Explanation

Why the format is shaped the way it is. Nothing here is needed to use the package — it is here because the decisions below are the ones most likely to be mistaken for arbitrary.

• The data model — a sample is a subject, not a scan, and most of the rest follows from that: grids, timepoints, coverage, integrity.
• What the converters refuse, and why — every import and export that fails on purpose, and the reasoning. A refusal you do not understand looks like a bug; a refusal you do understand is usually telling you something true about your data.
• Threads, processes and open files — one HDF5 lock per process, what runs outside it, fork and exec, writers, and when threads help. Read it before adding either.
• Design rationale — what 0.x could not express, the alternatives weighed for each load-bearing decision, the costs accepted, and the non-goals.

The design rationale's 1.1 section covers the clinical profile and the task and cache contract.

For the normative version, see the specification and 1.1.