Evaluating Client-Side Proving Systems

Duration: NaN:NaN:NaN

Speaker: Alex Kuzmin

Type: Workshop

Expertise: No prior knowledge

Event: Devcon

Date: Nov 2026

Participants analyze a saved client-side proving benchmark dataset covering 16 systems and various workloads (SHA-256, Keccak, ECDSA, Poseidon). First, they fit proving-time models against reported constraints or VM cycles and test predictions for a larger workload. Second, they construct Pareto frontiers for proving time versus RAM and proof size, then apply resource budgets and ZK/PQ filters to select eligible configurations. Discussion addresses prediction error, model selection versus independent validation, workload and configuration differences, and what to measure before adopting a system. The aim is to make application-specific choices from published measurements. Exercises use Python, NumPy, and Matplotlib, with optional coding-agent assistance