The discovery of a formal process model from event logs describing real process executions is a challenging problem that has been studied from several angles. Most of the contributions consider the extraction of a model as a semi-supervised problem where only positive information is available. In this paper we present a fresh look at process discovery where also negative information can be taken into account. This feature may be crucial for deriving process models which are not only simple, fitting and precise, but also good on generalizing the right behavior underlying an event log. The technique is based on numerical abstract domains and Satisfiability Modulo Theories (SMT), and can be combined with any process discovery technique. As an example, we show in detail how to supervise a recent technique that uses numerical abstract domains. Experiments performed in our prototype implementation show the effectiveness of the techniques and the ability to improve the results produced by selected discovery techniques.