I think we'll actually want to check if we can do better than COinS (which is a very limited format). I suspect we can, e.g. using the provided MODS

Do you have some kind of reference on the hierarchy between formats? This would be useful
Indeed they seem to have XML endpoints (https://discover.libraryhub.jisc.ac.uk/support/api/) I suppose I can feed this to MODS.js in the same way as the SAGE Journals translator fed it to the RIS translator?

Do you have some kind of reference on the hierarchy between formats? This would be useful

Not a reference but broadly:

1. MARC/MARCXML and MODS
2. Rich custom formats (like PNX for Primo) but those are a lot of work
3. RIS / BibTeX (both have pros and cons, depends on implementation)
4. Embedded Metadata of high quality (e.g. complete highwire tags)
5. COinS
6. Embedded Metadata of low quality (e.g. basic DC)

Indeed they seem to have XML endpoints (https://discover.libraryhub.jisc.ac.uk/support/api/) I suppose I can feed this to MODS.js in the same way as the SAGE Journals translator fed it to the RIS translator?

Should work exactly the same, yes.

I tried to implement it using the MODS endpoints but this is really not satisfactory, because the type is rarely well detected. Eg this item which is obviously a feature film, is detected by MODS as a "video recording", COinS thinks everything is a book and RIS / Bibtex find the right "film" type, but I did not find any unambiguous element on the page to detect the right type (the schema information is "CreativeWork" and the Library Hub info is "visual", which is shared with a lot of categories).
It does work well for some items eg sound recordings or maps where the ambiguity is reduced. I'm not sure how to proceed forward