Risk-Aware Runtime Monitoring and Mitigation: Semantics, Abstraction, and Monitor Memory (AE)

It includes an executable infusion-pump model supporting exact finite-horizon predictive risk analysis and a FastAPI-based prototype demonstrating history-dependent hazard monitoring, consequence-based risk computation, and policy-triggered mitigation. Experiments demonstrate different risks for histories reaching the same declared system state, validate incremental monitoring against full-prefix references, compare exact predictive monitor quotients, and measure runtime and memory costs.

Conservative bucket-monitoring experiments quantify extra hazard and mitigation verdicts, including their episode counts and durations, and check that exact violations are not missed. The package includes prebuilt Docker images for ARM64 and x86-64, source code, automated tests, reproduction scripts, reference results, and documentation for evaluation and reuse. With Docker and the documented prerequisites installed, the default evaluation runs offline after download.</p>