Automatically determining the situation of an ad-hoc group of people and devices within a smart environment is a significant challenge in pervasive computing systems. Current approaches often rely on an environment expert to correlate the situations that occur with the available sensor data, while other machine learning based approaches require long training periods before the system can be used. This paper presents a novel approach to situation determination that attempts to overcome these issues by providing a reusable library of general situation specifications that can be easily extended to create new specific situations, and immediately deployed without the need of an environment expert. The architecture of an accompanying situation determination infrastructure is provided, which autonomously optimises and repairs itself in reaction to changes or failures in the environment.