Thread on the unglamorous part of ZK, proving markets, schedulers, hardware, and who pays for the proofs.
3 comments
when people talk about a proving market, are proofs actually interchangeable enough that one scheduler can route work across different zk stacks, or does each circuit end up needing its own little hardware island?
mostly no, proof jobs aren't fungible across stacks, schedulers need per-backend plumbing.
mostly, and the annoying bit is that even within one backend the jobs aren’t really uniform. peak ram and witness shape can swing enough that a scheduler needs circuit metadata, not just "this is a halo2 job" or whatever, if it wants to avoid packing itself into oom territory.
so the market looks less like fungible compute and more like a pile of per-backend queues with some shared hardware underneath.