fine. Someone will be competing to outperform them and take their slot by offering a better service with a better delegator return.

  1. Unbonding scope. The extended unbonding period (90+ rounds) is motivated by “penalizing misbehaving nodes with long capital lockup.” Does it apply only to node/validator bonds, or to ordinary delegators as well? If passive delegators keep a short unbonding period, that materially changes the liquidity calculus for LPT holders.

Delegators on the validator set keep a 7 day unbonding period. No change in their liquidity risk calculation.

Node operators have extended period on their fixed-node-bonds, since they need to be disincentivized from harming the network, through a long capital lockup.

  1. Migration. At transition, what happens to existing delegations — do they carry over automatically to the same operator (as validator), and under which unbonding regime during the migration window?

Migration is a big topic to be designed, and these are good questions. No firm answer yet, though I favor easy continuity since it’s unlikely we’ll get all delegators to take action in a short time window.

1 Like
3

Following up on the concentration angle since I think it’s worth spelling out more explicitly. The fixed bond per node addresses who can register an identity on the network, but nothing in the proposal caps the actual GPU capacity a single node can expose. That matters because, as things stand today, one operator can already run the majority of the network’s GPU capacity behind a single node identity, without that having any relationship to their stake. Moving more of the reward weighting toward fees actually served, rather than stake, doesn’t change that dynamic, it reinforces it. If an operator already captures most of the real demand because they have the most capacity, then shifting the majority of issuance toward fee based rewards just means they capture the majority of that issuance too.

And that’s where I think the real risk lives, not in a static 50 percent share, but in the flywheel it creates. That operator earns the bulk of the LPT rewards because they serve the bulk of the fees, that LPT (or its value) gives them the means to rent or buy even more GPU capacity, more capacity lets them serve even more demand, and the cycle repeats. None of the four proposed mechanisms puts a brake on this. BME, validator scoring, the fixed node bond, the long unbonding period, all of them tie rewards more tightly to real work performed, which is exactly right for solving the ghost node problem, but they do nothing for the opposite failure mode: an operator who is working legitimately, at scale, and simply gets structurally bigger every round as a result.

The validator design is built to catch dishonest or non-productive nodes, but a dominant operator doing genuinely honest work would score perfectly fine under that system, there’s no mechanism looking at market share itself. It feels like we’re solving for decentralization of honesty without necessarily solving for decentralization of capacity, and those are different problems.

A couple ideas worth putting on the table for the topic specific threads: something like j0sh’s min(fee share, stake share) formula, but reframed around an explicit cap on the share of network fees or rewards any single entity can capture rather than just tying rewards to stake. Or alternatively, some kind of diversification constraint built into how Livepeer Agent routes work itself, capping the share of volume any one identified operator can receive over a given window, even if that’s not the locally optimal routing choice in the short term.

Would like to hear if this has already been thought through and I’m missing something, but as written I don’t see anything in the current design that prevents the network from ending up more centralized in practice than it is today, even if every individual actor is behaving honestly.

2 Likes
4

Thanks for the comments. Yes I agree we should start a topic specific thread on the min(fee share, stake share) idea. It’s been suggested many times, even going back to the project’s origins, that a node should have to have X% of stake to claim up to X% of fees as a driver for LPT demand. However it makes it provides incentive for a node to stop working as soon as they earn their stake share in fees, (which is technically bad because someone has to outperform their stake to counterbalance any underperforming node who doesn’t perform up to their stake).

Regarding your bigger point on centralization…

I would argue two things…

  1. It is NOT bad for the users of the network if the biggest operator, offers the lowest price, the highest reliability, all the capabilities, and meets their needs on scale. If no other node can compete on any of these vectors, no user should be forced to choose another node. You may not like it from a decentralization perspective, but the network has delivered the best possible experience to users.

  2. But that is impossible for a single centralized actor to do that. Decentralization brings many advantages including localization and low latency to the user, price competition against that centralized provider, niche services that the centralized provider can’t operate on, reduction of single points of failure seen in the centralized provider, redundancy options, and more. Agents will be better at identifying these advantages in realtime than humans will be - so if your node delivers a better experience on any vector here that an agent prioritizes, it will get work.

Or alternatively, some kind of diversification constraint built into how Livepeer Agent routes work itself, capping the share of volume any one identified operator can receive over a given window, even if that’s not the locally optimal routing choice in the short term.

While I don’t think the protocol can enforce this, I do think that it’s part of the client-side logic of the Livepeer Agent (or other vertical specific agents built to use the capabilities of the network) that they WOULD prioritize this. If they’re optimizing for redudnancy and fast failover, they’ll need to have multiple nodes in their selection queue. They’ll need to be evaluating multiple nodes through randomization and learning. They’ll switch quickly if they see a delay or capacity issue on a single node, etc.

Lastly, if validators really really really want to prioritize decentralization over QoS, they have the option to reduce the reward multiplier score for a large operator. I’m not sure this would be justified and helpful to the network, but if the centralization were deemed harmful, it’s an option. Dropping their score to say 0.95 would still route a ton of rewards, but incentivize them to post more fixed-node bonds for diverse capabilities or locations or whatnot at 1.0 scores than just back all the work from one node with a 0.95 score. Still, at the end of the day, I don’t think nodes that are worse on all vectors, and therefore providing no unique utility to the network at all, and catching no usage, should be entitled to rewards just for showing up. I think it’s likely all well intentioned nodes will find their niche or vector to compete on vs any big centralized node operator however.

