The introduction of Data Services as first class citizens in DCAT 2.0 raised questions about the usage of Data Services and Distributions. This section provides a guideline for publishers what to consider as a Distribution and what as a Data Service.
A first distinction between distributions and data services is their dependency on a dataset for their existance. A distribution cannot exist without its dataset. It is a specific representation of a dataset (cfr definition W3C https://w3c.github.io/dxwg/dcat/#Class:Distribution). Whereas a data service is an entity in its own right. It provides access to datasets or it provides data processing functions. The independence also holds between the distributions of a dataset, and the data service which provides access to that dataset. The distributions are not required to be the result of the data service operations. However, they may.
Many of the properties of distributions are file oriented (downloadURL, format, byte size, checksum, modification date, …). The relevance of this information is reduced for data services, related information is present in a very different form and thus under different terminology. For instance, data services do and can provide format transformations, language transformations and schema transformations on request. Also the handling of trust is different. While tampering of downloadable content is detected by e.g. checksums, data services create often a trusted channel using security measures such as authentication and encryption. This reduces the need for additional trust checks on the data.
The difference between downloading a file or accessing the data through a service have resulted in the following guidelines:
Orthogonal to the nature clarification of distributions and data services, there might be need for a granularity clarification between datasets and distributions. Commonly, at first sight, it is expected that all distributions of a dataset are indentical in content, only differing in the representation of the data. But the when considering dataset series, this interpretation seems not valid anymore. In the upcomming release of DCAT 3.0, dataset series and dataset versioning are addressed. Implementors are advised to already take this proposal into account when creating guidelines for distributions. Note that this is less an issue between datasets and data services as both are independent entities. Data services usually address the granularity by providing the necessary query interface language so that the user can get the data according its needs.
These guidelines will be able to capture many access patterns, corresponding to most users’ expectations. However there might be cases that are more vague. In that case the DCAT(-AP) community can be questioned for a recommended approach.