We present a theory of fast and perfect detector components that extends the theory of detectors and correctors of Arora and Kulkarni, and based on which, we develop an algorithm that automatically transforms a fault-intolerant program into a fail-safe fault-tolerant program. Apart from presenting novel insights into the working principles of detectors, the theory also allows the definition of a detection latency efficiency metric for a fail-safe fault-tolerant program. We prove that in contrast to an earlier algorithm by Kulkarni and Arora, our algorithm produces fail-safe fault-tolerant programs with optimal detection latency. The application area of our results is in the domain of distributed embedded applications.