Unfortunately this forum is mostly problem-based. Peoples' opinions on the best practices that work for them don't get enough discussion.

If I were to describe my rules of thumb they would be maximize tags and minimize sub-collections. And make full use of Saved Searches.

Tags are most powerful and easy to exploit if you derive most of your tag list *before* you start collecting references (or as early as possible in that process). It is hard to make wholesale changes to tags once you have a large library, if those changes will require finding a lot of items that need tag changes. Your tags may evolve, but in general it's probably better to have a little too much detail in your tag system than not enough. I don't worry too much about redundancy in my tags. I can always search for several tags that I know might be relevant. It's easy to combine redundant tags later (in the left tag pane, just rename the one you don't want to the same name as the one you do).

It requires discipline to assign tags to each new item (as soon as possible after you add the item). But the payback when you can pull a pile of relevant tagged items from your library when you need them is great.

Hierarchical tags would be great, but are not currently available in Zotero (they are in at least one plugin).

I try to minimize the creation of new sub-collections (collection bloat is a problem that a lot of people fall into, which can be hard to come back from). And I never have items in a sub-collection that are not also in a top-level 'domain' collection. That way I can delete or archive sub-collections that have outlived their utility, without worrying that there might be items whose overall categorization will be lost if I delete the sub-collection. I have View\Show Items from Subcollections set to OFF.

He means that if you have a library structure
Library
---Domain
------Subdomain A
------Subdomain B

Every item in the subdomain collection is also placed in domain. FWIW, I typically do the opposite and try to avoid duplication in a tree. I also use tags vs. collections the exact opposite way from Tim (I use tags sparingly and do all major organizing with collections and then search within those), but there's obviously not right/wrong here. Consistency is much more important than any specific mode of organization.

Indeed, it's your choice that the item is also in domain or not. All items are in the "root collection" (that is, the library), but they do not need to be in parent collections of subcollections. Through "View\Show Items from Subcollections", you can view the items in that way, so you would view an item in any subcollection (for example subdomain 1) in all the parent collections (domain)

Just to elaborate on my workflow, what I firstly meant was that everything I add to my Library initially gets put into the top-level knowledge Domain "topic" collection to which it most logically belongs (which for my interests is typically only one Domain, but occasionally more than one). I don't really have "Project" collections, which would probably require a different organization.

The vast bulk of my Library's items are in one of those top-level collections (only). Under those top-level collections there is typically only one level of sub-topic subcollections, that are created only as I need them (and then remain). At that time, I put into that new subcollection just the papers that are best for that subcollection. Those items may get dragged from the broader top-level collection (if I remember what they are), or from a tag search or advanced search.

The number of those sub-topic/sub-collections obviously grows over time (the number of top-level Domains does not change very often). So when good new papers later come into my Library, occasionally I will add them not only to their top-level Domain but also to any sub-collections to which they logically belong. I only really have two levels of collections, with limited overlap at each level, so that is not too onorous.

My heavy reliance on tags for finding items in my Library is largely due to having worked out a thorough tag system very early on in Zotero, having learned the immense value of that, going all the way back to Reference Manager in MS-DOS. I am regularly very pleasantly surprised when I retrieve items that I would not have found in any other way. But tags do require quite a lot of work (reading all new items and adding all relevant tags).

But there is obviously no "one-size-fits-all" for Zotero collection approaches. Which is why it would be good if there was more discussion of workflows around here.

As a major example of very different approaches, I said I do not use "View\Show Items from Subcollections". It just seems like a illogical option to me (and incidentally commonly the source of problems posted here, by people who maybe forget they have it set to ON and see some deletion behaviour they don't understand). Why would one wish to see items in a collection that are not actually in that collection ? It's obviously a completely different way of thinking about one's Library. Perhaps on the plus side it populates parent collections with all their subcollections' items, without having to do any explicit work (which my workflow achieves in another way). But it makes it difficult to delete anything from subcollections, for fear of losing items from all collections above that subcollection (but obviously not from the Library) - because items can only be seen in those collections when they exist in the subcollection. So I suspect it contributes to collection bloat. In my scheme I can delete a subcollection I no longer need without it having any ramifications for its items' membership of higher-level collections (because those items actually *belong* to those collections too).