Cadence-mini

Mini implementation of Cadence (arxiv.org/abs/2607.02275), the multi-proposer BFT consensus protocol Category Labs published for Monad. It runs the protocol’s three core ideas end to end in a deterministic Rust simulator: extreme pipelining (every block finalized in its own independent consensus instance), multiple concurrent proposers per slot with censorship resistance and hiding, and 2-round speculative finality revertible only on a provable equivocation. 24 property tests verify the paper’s claims , safety, the optimal 3-round fast path, resilience at the f-bound, pipelining independence, speculative soundness. The tests themselves are tested: a mutation script breaks the protocol four different ways and proves a named test catches each. Same seed, byte-identical trace, every run. The latency experiment (200 simulated validators, geo-plausible delays) lands at a 0.71 speculative/final ratio vs the paper’s ~0.76 — REPORT.md includes the full round decomposition and the measurement bug I diagnosed to get there. NOTES.md logs every ambiguity in the paper and the interpretation I chose. Mock crypto, no networking , it’s a reference implementation for understanding the protocol, not a node.