Network operation
Market infrastructure must recover without guessing which balances or orders were correct before a failure. Novrinex operation begins from reproducible network definitions and recovery paths that return each service to verified state.
Reproducing a network
Section titled “Reproducing a network”Each environment pins its chain identity, protocol version, genesis state, validator set, peer topology, application release, service configuration, protocol parameters, and oracle definitions.
Release artifacts carry checksums and signatures. Genesis construction is deterministic, and private credentials remain outside versioned public configuration.
These inputs let an operator reproduce the intended network rather than reconstructing it from a running server after an incident.
Recovering nodes and data services
Section titled “Recovering nodes and data services”A validator or full node can restart from its local state, restore a verified snapshot, or join through state sync. Archive blocks provide the history needed for a complete replay.
Indexers rebuild from committed blocks and versioned events. RPC and WebSocket services can fail over to healthy instances. Oracle-publisher recovery focuses on restoring the required reporter quorum before affected markets accept new exposure.
A recovered service verifies its chain identity, height, and state commitment. It does not treat an indexer cache or user interface as the source of account truth.
Upgrades
Section titled “Upgrades”An upgrade identifies its activation height, exact release and checksum, protocol and schema versions, deterministic state migration, operator procedure, validation checks, and rollback boundary.
Operators rehearse the transition against a persistent network before activation. Every validator must produce the same migrated state from the same starting point.
Observing the transaction path
Section titled “Observing the transaction path”Monitoring follows an action from network receipt through mempool admission, block inclusion, finality, Core execution, indexer publication, and WebSocket delivery.
Order, cancellation, oracle, liquidation, and builder transactions are measured separately. Latency distributions include p50, p95, and p99 under sustained traffic and bursts, because an average can hide the delay experienced during a volatile market.
Failure exercises
Section titled “Failure exercises”Operators rehearse validator loss, unavailable voting power, slow peers, network partitions, RPC overload, oracle disagreement, corrupted indexers, snapshot recovery, failed upgrades, and concentrated liquidation traffic.
The network should fail according to known financial rules. When prices are stale or uncertain, new exposure pauses. During congestion, FairFlow retains capacity for cancellations and solvency actions. When a projection is corrupt, it rebuilds without changing the underlying ledger.
Public performance figures describe an identified network configuration and workload. A component benchmark does not stand in for the end-to-end experience of submitting, finalizing, and observing a trade.