GPU capability decides which compute lane can serve a job.

Hardware determines the available lane

A node agent benchmarks the GPU and detects confidential-compute capability during onboarding. The specification separates sealed-lane eligible confidential hardware from consumer GPUs and other open-lane hardware. A scheduler routes each job according to both the workload and the node’s capabilities.

Match jobs to suitable nodes

Routing considers compute lane, VRAM fit, stake weight, latency and reputation. The sealed lane (in testing) is for private jobs on the attested stack. The open lane handles non-sensitive workloads such as public-model batch inference, embeddings, evaluations and image generation.

Keep earnings tied to buyer demand

The rate card is posted by GPU class, lane and work unit. Providers earn from verified jobs funded by buyer spend. Demand can change, and available hardware does not imply guaranteed utilization or income. The specification’s temporary availability floor is capped and decays to zero.

Follow the path through the dashboard

The dashboard design brings node state, lane, receipts, epochs and claim routes into one place. This preview illustrates that information architecture. Values and charts are labelled as examples rather than live network activity, current earnings or a service-level commitment.