Didn't see any references to semantic web, ontology, linked data standards. They would have been a great fit, for instance, for time-series datasets there is W3C recommendation: https://www.w3.org/TR/tabular-metadata/. If you're an owner of the dataset, and you need to expose it to GOODS, just author a my-dataset.jsonld and feed into a centralized dataset search service.

They spend a lot of time talking about how dataset owners don't have to do anything to expose it to GOODS. It sounds like a big part of the work that goes into GOODS is automatic metadata inference.