The evolution of cognitive (secondary) networks to enable more efficient spectrum usage will rely on fast and accurate spectrum sensing/mapping, supported by a suitable architecture for data integration and model building. In the first part of the talk, fundamental aspects of the wide-area RF mapping problem as a grand challenge will be highlighted; and some recent work at UW that clarifies sub-system level trade-offs (between scan latency and channel status estimation accuracy, for example) will be described. Next, the evolution of a hybrid architecture – decentralized client-side sensing assisted database updating – is explored. Within this, model-based answers to fundamental questions such as “how much white space capacity is available” as a function of location for U.S. TV bands are developed. The talk will conclude with a description of current efforts for spectrum sharing (co-existence) just underway in the 3 GHz band (broadly) between different primaries (largely government operated communications such as military and non-military radars) and commercial networks (802.11 and 4G LTE).