Instead of relying on logs, replay, or historical coordination, nodes exchange full state and correctness emerges through selection and erasure. Multiple candidate states may exist briefly; one survives, the rest are discarded.
The demo runs five independent nodes over UDP. A node can be severed mid-run and later rejoin without replaying history. The repository includes a short screen capture showing the system running live.
This is not production-ready and intentionally surfaces failure modes like packet loss and jitter rather than hiding them behind retries or buffering.
I’m posting this as an exploration of a primitive, not a finished system. Happy to answer technical questions or hear critiques.
(For longer technical follow-ups, I’ve enabled GitHub Discussions in the repo.)