Scope & threat model

Velarium’s security posture is defined by three moving parts: Shipyard (compile/synthesis), Celerity (runtime/scheduler), and Vehicles (sandboxed native execution units). The trusted computing base (TCB) is the operator’s environment (host OS/kernel/application and system services/hardware) plus Velarium’s own runtime, compiler pipeline and system code within Vehicles. The adversary is the workload: contract bytecode and transaction inputs that may be malicious, and transaction patterns that attempt to stall execution or amplify resource usage.

Trust boundaries

Velarium draws a hard boundary between untrusted contract code and the trusted executor. Recompiled contracts execute only inside Vehicles. The host runtime (Celerity), the compilation pipeline (Shipyard), and the integration adapters are trusted by the deployment. In clustered deployments, the intent is to tolerate runtime faults (nodes disappearing, link loss) without producing incorrect outputs.

Compile-to-native and native-code-execution attack surface

Compiling adversarial bytecode into native code expands the attack surface. Malformed or adversarial contracts may attempt to trigger compiler/codegen bugs, access restricted memory or cause undefined or faulty behavior at runtime. Responsibilities for mitigating this is split between Shipyard and Celerity

Shipyard mitigations:

Celerity mitigations:

Adversarial contracts & DoS resistance

A contract can be “malicious” without exploiting memory safety: it can attempt to stall the pipeline, create hotspots, or amplify resource usage (CPU, memory, trace volume). In addition to techniques used to mitigate normal contention, Velarium has tools to detect and report on pathological contention allowing the system to take preventative actions: