Erlang is a concurrent functional language based on the actor model of concurrency. In the purest form of this model, actors are realized by processes that do not share memory and communicate with each other exclusively via message passing. Erlang comes quite close to this model, as message passing is the primary form of interprocess communication and each process has its own memory area that is managed by the process itself. For this reason, Erlang is often referred to as implementing “shared nothing” concurrency. Although this is a convenient abstraction, in reality Erlang’s main implementation, the Erlang/OTP system, comes with a large number of built-in operations that access memory which is shared by processes. In this paper, we categorize these built-ins, and characterize the interferences between them that can result in observable differences of program behaviour when these built-ins are used in a concurrent setting. The paper is complemented by a publicly available suite of more than one hundred small Erlang programs that demonstrate the racing behaviour of these built-ins.