Updated GNOME User Interface Guidelines are now official

None of that was true when I last used i3. YMMV.

I can't exactly put my finger on why it feels that way

Guestures on phones are considered intuitive, but take my phone to 1980 and it would be considered quite difficult to use.

Heavy users of computers develop intuition for more computery, abstract interfaces through exposure, but this is a small slice of people—the rest follow tried & true patterns but just hope for the best, with little confidence or understanding, which is why UI changes are so damaging for them—and is kinda related to how various conventions in video games can read as intuitive and pick-up-and-play to seasoned gamers but are not at all to newbies—watch a young kid with no gaming experience struggle with Super Mario Brothers 1-1 with a D-Pad (four discrete directions) and just 2 buttons. This is a level that acts as a natural tutorial to the game, but it's still a bunch of trial-and-error as an experience, and is difficult because the mapping of the six button presses to actions on the screen is (I think this isn't well-enough appreciated) very abstract. Compare with watching a kid learn to drive an electric kids' car (like a Powerwheel) with forward, reverse, and steering. Similarly complex, but in my experience they pick it up faster. The controls map more closely to the effect, I think is the reason, even though the wheel is fairly abstract. Stick forward, you're going forward. Stick back, you're going back. Push pedal, go. Turn wheel right, go right. Turn wheel left, go left. The least-intuitive thing is easily the brake pedal (if there is one—often they just stop when you let off the "gas"), and of course reversing messes up adults, even, by switching everything around (right to go left, push pedal forward to go backward, et c.).

Examples that I used on here just the other day: the iOS 6 and earlier unlock slider (obviously interactive, looks like a familiar control from the real world that's sometimes used to undo a latch, the only obviously-interactive thing on the screen, follows the motion of your finger as if you were moving a real object—you hardly even need the instructions "slide to unlock"), and the way "folders" worked on same, mostly the way that they made them feel like a physical thing existing under the Home Screen so there was no state change across the entire screen—the screen "split apart" to reveal what was ("always") "under it", leaving part of the Home Screen visible, and sure enough, touching that still-visible part would slide the folder shut again. The folders were also limited in capacity—no abstract scrolling, they were like a physical space, with limits.

Spatial finder on pre-OSX macOS is another, more subtle example. 3-pane browsing, file managers that change as one browses folders rather than opening a new window, and file managers that do open new windows but don't retain their position on the screen, are all more abstract to various degrees. They break the connection between action and effect, and the connection between what's visible and what's being operated on ("this window was that folder, but now it's this other folder? Where'd the first one go?"). In that case (and many), the more-intuitive option may be strictly at-odds with more efficient operation for seasoned users.

Basically: can a user leverage what they know, or do they have to learn how their actions connect to whatever thing you've made those actions do? Everyone exists and grows up in a physical world, which remains fairly static over long periods of time, so things they know from that work well. Computer interfaces have developed some long-lived conventions, but a lot of it churns way too much to be reliably "intuitive".

Sure, but everyone brings some familiarity with the physical world to the table. The first things babies learn are the effects of pushing on things, some idea of weight & momentum & gravity, et c. The best UIs build on that when possible. If you push two things together and they "click", they should stay stuck together until "pulled" apart somehow. This is why touch screens can be (but aren't often, since iOS took a big shift in version 7, anyway) very intuitive—they allow more natural and direct interaction with what's happening on the screen. You can "push" two things together, you can "pull" them apart, you can "squeeze" things, you can press on them, and there's not even the abstraction of a mouse in the way. Even with a mouse, clicking is abstract (I'm clicking down here, something's happening up on the screen), motion is surrogate, "dragging" is very abstract when you think about it, et c, and that's where you start with desktop UIs, before you even throw in anything else.

I agree that discoverability is great, and it's also something computer UIs tend to be terrible at. The #1 distinguishing characteristic of a discoverable UI is that the user feels very safe experimenting. If they're afraid to try things, they can't discover much. UIs rarely achieve this, and you can see the effect in how anxious most people get doing anything even slightly unfamiliar on their computer.