Consider a group of servers that agree to carry out an intensive computation task that spans several days. Furthermore, assume that the computation requires multiple rounds of interactions and the full participation of all participating servers, and otherwise fails. Here, we would like to guarantee that the servers exchange information as agreed upon without defaulting. In such a setting, it is critical to ensure that the computation is carried out as intended, and that no server invests computational effort only to learn that a different server abruptly decided to not continue the computation any more. Observe that the problem description as is does not involve money. Still our formulation of SCD allows us to capture such a setting and offers a meaningful solution to this prob- lem, namely that a defaulting server will be forced to pay a penalty to everyone else. Such a solution can be achieved by a straightfor-ward use of a verifiable computation scheme in combination with our see-saw transaction mechanism.
Next, consider a group of agents who participate in a set of fi-nancial transactions over the internet. For example, these could be agents in a prediction market (possibly with dark pool trading ca-pabilities) who place bets on the occurrence of sets of events, and may adaptively vary their choices depending on whether a previous event in the set happened or not. One must also consider what hap-pens when a malicious agent stops participating during the process. A naïve solution would require that the agents make a deposit at the beginning of the protocol which they would forfeit when they abort.To make this idea work in a decentralized setting, one must develop a method to put the deposits in escrow, and make sure that in the event of an abort (1) honest agents can always retrieve their de-posits from the escrow, and (2) honest agents obtain penalties from the escrow when a dishonest agent aborts. Implementing such a de-centralized escrow when a majority of agents are dishonest is not straightforward. Our formulation of SCD exactly allows the capa-bility to maintain a decentralized escrow across multiple stages of a computation and hence our protocol implementing SCD provides a solution to the prediction market problem described above