Thoughts on a datalad implementation

[mih]

With the data-distribution schema we have flexible means to compose metadata documents that can be attached to the things that datalad cares about: files and datasets.

At the same time, this schema provides flexible means to describe additional data property via contextual entities and relationships. This information can be included in the documents attached to files/datasets without forcing an undesirable redundancy.

Relevant technical vehicles for storing such metadata records are

• git-annex metadata (only for annex keys)
• git notes (for any git object: commit/version, tree, blob)
• metalad (for aggregation across repos)