[PROPOSAL][ABANDONED] Cosmos Validator Diversification Reward
This proposal draft addresses a potential solution to the problem of Validator centralization. Any feedbacks/suggestions/discussions are welcomed. Premise • Re-code the rewards algorithm so that a Gini coefficient is calculated to see if each accounts’ staked Atoms are suitably diversified across [TBD] number of separate Validators. If the account meets, or is less than, the agreed upon threshold Gini coefficient then the account receives XY% “extra diversification rewards” in addition to the normal rate of staking rewards. • Decentralization of Validators is a good thing for Cosmos’ current and future security. • Decentralization is also a good thing for the greatest amount of Validators, both present and future. Problem • As evidenced by Chorus One’s recent data pull ( Chorus One – Medium ) there is very little Validator diversification by most accounts. • So far there has been little to no economic reason for Delegators to diversify across multiple Validators. • With most Validators having experienced very good uptime so far and the penalty for lack of uptime and liveness being relatively minimal there is little economic reason to diversify. •…
Excerpt (1185 of 2553 characters). Read the whole post on the forum ↗
I’ve found using the Gini coefficient for measuring diversification to be somewhat unsuitable. Example: Delegator A has 2 delegations split 50:50 across 2 validators, his Gini coefficient is 0 (perfect equality). Delegator B has 5 delegations split 40:15:15:15:15, his Gini coefficient is 0.2, but I would argue Delegator B is better diversified and more suitable to earn rewards. Also having a very small delegation to one validator can totally skew your Gini coefficient. I’m still trying to figure out if there’s a better way to measure this.
I did not know about the Gini coefficient before reading this proposal … Anyway as found out by @FelixLts it is not suitable as it is. Simple solution that comes to my mind : ponderate this Gini coefficient with a coefficient calculating the ratio between the number of delegations and the total number of validators : ([current total number of validators]-[number of delegations])/[current total number of validators]) This coef would be between 0 (delegations split across the total number of validators) and approach 1 if there is only one delegation Regarding these other potential problems in the draft proposal : Large Validators might divide their operations down across multiple smaller Validator positions in order to allow Delegators to achieve artificial diversification by choosing multiple Validators who are in fact run and managed by the same Validator. This is not a problem specific to this proposal. According to me there should be off chain written rules that forbid the same entity to run multiple validators without making it clear they are actually controlled by the same entity. If such a behavior is discovered there should be a community vote to punish/remove…
Excerpt (1198 of 1724 characters). Read the whole post on the forum ↗
Good points on Gini.
If it was stipulated that accounts would need to have some minimum number of total Validators, like 8, would that mostly fix these issues?
I think @ddrdrck’s idea of somehow incorporating the # of validators and creating a new coefficient is intriguing. Setting the total number high could be a simple way to achieve the goal too.
Also considering the sybil attack risk, maybe self-bonded Atoms should receive diversification rewards too. Feels like otherwise this would disincentivize validator skin in the game.
Could diversification rewards incentivize custody-based staking services?
Yes, per @ddrdrck, incorporating a formula that looks at number of Validators and staked Validators in the account makes sense to use.
Now per self-bonded Atoms, they seem to be “damned if they do and dammed if they don’t” in that it’s definitely a marketing strategy for Validators to claim a high self-delegation but by self-delegating they’re missing out on this diversification reward.
I don’t think there’s a good answer here beyond the fact that it comes down to Validators either getting to market their safety by putting their own money ‘at-risk’ in their own Validator in order to prove reliability or they put their money ‘at-risk’ with many other Validators (“the network”) in order to gain a better return.
Frankly I’d completely understand and would actually prefer a Validator who delegated to the network to get the diversification reward rather than the Validator who doesn’t trust other Validators or is self-bonding for the marketing impact of it.
