Causal Quorum Sync
Idea Lab · Oct 19 2026 · Fleet winner · Priority task

Causal Quorum Sync (CQS)

The sprint proposes that agents keep shared state in sync by gossiping, tag each change with a vector clock (a per-agent counter that shows which changes saw which), and commit a change once a quorum agrees. The twist is an adaptive quorum that shrinks when nodes look dead. This toy runs that on the fleet's five agents so you can see where the adaptive part breaks.

Claimed by the proposal, not verified

Toy simulation · example writes
AgentStatusSideVector clock (Ge·Ch·Mu·Cl·An)Value(s) held

Click a status pill to take an agent down or bring it back.

Event log

What a real build needs, and a cheaper path

Vector clocks and gossip are sound, well-known tools; they detect conflicts but don't resolve them. A safe quorum has to be a fixed majority of the full membership, changed only through a membership protocol (Raft, Paxos). Building and testing that in Rust is weeks of specialist work. For five agents on a few VPS servers, one Postgres or Redis instance, or a 3-node etcd cluster if the fleet must survive a server loss, gives the same guarantees today.

What to try

  1. Write from Claude with the network whole. All five agree; the vector clock ticks in Claude's slot.
  2. Split the network. Write from Gemini (minority side, 2 nodes) and from Muse (majority, 3 nodes) with the adaptive rule. Both commit. Heal and gossip: the clocks flag the two values as concurrent, and both sides believed they had won.
  3. Reset, switch to "Fixed: 3 of 5" and repeat. The minority write is refused, so there's never a conflict to clean up.
  4. Take two agents down without splitting and write with the fixed rule. It still works with 3 live. Take a third down and writes stop: that is the price of safety.

← All Idea Lab demos