Learning Object repositories play a key role in the vision of reusable learning contents and learning designs, serving as providers for learning-oriented artefacts. Nevertheless, current metadata creation practices result in artefact collections that lack machine-understandable metadata, which seriously hampers opportunities for reuse. Semantic Web ontologies can be used to improve the quality of learning object metadata records, but they are not enough by themselves. In order to respond to requests by returning the adequate resources, the repository is required to be aware of the amount, type and quality of the metadata records it stores. In this paper, the design of a learning object repository approach to "semantic lifecycle" is described and illustrated through the concrete architecture of the prototype of the SLOR repository.