Spectrum sensing schemes focus on detecting active transmitters, but not incumbent passive receivers, which have nevertheless to be protected from excessive interference. The present paper advocates the notion of receiver map as a tool for unveiling areas where licensed receivers are likely to reside, and for limiting the interference accordingly. Receiver maps are tracked using a Bayesian approach, based on a one-bit message - here referred to as 'interference tweet' - broadcasted by the primary receiver whenever a communication disruption occurs due to interference. The resultant maps are utilized in a cross-layer resource allocation scheme, which is designed to optimize the performance of an underlay multi-hop cognitive radio network under long-term probability of interference constraints. Although non-convex, the problem exhibits zero duality gap, and it is optimally solved using a Lagrangian dual approach.