How it differs from a crawl
A crawl builds or refreshes what a system knows; retrieval uses it for one question. From a server's log the two often look the same, which is why a retrieval claim needs a fetch traceable to a user-triggered query, not just a request.
Why it is rarely measured
The question that triggered it is usually invisible to the site. The ladder at /evidence keeps it as its own step for that reason: it should not be assumed from a crawl or inferred backwards from a citation.