2 Likes
5

Looking forward to that topic-specific thread :slight_smile:

My idea was actually the other way around: an orchestrator’s per-round rewards would be capped, not the fees. So: reward % = min(fee %, stake %) That is not a (dis-)incentive to stop processing jobs - fees are uncapped - but it is an incentive to make themselves more attractive to delegation in order to earn even more, in LPT (or they’d have to buy more to self-stake). This seems like it’d lead to a much more dynamic and interesting mix of fee-reward shares and more delegator movement as they chase yields. That also reinforces the original thesis of delegation driving network activity.

I could go on, but will save it for the topic thread. However, it would be a much smaller protocol change than spinning off a new validator set and staking towards validators, etc. The current proposal seems to relegate delegators to mostly-idle speculators which is somewhat disappointing.

I’m a delegator, and the reason I don’t move stake around more often is because it wouldn’t actually do much right now. Validator staking diminishes that role even further (eg, no fees). Delegation was one of the more interesting yet under-baked aspects of the original protocol. I’d love to explore that more if we can, rather than port over the role in an even smaller capacity.

2 Likes
6

Thanks for the detailed answers, and for bringing them into the forum — the 7-day clarity for delegators removes the main reason many of us would have considered unbonding ahead of the transition. That alone buys retention.

On the big open question — can delegators post the fixed-bond?

I’ll be honest that as a delegator I have a direct interest in “yes,” but the cons you list are real and shouldn’t be waved away. Two especially: losing the node-operator buy pressure defeats part of the point of BME, and putting someone else’s capital at risk weakens skin-in-the-game.

A middle path that might keep the upside without the downsides: allow delegator-funded bonds, but only as a minority portion of each node’s bond, and with the node’s long unbonding period applying to that delegated portion too.

  • A cap (say ≤50% delegator-funded) keeps genuine operator capital at risk — skin in the game preserved, and operators still need to buy LPT at the margin.
  • Applying the long lockup to delegators who opt in creates two clean, self-selecting classes: validator delegators (7 days, modest yield, security-focused) and node-bond delegators (long lockup, fee-derived upside, risk-taking). Each prices its own risk, and nobody is forced into the illiquid tier.
  • It also answers the validator-security concern: the illiquidity premium makes node-bonding a deliberate choice rather than a default drain on validator stake.

Compared to leaving it fully outside the protocol: an informal lending market puts counterparty risk on holders with no protocol enforcement, and the yield it pays would be opaque. In-protocol, the terms are visible and the risk is explicit.

There’s also a capacity-side argument for it. The operators most likely to be capital-constrained are small ones with real GPU capacity but little LPT — exactly the long tail a fee-weighted model risks starving. If bonds can only come from an operator’s own balance sheet, the viable operator set is filtered by who already holds LPT, which correlates strongly with who is already large. Letting delegators fund bonds (within a cap) widens the operator set rather than concentrating it, and gives delegators a way to back network diversity with capital rather than only with votes.

On min(fee share, stake share) — I’d back Josh’s version, and specifically for delegator reasons.

The clarification matters a lot: capping rewards rather than fees is the right split. Fees are a market verdict and shouldn’t be touched — telling the best provider it can’t serve demand it won just pushes users off the network, and equal treatment of unequal contributions is a tax on performance. Issuance is different in kind: it isn’t paid by customers, it’s paid by every holder through dilution. So “should the best operator earn more?” (yes — through fees) and “should everyone’s dilution fund the compounding of a lead that’s already been won?” are two separate questions with separate answers.

What makes Josh’s formula interesting from a delegator seat is the second-order effect. Under the litepaper as written, stake no longer routes work and no longer lifts any ceiling — so a node has no economic reason to want my delegation, which is precisely how delegators end up as mostly-idle speculators. Under min(), a node that out-earns its stake share has a direct, quantified reason to attract delegation: it’s the only way to lift its reward cap. That turns delegated stake into a scarce resource nodes compete for, which is what gives delegators both a role and bargaining power. It’s also the LPT demand driver mentioned above — nodes either attract stake or buy it.

One refinement worth considering, if the concern with a hard min() is the flat region above stake share: a smooth version, e.g. reward share ∝ (fee share)^α × (stake share)^(1−α), normalized across nodes. α=1 is the current proposal (pure fee weighting), α=0 is today’s stake-only model, α=0.5 is a balanced geometric mean. It preserves Josh’s core property — more stake always lifts your rewards, so delegation always matters — while removing the kink: every marginal unit of fee-earning still raises issuance, just with diminishing returns, so there is never a region where additional work earns zero additional LPT. Governance gets one legible dial to tune how much issuance follows work versus capital at risk.

It also happens to address the concentration concern without any explicit cap or forced equality: the dominant operator remains the best-paid node on the network in both fees and issuance, but the protocol stops paying an accelerating subsidy on top of a position that’s already established.

On the reframing — “yield as a cherry on top rather than the reason to hold” — I agree that’s the healthier framing, and it matches how I’d want to hold LPT. The caveat is that the cherry still has to clear a bar, because delegating means committing capital that could sit liquid instead. Your 70/25/5 example lands delegator yield somewhere in the low double digits depending on participation and the new issuance curve — which is a workable zone. Much below that, and rational holders simply stay liquid, which starves exactly the validator stake the set needs for security.

1 Like
7

FYI, I’ll be hosting an open community discussion on the Delegator 2.0 topics on Wednesday, July 29th, and 2:00pm ET in the Discord. If you’re interested in these topics, please come prepared to share feedback and discuss in detail: Discord