We have an automatic de-duplication system when you import. Basically how it works is that any reference with an identifier (DOI or PMID) will only have 1 library item, so when we detect an article in your library (or in a file you’re importing) with a matching DOI or PMID we won't import it again. If we don’t match identifiers we compare Title, Authors and Journal (with a 96% accuracy match) to determine duplicates. Even if you've 'copied' the same article into multiple projects and subprojects they're essentially a mirror of the original library item, rather than separate duplicate references so only 1 instance of the article will appear in the All references section.
We also have a option to take care of any duplicates that do make it through the system.
This will give you a list of all detected duplicates and let you choose which to merge. In a given set of duplicate references, any that have an identifier (either DOI or PMID) are considered the master reference, and the notes, tags and highlights from the duplicates will be merged to the master. If we don't find an identifier, the reference with the most complete metadata is considered the master.