I think that’s basically the distinction I’m trying to get at. If the concern is intermediation and staking composition, total stake feels like a very indirect proxy for the actual problem.
Native delegation seems like a much more direct lever. I’m not necessarily convinced Ethereum needs native liquid staking too, but I do think the delegation primitive is worth reconsidering.
I opened a separate thread exploring that and whether native delegation could be structured so that commercial staking concentration does not automatically become additional consensus concentration:
Update (Aug 14): The mechanism in that thread has changed substantially again.
The earlier concentration-sensitive reward model, bond-weighted growth version, operator-family construction, quadratic concentration reserve, runoff mechanism and fixed leverage-cap version have all been removed from the core design.
The main problem I kept running into was identity splitting. If a protocol gives smaller pseudonymous operators better economics through a nonlinear per-operator rule, a large hidden operator can generally reproduce those economics by dividing itself across more identities.
The current construction takes the opposite approach: instead of trying to identify or economically punish large hidden operators, make the allocation rule itself neutral to identity splitting.
Native delegated ETH would enter a protocol-level pool and be routed by the protocol rather than by the delegator’s choice of commercial staking provider.
For validator i:
V_i = ordinary native effective balance
If the validator is NED-enabled:
B_i = V_i
otherwise:
B_i = 0
B_i is deliberately not “operator-owned stake.” Ethereum can observe native effective balance, but it cannot reliably observe beneficial ownership.
Let:
D_i = protocol-assigned NED delegated principal
The balanced routing target is:
D_i = u * B_i
Because the allocation is linear, splitting the same eligible base stake across additional validator identities does not increase aggregate target allocation:
Σ_i D_i = u * Σ_i B_i
So Ethereum does not need to determine whether several apparently independent validators secretly share an owner in order to make the routing rule split-neutral.
There is a second problem though.
Even proportional allocation can amplify a concentrated subset of the validator set if only that subset participates.
The latest version handles that with a Delegation Concentration Envelope, or DCE.
Let:
S = total ordinary active base stake
E = total NED-eligible base stake
D = total assigned NED delegated principal
and:
e = E / S
d = D / S
Normalize each participating validator:
b_i = B_i / S
y_i = D_i / S
For an eligible-base mass m, define:
C(m) = maximum delegated principal that can fit inside any eligible-base slice of size m
Operationally, this is a fractional-knapsack bound. Validators are ordered by D_i / B_i, highest first, and the protocol asks how much delegated principal can actually be packed into a base-stake slice of size m.
This replaces an earlier version that used only the single highest local leverage ratio.
That earlier bound was safe but too sensitive to one outlier. The DCE is tighter because a tiny high-leverage validator contributes only the delegated principal it can actually carry instead of its ratio being multiplied across unrelated stake.
Let κ < 1/3 represent the largest pre-NED base-stake coalition NED is required to prevent from being pushed across one third solely through delegated-weight amplification.
Define:
m = min(κ, e)
The design also includes a global effective-weight multiplier γ, normally:
γ = 1
Effective delegated consensus weight is:
Q_i = γ * D_i
Then require:
κ + γ * C(m) <= (1/3) * (1 + γ * d)
For any hidden coalition starting with at most κ of ordinary base stake, its eligible base share can be no more than m.
By construction of the DCE, its delegated principal can therefore be no more than C(m).
So its resulting consensus share is bounded by:
q <= (κ + γ * C(m)) / (1 + γ * d) <= 1/3
The protocol does not need to know which validator identities belong to that coalition.
The interesting consequence is that safe NED capacity becomes dependent on how broadly the validator set participates.
In the balanced state:
D_i = u * B_i
so:
C(m) = u * m
and:
d = u * e
For an illustrative protected threshold of κ = 32%, the concentration-only capacity is approximately:
| NED-eligible coverage | Maximum D / S from concentration |
|---|---|
| 40% | 2.86% |
| 60% | 6.67% |
| 80% | 20% |
| 88.89% | 50% |
| 90% | 60% |
| 92% | 92% |
So if only a narrow subset participates, NED is deliberately small.
As participation becomes broader, the amount of native delegation the mechanism can safely support increases.
At universal proportional participation, relative amplification tends to zero because the delegated distribution increasingly reproduces the validator set itself.
The shorthand I’ve been using for this is:
Flanders earns scale by earning coverage.
The proposal also now separates three different risks rather than trying to control all of them with one leverage parameter.
The DCE bound handles hidden-coalition concentration.
A separate local limit:
D_i <= ℓ * B_i
bounds delegated principal per unit of eligible native stake.
And a global limit:
D / S <= Λ
can bound total NED exposure even when near-universal participation makes the concentration constraint extremely permissive.
So the direction of the proposal has changed quite a bit.
It is no longer trying to make large staking providers progressively more expensive or force them to decentralize themselves.
A provider could remain commercially very large.
What its customer base would no longer provide automatically is an equivalent share of native delegated consensus weight.
In shorthand:
commercial staking share != native delegated consensus allocation share
A provider that wants more NED allocation still needs more NED-eligible native base stake, and the resulting distribution remains subject to the full-network concentration bound.
There are still substantial implementation questions.
The current reference direction treats:
W_i = V_i + γ * D_i
as the validator’s NED-weighted consensus balance for the stake-weighted roles NED participates in.
That avoids creating separate attestation and proposer stake distributions, but it means execution-layer proposer revenue cannot all be forced back into the NED pool. The current design leaves that execution-layer revenue with the operator as an explicit operator rent rather than pretending the protocol can reliably measure and redistribute all MEV.
NED yield may therefore be lower than the full economic return of direct validation or some LSTs. I think that is an adoption question that needs to be modeled rather than hidden with a subsidy.
Withdrawals and slashing also require a bounded-liability model.
Ordinary Ethereum slashing evidence can surface well after the underlying message was signed, so NED cannot simultaneously provide unbounded historical delegated liability, finite final withdrawals with no clawback and exact assignment of every arbitrarily late loss to the original pool participants.
The current reference design therefore gives the delegated component a finite NED slashing claim window:
W_NED = 8192 epochs
A timely NED slashing notice keeps the relevant pool liability open. A withdrawal cannot settle until its delegated exposure has retired, the claim window has passed and any timely pending notices have resolved.
There is still a lot of consensus-layer work before I would call this EIP-ready, including parameter selection, Pyspec lifecycle/accounting, effective-balance integration, accountable-safety analysis for exceptional γ reduction, sync-committee semantics, delegated slashing penalties, historical proof costs and adoption/issuance modeling.
It also does not solve hidden beneficial ownership of ordinary validator stake or stop custodians from staking customer ETH outside the mechanism.
But I think the narrower question is becoming clearer:
If the concern is staking composition and intermediation, should delegated consensus weight itself be the thing we design around before changing aggregate staking yield for everyone?