As Karen has said, there are different structures for different situations.  One common situation is that a work lands on the cataloger's desk bearing series-like information, which the cataloger records in the catalog to help people find it if all they know is that it appeared in that series.

The cataloger does not however have enough information to be able to create a record for the series: when did it start, what does it contain, has it all been published by the same publisher etc., what is its title (frequently different from work to work), do some of the works belong to more than one series, did the series split into two, etc.? The cataloger often has no information about the series except that it contains the work that has just been cataloged.

In that situation the cataloger makes the series title (insofar as it is known) an access point in the record for each member of the series, but is not usually in a position to make a record for the series itself.  There is therefore no "host work".   The series-like information is included in the catalog as part of the description of the work cataloged.

Another situation is the item level collection, such as Jefferson's library.  Here we are talking about individual copies of books or other multiples.  If the architecture is to be kept clear, the data referring to the collection in this sense ought not to be in the bibliographic record, as the Item record is the place for copy-specific data.  Some catalogs do however allow a kludge by putting the data into the bib record, in local fields.  I don't expect this to be relevant to citation-bibliographies, except that it could muddy the waters.

On the relation between an archive and a collection, there are differing views ranging along a spectrum.  At one extreme there are those who consider an archive to be a record of transactions made by the compiler of the archive, in which case there are no "odds and ends": everything has its place in a hierarchy, and the next level up the hierarchy (field 773 in MARC terms) is the host work.  E.g. a letter dated 1915 may be part of "Correspondence 1914-1918", which may in turn be part of "Correspondence".  At the other extreme there are those who consider it to be virtually the same as an item-level collection, a more casual affair.  If the items are unique instead of multiples, as in the case of a collection of manuscripts with different provenances,  linking them to an artificial host work (the name of the collection) or linking them via a common access point (treating them like a series) is equally innocuous: a librarian might choose the same structure as would be used in the same catalog for multiples.  However, no archivist that I know would call that an archive.

William Schupbach
Wellcome Library, 210 Euston Road, London NW1 2BE, England
E-mail: [log in to unmask]
Catalog: http://catalogue.wellcome.ac.uk/search/X
[British registered charity no. 210183]