It's mostly navigating the PDF directories or notes repository, full-text search of my notes, or (less frequently) searching Zotero for bibliographic data. I don't use tagging for this and I'll address full-text search of the documents in a bit. I can't say that either direct navigation or text search of the notes is dominant as I do a lot of both. Having multiple ways to find information is good for redundancy as if one way fails, you can try another. So I don't think the balanced approach I have will change in the future.

For navigating the directories, I have a Python script called cdref that will search the directory names, which has proved to be very useful. If there's one match, it'll go directly to that directory, and if there are multiple, a TUI will pop up and allow me to select the directory I want.

I haven't found full-text search of the documents themselves to be particularly useful because terminology varies, frequently what I'm looking for isn't in the text (could be a figure, for instance), and probably thousands of my documents haven't been OCRed. I think that relying too heavily on full-text search of the documents assumes that other people will organize information in a way useful to me, which isn't realistic [1]. Full-text search of the documents is a part of my system, still, but it's mostly used to find things to put in the directories or notes so that I can easily find the documents again without having to remember the right keywords. (Though I also often keep track of useful keywords.)

Often I won't remember where I keep some things or even if I have a directory or note on something at all. So I might accidentally create a redundant directory or note. But frequently I later realize that and use it as an opportunity to increase the connectivity of my directories and notes through symlinks. Then if I go to the "wrong" place, a symlink will send me where I should go. And if something pops into my head as related, I add a symlink or a note in the README file for a particular directory. (The README files in the directories are separate from the version controlled notes but will eventually merge, as I indicated.) Over the years, I've accumulated a lot of connections like this.

With all of this said, I think the important thing is to find a system that works for you that you can slowly scale over time. It doesn't need to look like my system. I've iteratively developed a system that works for me over 10+ years at this point. The scale is easy if you have a system you contribute a bit to on a regular basis over a long period of time.

[1] I've been also looking into having a large local bibliographic database to in part as an alternative to online scientific search engines like Google Scholar because I don't want to assume such services will always be available.