Domestic ubiquitous computing systems often rely on inferences about activities in the home, but the open-ended, dynamic and heterogeneous nature of the home poses serious problems for such systems. In this paper, we propose that by shifting the responsibility for interpretation from the system to the user, we can build systems that interact with people at humanly meaningful levels, preserve privacy, and encourage engagement with suggested topics. We describe a system that embodies this hypothesis, using sensors and inferencing software to assess 'domestic wellbeing' and presenting the results to inhabitants through an output chosen for its ambiguity. In a three-month field study of the system, customised for a particular volunteer household, users engaged extensively with the system, discussing and challenging its outputs and responding to the particular topics it raised.