Embedded control systems consist of multiple components with different criticality levels interacting with each other. For example, in a passenger jet, the navigation system interacts with the passenger entertainment system in providing passengers the distance-to-destination information. It is imperative that failures in the non-critical subsystem should not compromise critical functionality. This architectural principle for robustness can, however, be easily compromised by implementation-level errors. We describe Safe- Flow, which statically analyzes core components in the system to ensure that they use non-core values communicated through shared memory only if they are run-time monitored for safety or recoverability. Using simple, local annotations and semantic restrictions on shared memory usage in the core component, SafeFlow precisely identifies accesses to unmonitored non-core values. With a few false positives, it identifies erroneous dependencies of critical data on noncore values that can arise due to programming errors, inadvertent accesses, or wrong assumptions regarding the absence of difficult-to-detect implementation errors such as data races and synchronization. We demonstrate the utility of SafeFlow by applying it to discover critical value flow dependencies in three prototype systems.