Skip to main content

Architecture

Senticore separates fast execution from trustworthy settlement by running matching, state materialization, and custody on distinct layers.

Component map

Six integration surfaces

Most crypto exchanges expose two ways in: REST and WebSocket. Senticore exposes six logical integration surfaces, each routed through the same canonical sequencer.

SurfaceForWhy
Public Market Data HTTPApps, dashboards, data consumersCacheable, unauthenticated, separated from trading paths
Trading HTTPRetail traders, app backendsJSON over HTTPS, EIP-712 signed actions
Public WebSocketReal-time market dataSequence-numbered, gap-recoverable
Private WebSocketUser account streamsAuthenticated handshake, per-user channels
FIX GatewayInstitutional desks, OMS, quant fundsIndustry-standard protocol with session sequence recovery
Market Maker APILiquidity providers, quote enginesBulk operations, binary submission, cancel-on-disconnect, drop-copy stream, quote obligations

All surfaces converge into the same matching engine, risk checks, canonical state, and settlement path.

How this compares to other DEXs

SurfaceHyperliquiddYdX v4Senticore
Public Market Data HTTPYesYesYes
Trading HTTPYesYesYes
Public WebSocketYesYesYes
Private WebSocketYesYesYes
FIX GatewayNoNoYes
Binary Order EntryNoNoYes, through the Market Maker API
Dedicated MM bulk + Drop-Copy + CODNoNoYes

Comparison is based on publicly documented integration surfaces as of May 11, 2026. A venue may expose adjacent generic actions; the table marks "Yes" only when the surface is publicly documented as a dedicated integration surface. Sources: Hyperliquid API docs, Hyperliquid WebSocket docs, dYdX Integration Documentation, and dYdX WebSocket docs.

Sequencing layer

The sequencing layer validates signatures and credentials, accepts actions into the mempool, assigns deterministic order, and writes accepted events to a durable log before execution state is materialized.

The durable event log is the replay boundary. It supports recovery, state verification, and incident analysis without making API read models the source of truth.

Execution layer

The execution layer runs matching and risk.

ComponentResponsibility
Matching engineMaintains order books and generates fills
Risk engineApplies collateral, position, exposure, and product constraints
State materializerConverts accepted actions and fills into canonical account, order, balance, and settlement state

Persistence layer

Three database boundaries keep concerns separated:

BoundaryResponsibility
Canonical stateAuthoritative execution and ledger state, write-heavy, replicated
Read projectionsQuery-optimized views for public API, private API, streams, and the web application
Metrics and audit eventsOperational observability, business metrics, and incident analysis

This boundary keeps trading-state writes, user-facing reads, and operational control surfaces separated.

Settlement layer

Settlement is a separate layer, not a sequencer subroutine. It prepares checkpoint payloads, coordinates publisher quorum signatures, and commits roots on-chain.

ContractRole
StateCommitmentStores state and withdrawal roots signed by the publisher quorum
MultiCollateralVaultHolds collateral and executes proof-backed withdrawals
SPT TokenProposed token for incentives, governance, and fee alignment

Data flow: a typical trade

  1. A user or delegated credential signs a trading action.
  2. The client submits through HTTP, FIX, or the Market Maker API.
  3. The access layer validates credentials, action shape, and permissions.
  4. The action enters the mempool and receives deterministic ordering.
  5. The accepted event is recorded in the durable log.
  6. Matching and risk checks produce fills, rejects, or state transitions.
  7. Canonical state is materialized and read projections are updated.
  8. WebSocket streams publish market and account updates.
  9. Settlement roots are built and signed by the publisher quorum.
  10. Roots are committed on-chain for verification and withdrawal execution.

What we do not show

For clarity the diagram omits:

  • internal admin and operational interfaces,
  • per-process detail of sequencing primitives,
  • monitoring, alerting, and incident-response infrastructure,
  • staging and pre-production environments.

These are part of the operational surface and remain in internal operations documentation.

Continue with the trust model, threat model, and integration overview.