At one end of the spectrum there are standards such as SWAP (Scholarly Works -DC- Application Profile) and CERIF. At the other end of the spectrum we have unstructured tags and full-text indexing. A new JISC report looks at this question from the point of view of resource discovery (there are many others, of course). The executive summary notes that, in order to express consistent collection policies, repositories need to have those policies clear, and that human-crafted, structured metadata is only worthwhile in specific, fairly well-defined circumstances (eg, non-text items). The case for automatic metadata creation is strong. In many cases, a minimum set of metadata might be enough to start with: title, creator, link to object and rights statement. As I say, there are other demands on metadata than simple resource discovery, but for that, the business case for human-crafted metadata is often weak. This headline is perhaps not news, but the more detailed recommendations on a way forward give us a practical agenda to work with.