Censorship Resistance and Slashing

In the event of Arx node misbehavior, all of a Node's delegations (including the operator's self-delegation) are slashed by the same rate as punishment. The slash acts as a strong disincentive for misbehavior, and also provides a mechanism to pay for the costs associated with a failed Computation (to compensate the other Nodes that did not misbehave for their work).

Types

There are two types of punishments for Arx nodes in the Arcium Network, non-participation (e.g. downtime or intentional inaction) and cheating (providing incorrect computation results). Both types are punished via slashing, however, non-participation receives a heavier punishment because the cost to detect non-participation is greater than that of cheating. Cheating is detected cryptographically, however, non-participation detection often comes with the increased cost associated with requiring a broader set of Arx nodes (not just those in the Computation's Cluster) to run a consensus protocol that verifies the withholding claim (see the Detection and Disputes sub-section below).

Slashing penalties are rate-based, meaning that the amount slashed is always proportional to the size of the Computation that was interrupted. As such, larger slashing penalties are applied to misbehaving Nodes with larger hardware specifications and therefore larger total stake delegated to them (since more stake is required to enable the hardware); thus, the increased stake functions as proportional collateral for increased slashing penalties (when needed).

Detection and Disputes

Since non-participation cannot be detected cryptographically (unlike cheating), Arcium Network's non-participation dispute resolution mechanism detects non-participation via a broader consensus of Arx nodes. The protocol allows any participant to accuse another of non-participation, prompting the accused to prove their participation by sending a specific message to all peers within the Cluster, which then act as relays. If initial attempts fail, the accusation can be escalated to a broader set of Arx nodes outside the original Cluster, ensuring broader verification and maintaining transparency. This strategy not only eliminates the possibility of censorship or disruptions but also aligns node incentives with network health, promoting a stable and reliable computational environment.

Cheater detection

Another form of misbehavior for participants is sending incorrect data to each other, which will cause the receiving peer to detect this and trigger an abort of the protocol. In classic BDOZ, it is impossible to tell apart a legitimate abort from a malicious one, opening up the possibility of nodes using this mechanism to selectively censor computations with virtually no repercussions. Thankfully, protocols like these can be augmented such that attempted cheating can be cryptographically demonstrated to other peers as well as to a third party: coupled with a slashing mechanism, as well as the knowledge that another cluster will pick up the computation subsequently, this all but removes any incentive to attempt censorship of a computation.