How would this work in practice? Solvers give attestations to txs they see?
Attestations are one way (in particular to vet that settlements don’t execute any malicious interactions with the funds stored in the settlement contract), but even just having governance token holders run a some client we proveide to detect misbehavior and launch a slashing proposal in the DAO could be an option.
Separating solver responsibilities like you suggested is an interesting thought (which is more related to our vision for https://forum.gnosis.io/t/gpv2-road-to-decentralization). One reason why we wanted to have each solver also run independent infrastructure is that - unlike in year strategies - solvers don’t run on Ethereum but in some local setup (AWS, self hosted, etc). By having fully independent solvers the protocol should be more resilient (in the past we have seen that e.g. a node misconfiguration on our end caused a downtime of the entire protocol).
Would have some clear rules for when should someone be slashed, might be contentious later on when there are various actors with different incentives if the conditions are not clearly defined. But maybe its enough to clearly define them for start just in verbose.
Fully agree, the idea was to have just separated solvers, who are running their own infrastructure from strategy creation and implementation.
The solvers would be fully independent and responsible for correctly running the best strategy against the current batch. They still would get penalised for downtime / misbeviour. the only and main difference would be that they would not be responsible for the strategy logic. currently they are for both. logic would be responsibility of governance as I outlined above. this would have various nice properties.
Do you see probable attack vector / problem with this?
The question / problem may be how to ensure that the solvers are in fact running the vetted strategies by governance.
This is basically the same issue that L2s are solving, ensuring that execution was valid. only in this case it is execution of valid strategies. There could be fraud proof game, or even better zksnark provided by the solver with every executed batch.
Specifically, instead of maximising the sum of user surplus as a primary objective criterion we could maximise the product of surpluses. The philosophical difference between maximising the sum and product of surplus/utility is well summarised in the Introduction:
Nevertheless, the idea of maximizing the Nash social welfare (that is, the product of utilities) seems natural in and of itself [19, 37]. Informally, it hits a sweet spot between Bentham’s utilitarian notion of social welfare—maximize the sum of utilities—and the egalitarian notion of Rawls—maximize the minimum utility.
Maximising the product of surplus (nash social welfare) may also give us other fairness (like envy-freeness) and efficiency guarantees (Competitive Equilibrium) - or at least approximations of such - for free:
The CEEI allocations (which are what we care about, as the prices are virtual) exactly coincide with allocations x that maximize the Nash social welfare i vi (xi ) [5, Volume 2, Chapter 14]. Consequently, a CEEI allocation can be computed in polynomial time via the convex program of Eisenberg and Gale [22].
