Zeeka saysNarwhal disseminates transactions as a DAG of certified batches; Bullshark orders the DAG deterministically — decoupling data spread from consensus for huge throughput.
Traditional consensus puts a single leader at the center of every block, and bandwidth bottlenecks on that leader. Narwhal rejects that design: every validator independently certifies batches of transactions and gossips them, forming a DAG of certificates across rounds. Bullshark is the tiny consensus protocol that runs over the DAG just to decide a total order of certificates — not to disseminate them.
Decoupling data dissemination from consensus means throughput scales with the entire network, not one leader. The architecture powers Sui, Aleo, and other high-throughput chains. The demo builds a small certificate DAG and produces a deterministic linearization.
Power-ups you unlock
Narwhal: every validator certifies and gossips batches in parallel
Forms a DAG of certificates across rounds
Bullshark: small consensus protocol over the DAG order
Throughput scales with the whole network, not one leader
Powers Sui, Aleo, and other high-throughput chains
The Collision attacks — common mistakes
Thinking the DAG itself is the consensus — it is the data layer
Forgetting Bullshark provides the total order
Assuming DAG mempool removes all bottlenecks (it shifts them)
Underestimating the bandwidth required for parallel dissemination
Boss battleBuild a small DAG of certificates and produce a deterministic linearization.