Time synchronization is a crucial component of a large class of sensor network applications, traditionally implemented as a standalone middleware service that provides a virtual global time to the application layer. While this approach is simple and robust, it suffers from two important drawbacks: first, power aware operation is cumbersome to implement, and second, periodic resynchronizations induce an extensive messaging overhead. To resolve these problems, a new class of reactive time synchronization techniques have been proposed that establish a common reference time base post facto, i.e. after an event of interest had occurred. In this paper we present the formal error analysis of a representative reactive technique, the Routing Integrated Time Synchronization protocol (RITS). We show that in the general case the presence of clock skews cause RITS to scale poorly with the size of the network and identify a special class of sensor network applications that are resilient to the scala...