Skip to content

Architecture

How Nexwest separates consensus, execution, and settlement while keeping one-block finality and a single account state.

Updated Nov 4, 2026Builders2 min read
On this page

Nexwest is a purpose-built layer one for trading. Consensus, execution, and networking are designed together so that a signed order can be matched, risk-checked, and committed in one block.

The two layers

LayerResponsibilityInterface
ConsensusOrders transactions, produces blocks, finalizes stateNexBFT
ExecutionMatches orders, updates margin, settles funding and liquidationsDeterministic state transition

The execution layer is split into two namespaces that share the same state root:

  • Core — perpetuals, spot, margin, funding, liquidations, vaults, and staking.
  • EVM — a general-purpose contract environment that can read Core state through precompiles and write actions through a system contract.

One chain, two execution surfaces

The EVM is not a separate chain. It is secured by the same validator set and commits in the same block as Core. An EVM transaction and a Core order can be ordered deterministically relative to each other.

Consensus

NexBFT is a pipelined BFT protocol derived from HotStuff. Validators are weighted by staked NXW. A block is final when a quorum of more than two-thirds of stake signs it; there is no probabilistic confirmation window and no reorg.

  • Block time: 0.2 seconds median, 0.9 seconds at the 99th percentile.
  • Finality: one block.
  • Validator set: fixed per epoch; epochs rotate every 100,000 rounds.
  • Slashing: available for double-signing and invalid state transitions; not automatic for downtime.

Execution

Every block applies a deterministic transition function:

  1. Apply consensus-ordered actions.
  2. Match orders in price-time priority.
  3. Update positions, margin, and funding.
  4. Process liquidations and vault transfers.
  5. Commit the new state root.

The engine is single-threaded per market but markets execute in parallel where they do not touch the same account. This keeps ordering simple while allowing throughput to scale with the number of independent markets.

Latency

End-to-end latency is measured from the moment a signed order is sent to the moment the client receives the committed response.

PathMedianp99
Co-located client → committed fill0.2s0.9s
Public internet → committed fill0.4s1.4s
REST read → response80ms250ms
WebSocket update → client40ms120ms

Throughput

Mainnet currently sustains about 200,000 orders per second. The bottleneck is execution, not consensus; the networking stack is designed to scale further as execution is optimized.

State model

The state is an authenticated Merkle tree. Each account has:

  • a nonce,
  • a USDC balance,
  • an optional spot balance map,
  • open orders,
  • cross and isolated positions,
  • vault and staking entries.

The API exposes the same objects. There is no separate indexer required to answer “what is my position?”; the answer is a state proof away.

Was this page helpful?