Framework for dataset validation

[mih]

This has general utility, but also specific applications.

• verifying the completeness of an individual "sample" record in a study (all desired data types, all required annotations)
• checking whether a desired number of item of some entity is present
• ...

One way to implement this would be in terms of metadata validation. A comprehensive report on a dataset is generated, and this report is than validated for compliance with some schema. Once schema compliance is established, relatively simple declarative conditions can be evaluated, while benefiting from a known metadata structure and vocabulary.

The metadata-approach would also nicely connect existing facilities (metadata extraction), and further communicate additional details on the expected content of a dataset in a structured, machine-actionable form.