What was done to stored data is recorded in the stores' log books, not in dataset attributes¶

Each store keeps a log book of what was written: the source files, their hashes, and an ordered record of the preprocessing steps (operation, settings, result) with the versions of the packages whose code ran. Stored data carry no provenance attributes, so processed and unprocessed data have the same shape. A write whose preprocessing settings differ from what the store already holds is refused, and VOD is computed only from canopy and reference data processed the same way.