Three main features affect whether Nexus will check a remote site for new content:
Metadata Max Age
Component Max Age
Not Found Cache
Configuring these to perfect values in repository configuration to avoid build failures when new content is available at the remote site is not a science, but an art.
Currently there is a UI button an admin can use to Expire Cache on an entire repository. Where a proxy repository contains large numbers of already cached content, checking the remote for all of these is needlessly expensive, when only one specific package may have failed a build.
Using the component or asset delete API does not seem like a good solution either - it does too much and the interface requires knowing the internal UUID of a specific asset, which is going to require a separate REST call. The goal is not to delete what is already cached anyways ( may cause other build failures unrelated), merely ask Nexus to treat the cache as expired for existing cached items.
There should be a way to expire cache of specific paths and their children assets rather than only all or nothing.
There should be a repository format generic REST API that could be used to expire specific items in the not found cache and those assets subject to age timeouts configured on a repository.
Nexus 2.x had an Expire Repository Cache scheduled task that one could specify both the repository and the "path" to affect.