barnabe4d

Many thanks to @fradamt, @casparschwa, @vbuterin, @hxrts, @Julian, @aelowsson, @sachayves, @mikeneuder, @DrewVanderWerff, @diego, @Pintail, @soispoke, @0xkydo, @uri-bloXroute, @cwgoes, @quintuskilbourn for their comments on the text (comments ≠ endorsements, and reviewers may or may not share equal conviction in the ideas presented here).

We present rainbow staking, a conceptual framework allowing protocol service providers, whether “solo” or “professional”, to maximally participate in a differentiated menu of protocol services, adapted to their own strengths and value propositions.

 Why rainbow?We intend to convey that the architecture below is appropriate to offer services provided by a wide spectrum of participants, such as professional operators, solo home stakers, or passive capital providers.Additionally, the spectrum 11 may remind the reader of eigenvalues, while we intend for this proposal to partially enshrine a structure similar to re-staking.

TLDR

Operator–Delegator separation

Discussions around staking have pointed out the natural separation between node operators and capital delegators, inherited from the perennial distinction between labor and capital. The separation is natural. Many parties wish to obtain yield, and yield is generated from the issuance rewarding participants who place their assets at stake to secure FFG, Ethereum’s finality gadget, and more largely to operate Gasper, Ethereum’s consensus mechanism. Along with placing assets at stake, the work of validation must be properly done to obtain rewards (more yield) and avoid penalties (less yield). This validation work is however costly, and may be performed by operators on behalf of delegators. The operator set is decomposed in two classes:

Operators as part of a Liquid Staking Protocol (LSP) may offer to issue a liquid claim for the delegators, known as Liquid Staking Tokens (LSTs). These LSTs represent the principal which delegators have provided, along with the socialised rewards and losses collected by operators, net of fees.

We review recent proposals to enshrine the separation of operators and delegators further into the protocol. Such attempts naturally target the formation of “two tiers”, an operator tier and a delegator tier (see Dankrad 5, Arixon 2, Mike 1 on the topic). A recent idea suggests that by capping slashing penalties to only the operator’s stake, assets of delegators are no longer at risk.

rainbow-staking-02802×1152 212 KB

If this sounds too good to be true, then why should delegators earn any yield? Vitalik lists two possibilities in his post, “Protocol and staking pool changes that could improve decentralization and reduce consensus overhead 3”:

In other words, delegators in this model do not exactly contribute economic security to FFG, their stake being non-slashable, but they are able to surface discrepancies in the gadget’s functioning. Their denomination as “delegators” remains somewhat contrived. We see three issues with the model of two-tiered staking as presented above:

Heavy and light services separation

To resolve these issues, we take the view here of two distinct types of protocol services, each type inducing within itself a market structure of delegators and operators. The fully unbundled picture becomes:

rainbow-staking-13302×1152 278 KBThe two categories are functionally different, but both feature the Delegator–Operator Separation

Economics of heavy services

Economics of light services

rainbow-staking-23302×1152 350 KBBy bonding light operators, penalties may be doled out to service providers with bad quality of service.

 Link to re-staking"Provision of validation services” sounds eerily familiar. We now draw a direct link between the discussion above and re-staking.

When a party re-stakes, they commit to the provision of an “actively validated service” (AVS). In our model, we may identify different types of re-staking:

So we claim that the model above is a partial enshrinement of re-staking, in the sense that we determine a “special” class of protocol AVS for which rewards are issued from the creation of newly-minted ETH. We then allow holders of ETH to enter into the provision of these services, either directly as operators, or indirectly as delegators.

Staking economics in the rainbow world

This section presents a general discussion of the framework and its implications on the Ethereum network’s economic organisation.

We regard Gasper (and in the future, a version with Single-Slot Finality, SSF) as the heaviest protocol AVS, one that is rigidly constrained by its network requirements (e.g., aggregation and bandwidth limitations), and which receives the highest share of aggregate issuance minted by the protocol. As such, its locus attracts a great many parties wishing to provide the stake which the service demands for its core functioning, while inducing market structures of long intermediated chains of principal-agent relationships 2.

Yet, it is unnecessary to offer unbounded amounts of stake to the Gasper mechanism in order to make its security claim credible. In addition, given the induced intermediation of stake, it is necessary to prevent the emergence of a single dominant liquid staking provider collateralised by the majority of the ETH supply. This means that measures such as MVI are critical to target sufficient security, by creating sufficient pressure to keep the economic weight of Gasper in the right proportion. However, MVI drives competitive pressures ill-suited to solo stakers, who are, besides, mostly unable to issue credible liquid staking tokens from their collateral and have thus low capital efficiency.

What is the role of solo stakers in our system? Their economic weight means their group is not pivotal to the Gasper mechanism. In particular, they cannot achieve liveness of finality by themselves, which requires 2/3 of the weight backing Gasper. Solo stakers find solace in two core value propositions which they embody ideally:

rainbow-staking-31920×892 155 KB“Preference entropy” denotes the amount of information surfaced by protocol agents to the protocol. Agents who censor have lower preference entropy, as they decide to restrict the expression of certain preferences, such as activities which may contravene their own jurisdictional’s preferences. Collectively, the set of solo stakers who operate nodes to provide services is highly decentralised, and is thus able to express high preference entropy. This economic value translates into revenue for members of the set.

