Embedded systems may easily die silently - error-nous behavior is detected only when the services it should provide are requested and the system does not respond - but how do you figure out what happened. Many common failure scenarios are detectable if not only the status of a system is evaluated but tendency analysis is taken into account. In machin tool industry the problem of tool-wear has been successfully tackled by logging of tool related force and torque data and monitoring the tendency of these values - thus giving an early warning when tools need to be replaced or adjusted. If this strategy is to be applied to embedded systems then the amount of data that a system needs to provide goes well beyond simply status values - embedded GNU/Linux allows to monitor systems at runtime down to the kernel internals and provides a multi-level logging facility that is the basis of any tendency-analysis system. Making this data available to off-line systems is trivial and possible with low resource requirements. Off-site logging allows to perform tendency analysis not only over long terms but allows to detect correlations between events on different devices and frees such analysis of the resource constraints that apply to the embedded system itself. The potentials for early fault detection and maintenance response has, I belive, not been appropriately considered by embedded OS/RTOS developers.