> Several data providers respond to this service agreement with a bid in the form of a data reply — when enough data providers have responded, the majority response is taken (or average depending on the request), outliers are removed, and data is fed into the contract.
What's to stop me from setting up 10,000 different data providers that initially provide good data to get a good reputation score, but then slowly corrupt them over time? It doesn't matter how many data providers you average if I can set up millions of them in seconds. I don't see any way to solve Sybil attacks here.
Bingo. Most cryptocurrencies/contracts/anything in the field/realm -- they don't attack the problem of "person" vs address/wallet/account. ChainLink might think they are clever but like you said..when accounts in your network are free, then don't expect any kind of consensus to work. Accounts or rather, more abstractly, entry into your network -- needs to cost something that can't be easily done to gain majority. Another way to attack the issue is to do antes..so accounts dont cost anything unless the account holder is caught doing something bad -- ie. every entry requires a refundable collateral.
One thing that these networks don't seem to protect well against are fake peers that request data but dont provide it. These currencies and coins have been lucky in that regard and have prevented some of it by seeding their own trusted peers as the initial peer neighborhood. But is this true decentralization? Seems a bit obtuse. To add to the baffle, IPFS started using bitcoin and ethereum for storing the initial peer data for new clients to connect to. Its a web... maybe thats ok. Are things allowed to be this tangled? ;-)
In 2014, the IPFS protocol took advantage of the Bitcoin blockchain protocol and network infrastructure in order to store unalterable data, remove duplicated files across the network, and obtain address information for accessing storage nodes to search for files in the network.
There's a penalty payment that each node puts up into escrow for each assignment of data, and if the data is not accurate in relation to all the other providers of that same data, the node will lose the payment.
There will also likely be a small amount fo LINK required to start a node with enough reputation to gain assignments which would also increase the cost of a Sybil attack.
> There's a penalty payment that each node puts up into escrow for each assignment of data, and if the data is not accurate in relation to all the other providers of that same data, the node will lose the payment.
So, a prisoner's dilemma situation here? If one person objects, everyone loses their money? Who gets the payment? Are the coins permanently burned? If so, seems harsh in the face of accidents. If not, seems open to abuse if someone could be both the smart contract creator and a data provider. I create 100 data providers, and a smart contract, and when I detect someone new has joined my pool, cause them to lose their coins which are sent to me.
> There will also likely be a small amount of LINK required to start a node with enough reputation to gain assignments which would also increase the cost of a Sybil attack.
Ah, so an economic majority that successfully scams others and acquires a mass of tokens can use them to launch more data providers.
Mmm, what would stop me from just grabbing the data the other oracles made available, and pretending I did the actual calculation/check/api-call/whatever?
> Several data providers respond to this service agreement with a bid in the form of a data reply — when enough data providers have responded, the majority response is taken (or average depending on the request), outliers are removed, and data is fed into the contract.
What's to stop me from setting up 10,000 different data providers that initially provide good data to get a good reputation score, but then slowly corrupt them over time? It doesn't matter how many data providers you average if I can set up millions of them in seconds. I don't see any way to solve Sybil attacks here.