The definition of the concept of "learning object" is still subject to considerable debate and controversy, causing some degree of confusion and ambiguity in research about the topic. Proposed definitions differ in terms of broadness of scope, focus on reusability and requirements imposed on metadata, among other characteristics. Nonetheless, if we approach the definition from the narrower perspective of engineering software systems that deal with learning objects, they can be characterized by the scenarios they are capable of engaging in or by the services for which they provide support. This connects the concept of metadata quality with a situation-specific definition of learning objects. Consequently, a resource may be considered - from the automation perspective - as a valid learning object in a given scenario, while in others it should be discarded due to the lack of certain characteristics or metadata elements. Since automated processes and Web services can be unambiguously specified, this definition enables the specification of detailed conformance profiles with precise and specific requirements on learning objects. While this kind of normative definition does not eliminate the controversial issues about the best design practices or evaluation criteria for learning objects in a broad sense, it removes ambiguity in technical communication and provides detailed guidance for implementers, evaluators and users of learning technology systems.