Implementation scenarios: where does the metadata come from?
OAI-PMH repository implementations consist of request- and error-handling scripts that draw on a relational database (RDBMS) for the metadata they deliver. In practice, this is because implementations that relied on real-time extraction of metadata from documents in a file tree would be unacceptably slow. So for practical purposes the question becomes one of how the metadata finds its way into the RDBMS, and that in turn depends on the nature of the documents themselves,
What's here[NB: this is a guided tour of an experimental service. It bounces around a lot, so if it's not working for you, try again. Frustration at repeated, unsuccessful attempts can be vented at the author, tom-dot-bruce[somewhere in the vicinity of]cornell-dot-edu]
This "guided tour" will walk you through the prototype OAI repository server at the Legal Information Institute. It includes a little bit of explanatory text, but
All metadata standards are compromises between completeness and accuracy of representation (on one hand) and ease of formulation and implementation on the other. Gathering support for any particular standard or set of standards requires momentum that can easily be lost in debate over minutiae. There is a built-in tendency to make the perfect the enemy of the good, when what is need are functional standards that are designed not to stand in the way of their own refinement at some later time.