FAIR is a set of four foundational principles created by a network of researchers and stakeholders to guide researchers on the path toward Open Science.
A dataset that conforms to FAIR will be:
Findable - Data should be findable by both humans and computers. The dataset should have a unique, persistent identifier such as a DOI. The dataset should contain strong metadata that makes working with the dataset easier. The metadata should be registered within a searchable database or other resource..
Accessible - The data is able to be retrieved via its identifier (but authorization procedures can be in place as needed). Protocols should be open and able to be implemented by others. The metadata should remain available even when the actual data is not.
Interoperable - Metadata should be written in a common, accessible language with a broad range of applications, for example .txt, .json, .nii, .csv. Propriatory file types should be avoided when possible
Reusable - Metadata should be well-described and the data usage license should be clearly stated. The metadata should conform to standards applied in that modality/relevant community of researchers.
While it may seem that FAIR principles primarily apply to the end product, it is important to keep them in mind and comply to them throughout the project in order to save time later. It is easier to keep track of metadata along the way than it is to go back and remember what was done to the data. A dataset doesn't need to be open to be FAIR. Access to the data can be restricted as needed (Mons et al. 2017) in line with the "As open as possible, as closed as necessary" principle.