In a large scale distributed system with a large number of sensors interconnected through a wide-area network infrastructure, it is advantageous not to disseminate a query to all available sensors, but only to a subset of the most-relevant ones. We target an application scenario where mobile phones trace users' objects (which are equipped with small identification tags), distribute useful context information related to these objects, and are able to locate them when lost or misplaced. For locating a lost object, the proposed algorithm - parameterized with a data model of the application domain - is able to explore a wide range of heuristics based on history data present in the system (on objects, users, and their past location), similarly to the way a human user would re-iterate all that she/he knows about a lost object in order to locate it. As the proposed algorithm uses the data present in the system to parameterize its execution, it is generic enough to be applied to other application domains.