Recent research has abstracted diagnosis away from the activity needed to acquire information and to act on diagnosed disorders. In some problem domains, however, such abstraction is counter-productive and does not reflect real-life practice, which integratesdiagnostic and therapeutic activity. Trauma management is a case in point. Here, we discuss a formalization of the integrated approach taken in TraumAID, a system we have developed to serve as an artificial aide to residents and physicians dealing with multiple trauma. Among other things, the active pursuit of information raises the question of what is and what is not worth pursuing. In TraumAID 2.0, we take the view that the process of diagnosis should continue only as long as it is likely to make a difference to future actions. That view is formalized in the goal-directed diagnostic paradigm (GDD). Unlike other diagnostic paradigms, goal-directed diagnosis is first and foremost concerned with setting goals based on its conclusions. It regards the traditional construction of an explanation for the faulty behavior as secondary. In order to explicitly represent goal-directedness, the diagnostic process is viewed as search in a space of attitude-beliefs. From this, we derive a high-level algorithm that produces appropriate requests for action while searching for an explanation. A complete explanation, however, is not the criterion for terminating action. Such a criterion, we argue, is better treated in terms of goal-means tradeoffs. TraumAID's architecture, in so far as it embodies this goal-directed approach, assigns to a complementary planner the resolution of such tradeoffs.