Hi iain,
We are currently looking at this issue in relation to Primo.

For White Rose Research Online, I identified a few different scenarios that I think we're trying to cover:
1) (easy :o) Open Access copy in repository

2) Link to open access copy held elsewhere

3) Closed access, but with 'Request a Copy' button in repo

4) Metadata only record - no knowledge of an OA copy anywhere.

I think we're looking to only exclude the last of these from the OAI-PMH harvest set that Primo uses (got a meeting about that this Friday - will update you on any decisions).

One of the realted issues is that the harvested resource often links directly to the document URL.

Someone following the link will get a login screen, rather than any information about the item/availability.

I have a possible solution to this as well - which would help with (2) and (3) above if you still wanted them to be harvested.

Does that sound similar to your scenario?
If you don't use Request-a-Copy, or store external OA locations, I can suggest a 'local fulltext only' filter for an OAI-PMH set.

Cheers,

John