Potential negatives: Sybil attack: As mentioned earlier in the thread, it would be rational for large validators to split and run a number of smaller validators. This can only be stopped with off-chain rules and quasi-judicial processes. Formalizing rules, and attempting to use the governance system as a quasi-judicial system to enforce these rules will have wide ranging consequences. It’s one thing to use governance to change protocol parameters, it is quite another thing to use governance to make subjective judgments about individuals or entities. Efforts to police Sybil attacks would be extremely controversial, error prone and in the end would be unlikely to be successful. Incentivize custodial cartel formation: It would be rational for large token holders to form cartels. These token holders could be exchanges who custody tokens, or entities who hold tokens on their own balance sheets. A group of at least the minimum number of validators required to receive the bonus rewards would form, and replace a large self delegation with mutual delegations. There would be no net decentralization, yet these large holders would earn the bonus reward. Exchanges who already have a large…
Excerpt (1197 of 2119 characters). Read the whole post on the forum ↗
I’ve been procrastinating on replying because of this, but first let’s talk about the 600-lb gorilla here (and this goes for all top Validators who oppose this idea) namely the potential inherent bias here with regards to you currently being a top 10 Validator right now. More than likely I’m sure you realize that the net impact of sudden and mass diversification of the network could possibly mean far less bonded Atoms delegated your way as it gets spread out further downwards amongst more of the network. mattharrop: Efforts to police Sybil attacks would be extremely controversial, error prone and in the end would be unlikely to be successful. You’re not actually providing any supporting evidence here… And most of this is theory and assumptions which don’t hold their weight here. I mean the Game of Stakes example alone of nefarious Sybil players successfully being caught by 2 separate Validator in the community and then eventually getting forked out shows that the decentralized community can catch, come to agreement upon and then fork nefarious Validators. Also the recent failed network upgrade has shown that the community can remain decentralized yet come…
Excerpt (1197 of 2806 characters). Read the whole post on the forum ↗
Hey Felix, if it helps as an alternative to Gini coefficient, I’ve been using the Lorenz curve since July (with April as the baseline): https://figment.network/resources/cosmos-december-2019-network-analysis/
If you scroll to the bottom of that article, there is a list of links that correspond to each month’s ‘network analysis’ article. @JesseLivermore I’m finding a fairly consistent trend of Cosmos staking becoming less centralized. Thoughts? eg. doubts about methodology; rate of decentralization
Dude the gini ratio derives from the Lorenz curve there… And it’s still showing as terribly unequal.
Sidenote: I’m going to be proposing this officially soon.
Let me know if I can help with the proposal.
Sure! Are you able to download all the staked Atom amounts by validator for me? Figured I’d create an up-to-date Lorenzo and gini coefficient for the proposal.
I’m doing some modeling based on various scenarios to determine how many Validators at a minimum should be required to stake to in order to have the best economies of scale for this to work, but not require like a crazy amount of work on the Delegators’ behalf to delegate to like 50+ Validators or something like that.
Thanks!
Hi Jesse, I think that this is the data you’re looking for. Let me know if you need help with the proposal itself eg. feedback re: structure and/or wording
FelixLts:Also considering the sybil attack risk, maybe self-bonded Atoms should receive diversification rewards too. Feels like otherwise this would disincentivize validator skin in the game.
Been thinking on this point more and modeling it… Game-theoretically-speaking these large self-bonded entities would be wise to either stake out to numerous other Validators OR breakdown/divide their self-stake and create numerous mini-active-Validators… otherwise they don’t get these extra rewards.
If the extra rewards staked Validator minimum is set at 25, you could see 25 mini Dokia’s and 25 mini Polychain’s… In the end you could have 77 different Validators who make up 125 active Validators. I know @mattharrop mentioned this as well.
Overall I’m not seeing much of a solution here beyond starting a community campaign to fork out Validators who engage in this kind of bad faith behavior.
Suggestions though?
Yeah this is why I think that the stick approach game theoretically works better here, especially with limited validator slots. An implementation of correlated slashing as proposed by Sunny is the only thing I can imagine that could successfully deter this kind of sybil attack by making it more risky to split your stake across validators with the same setup. Meanwhile, it could still be benefical for the whale to run uncorrelated setups and split his stake across them, which would improve network resiliency (the desired outcome).
Forking out via community campaign seems excessively hard to justify, as the self-staking holder in question probably wouldn’t name validators Whale-1, Whale-2, etc. but come up with some plausible entity names, e.g. Staking Unlimited and Validator Incorporated or whatever.
@JesseLivermore just want to be clear that I’m supporting proposal creation in general, not that I support the substance of this proposal.
Edit: This post removed.
seems that this link leads us to nowhere.
I’ve been modeling various scenarios based on the required diversification count of Validators and have come to the conclusion that anything less than 50 Validators still wouldn’t “move the needle” much with regards to Cosmos’ overall Gini (which is currently 0.6907). For example an “extra rewards” requirement of 50 different Validators each having an equal stake for an account to get extra rewards (which could theoretically lead to the top 50 Validators each having about 3.3m staked Atoms) would still lead to a kind of high Gini of 0.527. It’s okay but its impact is kind of disappointing. Additionally, Polychain’s “Roman” has mentioned to me that they likely would have no choice but to split up (“Sybil”) their self-bonded stake into numerous Active Validators run by themselves if these extra rewards existed. And he mentioned any major fund/corporation who’s staking who has a fiduciary duty to maximize returns has to try to get the extra rewards, likely via Sybil nodes controlled by them. That’s very disappointing. So because of these 2 factors I’m going to suggest in the proposal that whatever the Active Validator count is that 80% of that total is the required minimum…
Excerpt (1196 of 2516 characters). Read the whole post on the forum ↗
Thanks, fixed it. Weird app versus browser error I think. I’m reposting below…
Perhaps one of the more fascinating outcomes is the discovery of an exact market-based interest rate and the yield curve. Smaller stakeholders can gain extra influence and higher yields by participating in the long-term staking pools.