I am using FeedAPI as a vetting/turking system to filter out possible unwanted feed items, and the problem here is that:
1) I have to have all feed items as nodes (published or unpublished, they have to be a node, which is messy)
2) If I delete one of these "nodes", the previously "published" nodes end up getting set to their default "unpublished" setting

So I propose a Poison DB, and before anything gets touched by FeedAPI Node, it checks said Poison DB, to make sure:
1) the same node isn't being re-created
2) If the same node exists, it doesn't get unpublished or rather re-created again

For my case, doing a simple check on of the the elements is enough to determine if there is a duplicate, meaning there is a field in the RSS feed that is unique, so it could act as a "key".