G) Governance and Incentives
The RNDR Network functions using both automated and user generated verification processes to confirm work. Creators are encouraged to promptly approve jobs that are correctly processed through the system. Final renders are withheld until full resolution watermarked previews are confirmed. As a result, creators are incentivized to expediently confirm correctly processed frames. If a frame is not successfully processed, a creator has the ability to reject the frame, and the frames will automatically be assigned to another node on the network for re-rendering.‌
Reputation scores are essential for both creators and node operators because they are central to the resource allocation process on the network. For example, creators with higher reputation scores will have access to more concurrent nodes while rendering. Similarly, node operators with better reputation scores will get jobs assigned faster than mining nodes with lower reputation scores. For example, if the network is not at 100% utilization, nodes with the lowest reputation scores will be last in the queue, incentivizing node operators to build a positive history. Reputations scores also serve as proxies for detecting malicious behavior. For example, if a creator regularly rejects work from a number of nodes with high success rates, these excessive failures will lead to a job being terminated. In such a case, the network will quarantine the job and a creator's reputation score may be negatively impacted. Similarly, if a node repeatedly fails to adequately process frames that are subsequently successfully rendered by another mining node, its reputation score will be negatively impacted, and the node will receive a lower ranking.
In addition to review, the network uses progressive rendering, in which creators can watch the render progress in real time as more samples are processed and the image goes from noisy to noise free. By watching a frame progressively render, creators will have additional tools to detect processing abnormalities early on in the processing of a job. The Network also automatically checks jobs by comparing a node’s potential and a job’s parameters through the client .exe application. If there are large disparities between a node’s OctaneBench potential and a node’s output, the RNDR network can detect these anomalies and reroute work to properly functioning nodes. Octane’s AI denoising algorithms can also be run on a render to detect “cheats” when samples have not been fully processed and noise filtering has been applied. Finally, the network .exe continually interrogates a node, allowing it to detect changes like underclocking. In the case of underclocking or poor performance relative to a node benchmark, the node’s OctaneBench score is recalculated and adjusted to reflect the underclocked value.
Finally, RNDR deploys incentives designed to reduce the risks of reputation gaming, in which a user with a poor reputation or a malicious user just creates multiple accounts to avoid the consequences of bad behavior. Creator accounts are tied to their unique Octane Account, which includes software license entitlements. Without these software entitlements, users cannot create renders for use on the network. Additionally, creators with better reputations get access to more nodes, creating an incentive to build history. For node operators, the network’s reputation system is used for the creation of higher tier node pools, which provide greater returns per compute cycle, with sustained positive reputation history. Thus, it is disadvantageous to start over with a new account. To make this system work, the network corrects for system error and includes recency bias functions in its scoring algorithm that allow users to self-correct over time.
Thus, disputes are resolved through a combination of user generated review and through automated machine learning processes. Finally, RNDR Support provides users an outlet to handle disputes and correct for idiosyncrasies or systemic network errors.
Copy link