This economic organisation redraws the lines of the MVI discourse. Today’s 1-D MVI optimises for competitive pressure to keep the weight of derivative assets such as LSTs in check (”minimum”), while under the constraint of safeguarding the presence of participants with lower agency over this specific mechanism, FFG (”viable”). With rainbow staking, we offer an alternative 2-D MVI where solo stakers are maximally effective participants on their own terms, achieving the goals which many have signed up for: Competitive economic returns with the ability to realise their preferences into the execution the chain, e.g., as ultimate backstops of chain liveness or as censorship-resistance agents. 2-D MVI no longer questions the ideal conditions of aggregate issuance with all participants lumped in a single battlefield, warring over a single scarce flow of revenue. 2-D MVI questions instead the adequate amount to allocate towards the remuneration of heavy services on one side and light services on the other.

We offer a strawman proposal for 2-D MVI determination. The first step is to port the current MVI thinking to the remuneration of heavy services, i.e., decide on an issuance curve compatible with a rough targeting of stake backing heavy services 4. This curve determines a level of issuance IH given some proportion of stake d behind heavy services. The second step, given this level of “heavy” issuance, is to decide the issuance dedicated to light services. This could be determined with a simple scaling factor, i.e., set IL=αIH for some small α (note that today, α≈3.6%). The level of aggregate issuance is given by I=IH+IL. The two questions 2-D MVI would be required to solve are then:

To provide orders of magnitude, issuance over the last year amounted to I≈ 875,000 ETH, close to 2 billion USD at current prices. If even α=5% of the yearly issuance was offered to the good performance of solo stakers for light censorship-resistance services, while 95% was offered to the heavy operator set, the set of light participants would collectively earn IL= 100 million USD. Assuming an investment of about 1,000 USD over the lifetime of a solo staker node (say 5 years, including upfront and running costs), we find that 100 million USD is enough to bring to marginal profitability up to 500,000 solo stakers as light operators. While this provides a generous upper bound in a world where all light service providers are also solo stakers, with no light delegators, we believe this order of magnitude is helpful to frame the terms of the debate.

rainbow-staking-41502×1252 220 KBProfessional operators are well-suited to provide heavy services due to their sophistication along many dimensions (economies of scale, capital requirements, knowledge, reputation). Meanwhile, solo stakers are well-suited to provide light services due to their high preference entropy, surfacing idiosyncratic signals to the protocol. The two groups overlap where heavy services earn credibility from the participation of solo stakers, or where light services earn efficiency from the participation of heavy operators.

Rainbow road

The next sections tackle individual facets of the rainbow staking framework, providing more colour. They are not essential to the text, and may be skipped by the reader in a hurry.

rainbowroad978×542 70.7 KB

Mapping the unbundled protocol

We offer the following map to represent the separation of services in our model:

rainbow-staking-52802×1852 479 KB

The map emphasises the resources provided by each type of operator. We also unbundled the “execution services” from the “consensus services”. To achieve MVI for the heavy operator layer, it is critical to limit reward variability, and “firewall” the heavy operators reward functions from rewards not accruing from their role as heavy service providers.

While Execution Tickets (ETs) may not ultimately be the favoured approach to separate consensus from execution, we believe their essence represents this separation well. In a world with ETs, where MEV (a.k.a., the value of the block production/payload service) is separated from consensus, heavy operators are less concerned with reward variability, timing games or other incentive-warping games which may occur. The map also shows that the payload service receives (at least) two inputs from the consensus services above: Heavy services finalise the outputs of the payload service, while light services constrain payload service operators by e.g., submitting them to inclusion lists (ILs), constructed along the lines of EIP-7547 or using different methods such as Multiplicity gadgets 4.

But how exactly do outputs of light services such as ILs constrain execution payload producers? In today’s models of inclusion lists, validators are both the producers and the enforcers of the list. Separating services, the light operators become responsible for the production of the list, but we still require the heavy operators to be its enforcers. Indeed, validity of the execution payload with respect to Gasper is decided by heavy operators, who attest and finalise the valid history of the chain, ignoring invalid payloads. We see two options to connect production and enforcement:

Additional analysis is required to understand the architecture of out-of-protocol services such as pre-confirmations or fast(er) finality (EigenLayer whitepaper, Section 4.1), in relation to the rainbow staking framework. Some of these services may still benefit from overloading heavy operators with additional duties, so while this is not a total elimination of all rewards outside of protocol rewards, it is a great mitigation participating towards making MVI more palatable.

