1
General Discussion / Re: Using Casper Protocol for Secure Price Feeds
« on: February 18, 2016, 01:16:21 pm »
There are two key differences between Casper's schelling game and a schelling game for something like a price feed that make the price feed case more difficult:
1. In a consensus schelling game, all that really matters is that you come to consensus on **something**. If someone pulls off a P+epsilon attack, and makes it confirm 0 instead of 1, that's fine; as long as it doesn't flip-flip forever and as long as it converges to 1 at least some of the time. In the price feed case, converging on the wrong price even once is BAD.
2. A price feed has a few failure modes that don't exist in the consensus case. Particularly, (i) there is a centralization incentive for the feed to converge on the feed of one single exchange, and (ii) if that exchange pulls a MtGox then it's hard to tell if the feed will actually be able to coordinate on decoupling itself from the exchange in time.
That said, I think that it should be possible to overcome both of these issues, for (2) there are a few multi-level antifragility tricks that I think could do the job; essentially you would have a second-level schelling game that only gets called very rarely, the theory being that it's called too rarely for it to be worth automating, and that game, if called, is what determines the rewards for the first round; this incentivizes the first round to do not what people think the first round will do, but rather what people think the second round will do, at last in extreme cases.
1. In a consensus schelling game, all that really matters is that you come to consensus on **something**. If someone pulls off a P+epsilon attack, and makes it confirm 0 instead of 1, that's fine; as long as it doesn't flip-flip forever and as long as it converges to 1 at least some of the time. In the price feed case, converging on the wrong price even once is BAD.
2. A price feed has a few failure modes that don't exist in the consensus case. Particularly, (i) there is a centralization incentive for the feed to converge on the feed of one single exchange, and (ii) if that exchange pulls a MtGox then it's hard to tell if the feed will actually be able to coordinate on decoupling itself from the exchange in time.
That said, I think that it should be possible to overcome both of these issues, for (2) there are a few multi-level antifragility tricks that I think could do the job; essentially you would have a second-level schelling game that only gets called very rarely, the theory being that it's called too rarely for it to be worth automating, and that game, if called, is what determines the rewards for the first round; this incentivizes the first round to do not what people think the first round will do, but rather what people think the second round will do, at last in extreme cases.