// the take
A Rust coordinator and miner-supervision stack for the quip-validator Substrate chain: it follows the chain head, stages Ising problems, dispatches them to miner subprocesses running simulated annealing, Gibbs sampling, or D-Wave quantum annealing, and submits proofs back on-chain. It's for people building or running mining infrastructure for this one specific chain, not a general-purpose solver or library.
The coordinator/miner split is clean — chain integration and session supervision live here, while each miner is a standalone binary talked to over a local socket, so a miner crash doesn't take down the coordinator. Devnet integration tests hit a real running node (snapshot read, proof submit, mempool decode) instead of mocking the chain client, and they're properly gated behind an ignored flag plus a QUIP_DEVNET env var so the default offline test run stays fast. The CUDA story is unusually well specified: documented compute-capability floor and ceiling, per-card JIT compilation, and an explicit fallback to the nearest lower architecture via PTX JIT for capabilities the shipped toolkit doesn't recognize.
This only has value if you're operating on the quip-validator chain — there's no standalone solver or library use case here. The README says 'experimental, no production warranties' and means it: core wire protocol and consensus primitives live in a separate quip-solver-core repo pulled from crates.io pinned at v0.0.0, which is not a version you build on. The actual miner binaries (CPU/CUDA/Metal/D-Wave) aren't in this repo at all — they're fetched from four other repositories via a shell script, so understanding or auditing the full mining path means tracking changes across five+ repos with independent release cadences. The D-Wave annealing path is only exercisable with access to real quantum annealing hardware, which puts a meaningful chunk of the stack outside what almost any adopter can actually test.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →