CSIP21 dependent on whether files exist or not?

[PhillipAasvangTommerholt]

When creating test corpuses we have created two packages:

One with files in the metadata/descriptive-folder
One without files in the metadata/descriptive-folder.
Both packages have no mdRefs in dmdSec in METS.xml.

I have trouble in understanding the CSIP21 requirement. Is it a SHOULD-requirement because there should be descriptive metadata in the package in general or is it a SHOULD even in both scenarios where files exist or not?

I would intuitively say that a mdRef-pointer MUST exist if there are files in the metadata/descriptive-folder and a SHOULD if there are no files.

[prettybits]

I would intuitively say that a mdRef-pointer MUST exist if there are files in the metadata/descriptive-folder and a SHOULD if there are no files.

I share this intuition, which the description of the test case also expresses (EDIT: I see that description has originally been added by you as well, so may not be as helpful a pointer. Still, the interpretation makes sense to me at least).

As to the general point why it is SHOULD I think this together with CSIP17 follows from CSIPSTR7. CSIP17 is SHOULD because it is allowed for there to be no descriptive metadata ("If descriptive metadata are available") and CSIP21 is SHOULD because CSIPSTR7 itself is, allowing in principle for metadata to not be placed in the strongly recommended place in the folder structure (or use an mdWrap element inline instead?).

[karinbredenberg]

Yes, if there are descriptive metadata the first rule to follow is CSIP17. Then its one dmdSec per descriptive metadata so therefore CSIP21 have 0..1 and a should.

[prettybits]

Then its one dmdSec per descriptive metadata so therefore CSIP21 have 0..1 and a should.

I'm not following, is my interpretation above that CSIP21 is SHOULD because, while strongly recommended, using a mdWrap instead of a mdRef shouldn't be an outright error correct?