 Would it be worth unbundling further? The payload service is itself a bundle of two distinct services: “Execution gas”, allocated via ETs, and “data-gas/blobs”, allocated via some DA market. The separation was discussed at a few points in the past. First, as a way to achieve better censorship-resistance for blobs, by creating a “secondary market” focused on blobs. Second, as a way to prevent timing games impacting the delivery of blobs. Practically, a mechanism akin to partial block delivery may be required for the consensus services to finalise the delivery of both parts of the full block, with blob transactions (carrying commitments to blobs and some execution to update rollup contracts) included in an End-of-Block section. While further analysis is required to understand the costs and benefits of this separation, we believe this direction will become increasingly relevant.

Towards SSF

We start the discussion with the most special protocol AVS: Gasper, the consensus mechanism of Ethereum. Its reward schedule is well-known, as well as its limitations regarding the size of its validator set, measured in “individual message signers”, and not in “stake weight”. To achieve Single-Slot Finality (SSF), a drastic reduction of the validator set size is required, but how can we achieve this without booting out the least valuable participants, as measured in units of stake-weight-per-message?

Vitalik’s post “Sticking to 8192 signatures per slot post-SSF: how and why” details various approaches, including the first model presented above (Approach 2). Some trade-offs are difficult, asking the question whether to go all in on decentralised staking pools (Approach 1), or rotating participation to allow for a larger set in aggregate, if not in constant participation (Approach 3).

We advocate for something mixing both Approaches 1 and 2 (Approach 1.5!)

The question of allocating one of the 8192 SSF seats to operators remains (note that this 8192 number may not be the absolutely correct one). Incentives may need to be designed in order to discourage large operators or protocols disaggregating and occupying more seats in an attempt to push out their competition. These incentives may once again favour more capital efficient players, who can amortise the cost of a seat over a larger amount of stake they control and revenue they receive. The number of available seats is constrained by the efficiency of cryptographic constructions such as aggregating signatures, but as cryptographic methods or simply hardware progress, the number of seats may increase, loosening the economic pressure to perform as a seat owner. Yet, the pressure of MVI keeping overall rewards to heavy operators low remains.

The heavy layer’s layers

Are we ruling out solo stakers from the heavy layer entirely? I do not believe we are. A re-phrasing frames well the terms of the debate, by calling solo stakers operating in the heavy layer solo operators instead. What we want here is for a low-powered participant to still perform a meaningful role in the heavy service.

Enshrining an LSM-like system, which looks a bit like Rocket Pool, already makes a significant dent into the requirements. We take the number of “seats” in SSF as given by Vitalik, i.e., up to 8192 participants. Assume that we intend for 1/4th of the total supply of ETH 4 to be staked under SSF, which is roughly 30 million ETH today. We would then require a single participant at the SSF table to put up about 3662 ETH, or about 8 million USD. Gulp!

With an LSM-like enshrined mechanism, we favour the emergence of partially collateralised pools (partial pools). Rocket Pool requires a minimum bond of 8 ETH for 24 units of delegated stake (1:3), while the LSM defaults to 1:250. Even a (more) conservative choice of 1:100, in-between the two, brings us back to an operator needing to put up about 36 ETH, roughly 4 ETH more than the minimum validator balance required today. Regardless of the limit allowed by the LSM defaults, an LSP may require a higher collateralisation ratio from their operators, e.g., Rocket Pool could still force the use of 1:3 shares.

Could we go further? Running the gamut of existing LSPs, Diva coordinates solo operators with low capital via Distributed Validator Technologies (DVT). A sub-protocol of Gasper could allow for the creation of DVT pools, where a virtual operator would behave akin to a solo operator does under the enshrined partial pool model. The virtual operator meanwhile requires DVT slots to be collateralised by partial solo operators, and the Ethereum protocol could offer the ability to condition the validity of a Gasper message on the availability of a DVT’d signature performed by a sufficient quorum of partial solo operators.

Regardless of how the solo operator is synthesised (either directly under the enshrined partial pool model, or via quorum under the enshrined DVT model), the solo operator is then provided with the ability to accept deposits from heavy delegators. Heavy delegators then tokenise their deposits as shares of an LSM-like system. These shares are received as assets by an LSP, according to the LSP’s own internal rules on the composition of its basket of operators. In exchange, the LSP mints a fungible asset, better known as an LST. Voilà!

Dynamics here are interesting. An LSP wishes for their LST to be credible, i.e., an LST is not credible if holders do not believe that the operators holding their principal are good agents 1. In the worst case, malicious operators may seek to accumulate as much stake as possible in order to perform some FFG safety fault, at the cost of getting slashed and devaluing the LST. An LSP curating a basket of good operators mints a credible LST, which has value to its holders.

Gadgets such as DVT bolster this credibility, as long as entropy of the DVT set can be ascertained. Importantly, this is not something that the Ethereum protocol needs to concern itself with. The protocol may not be able to procure itself an inviolable on-chain proof that a DVT set’s entropy is high, for instance because it is composed with unaffiliated solo stakers. Yet the open market may recognise such attributes, and value the LST derived from an LSP with good curation more than an LST with poor composition. While participating in the heavy layer may not be the most effective vehicle for solo stakers to express their agency on the network, the optimist in us believes this is a potent avenue and one that just makes good business sense 3.

Open questions