There are two ways to operationalize staking in crypto-economic systems. In one model, stake matters in relative terms, such that entities that stake more should be given more responsibility in the network. In another model, stake matters in absolute terms, such that entities that clear some threshold should be given as much responsibility as they can handle.
More subtly, the question around which model to use is tied to whether responsibilities are homogeneous or heterogeneous. If responsibilities are homogeneous, e.g. validators producing blocks for a Layer 1, then a model of relative stake generates more security. If responsibilities are heterogeneous, then a model of absolute stake is the only one that does not impair the network’s efficiency.
DoubleZero has both types of responsibilities. Network links are heterogeneous: a New York to London line is not substitutable for a New York to Tokyo line, regardless of the relative stake behind the line. But much of the overhead associated with the DoubleZero network — such as processing the DoubleZero chain — does involve substitutable compute, such that an entity that provides twice the stake should indeed carry twice the load.
This drives the duality of DoubleZero staking. Network contributors must stake some minimum number of tokens, after which their links are used freely for maximal efficiency. However, more traditional validators manage the DoubleZero overhead, with responsibilities proportional to the relative stake of themselves and their delegators.
Network contributors must of course stake tokens, as there are no guarantees of security without stake. In particular, each network contributor will have a required minimum number of tokens for its bundles of links. But a contributor does not need to stake tokens beyond this number for the link; and there is no advantage to doing so, as no rewards are paid to this stake explicitly. This allows the network to function leanly, from a capital standpoint.
The minimum number corresponds approximately to the importance of the links to the network; and as such, it should rise and fall, as the associated links become more or less important. This ordinarily would impose frictions on network contributors. However, the DoubleZero model proposes to set the stake needed for a network contributor’s bundle of links is exactly equivalent to the last k epochs worth of rewards. This — coupled with the operational detail that ongoing rewards are automatically deposited into stake — delivers all the desired properties. Above all, this lets a well-behaved network contributor to “set and forget” stake. Mathematically, a network contributor can never be below the minimum required stake as long as it has not been slashed. At the same time, it also allows links that grow in prominence to have larger stake requirements, and links that shrink in prominence to free up capital. Indeed, the network contributor can withdraw the excess deposited amount over the minimum stake with no penalty, which corresponds to it simply withdrawing lagged rewards or freed up stake.
Once the minimum thresholds have been reached and so the links are admissible to the network, staking will not be used as the basis of routing decisions or rewards calculations. This maximizes efficiency of the DoubleZero network. The alternative — by which stake influences or dictates routing decisions or rewards calculations — has some mildly appealing security properties, but it has horrible efficiency properties. Critical links may accumulate too little stake and thus have traffic throttled over it; or they may earn insufficient rewards and thus suffer underinvestment. Worse, since relative stake matters in this counterfactual world, an increase in stake at underutilized links will lead to a similar dynamic, even if there is no change in absolute stake tied to critical links. (These bad equilibria are self-healing over long periods of time by return-seeking capital, but the transitions can be long and the network performs poorly in the interim.)
Finally, while we focus on link staking, there should be an additional small amount of stake needed for identity purposes, at the contributor level. This is because the rewards methodology and burn parameter hinge partly on contributor identity, to establish redundancy value and to bound the attack on economic integrity respectively.
This section has largely focused on staking, but eventually the DoubleZero network will include slashing of that stake. Slashing can be motivated for two reasons. First, the links provided by a network contributor can be poorer quality than stated on four dimensions: lower reliability than stated, lower bandwidth than stated, higher latency or jitter than stated, or them censoring information. Second, network contributors may try to withdraw links during a period with an already-high amount of link exit. Neither of these is positive for the health of the network, and so both will incur penalties. This component of the model remains a work in progress for now.
Validators must similarly stake tokens to secure their contributions, either from their own account or from delegations by the DoubleZero community. However, there are no required minimums: responsibilities and rewards are proportional to their stake, relative to those of their peers. This is indeed the more traditional model used in blockchain.
Validators have responsibilities associated with their role. First, they must run the DoubleZero chain, which handles the overhead to manage the system in a decentralized way. Second, they manage the routing logic of traffic across the links. Third, they run the activator services to admit new links and offboard retiring links into the system. Finally, they compute the rewards that are owed to network contributors (and the rewards that need to be removed to preserve economic integrity) at the end of an epoch. In exchange for this provision of public goods, validators receive inflationary rewards.
The more traditional delegated stake dynamic also takes place here. DoubleZero community members can delegate their tokens to validators, who stake them to handle more responsibilities and earn more aggregate rewards (along with, for the validator, higher revenue through a commission on the delegated stake). This allows for broad participation in the core public good functions of the DoubleZero network.
Finally, this stake is naturally subject to slashing penalties if it acts adversarially, e.g. duplicate blocks, incorrect calculations for routing or rewards, censorship of new links, and so on. The exact details of this are also yet to be determined.
While the validator staking model may seem commonplace, the network contributor model is more unusual in crypto-economic systems. But there is a second benefit of this hybrid approach: it keeps the marginal cost of link provision low.
Permissioned systems have a short-term advantage over systems like DoubleZero, even if DoubleZero is more robust and resilient over the long run. Identity, reputation, and legal liability are substitutes for capital in building trust and honoring obligations. Contributors only need to install the hardware to participate in those networks; for DoubleZero, they also need to bring extra capital.
Thus, to keep DoubleZero competitive to network contributors in the short run, the staking model only requires enough capital to cover a few defections. Any requirement beyond that imposes costs beyond the benefits it delivers, and slows the growth of the network.