In fact, HoML2's design follows with PbtA, GM never rolls. If you attack, you attack, if you are PC and someone attacks you, you pick a defense, justify it fictionally, and invoke it to make a defensive check. (actually, being 4e-esque there needs to be some sort of mechanics here, I haven't quite figured that part out, defensive powers? Every power has a defensive function? Dunno...).
Ironically, "GM never rolls" is a feel choice though, right? The system doesn't care who rolls. Another way to think about feel is to understand what feel we're trying for, and then work our mechanic toward that. For me, saves is something about resistance to fate or anchoring: individual qualities that mean the same
fireball impacts different creatures differently, or that a creature can be quick on its feet while vulnerable to mind control. Attacks could be done as saves, and then with my immersionist preferences for me that would need to be symmetrical.
A possible benefit from giving each power a defensive function is that then perhaps the defense of the last power I triggered rides until the next one. Creating a mechanical game of - this is a great attacking power, but leaves me weak to X versus this is a not so great attacking power, but I'm set to withstand Y. And of course, you can give each character a default defense based on their race or class/level, and mix it up with some powers that override others for a time once triggered. I might predict some pain in balancing due to the coupling, and on the other hand, the coupling creates design space for expression.
Moves in AW tend to be either roll-based, or stat-manipulation-based (use Hard to Act Under Fire; gain +1 armour; etc), or a piece of gear. There are a couple of exceptions in the core playbooks: workspaces, and an arresting skinner, are the two I noticed on a quick skim through the book. So while they are not strictly uniform, there is a degree of consistency.
In this case I was thinking of the kind of limiting or definition relevant to implementation in a CRPG. You have what I might dub "magical" game systems, versus what might be dubbed "systematic". The magical elements are those that in a CRPG will result in numerous special cases. The systematic elements are all predicted from the system. Seeing as swathes of moves can't be predicted from any system, they are magical. The more magical they are, the more special cases. When others speak about efficiency and expressiveness, I think they are touching on desirable systematic qualities. Once we have a sufficiently expressive system, it is cheap to create instances of powers (just so long as you are happy with their limits). One of the more powerful and efficient drivers of expressiveness are combinatorial effects: meaning that you want sufficient diversity that those will layer up.
In my OP, that is what I was driving at. CRPGs are powerfully systematic, and then contain some number of special cases e.g. in their spell "systems" (funnily enough, and you can probably see how DW moves and 5e spells have a fundamental similarity). CRPGs struggle to go very far with magical game elements, whereas humans are wonderful at extrapolating with
sufficient consistency from the barest clues. The trick is to work out what those barest clues are, and how to engineer some systematic facets onto them. In DW, there is a straight-up meta-system for how moves are obtained, plus a number of light-touch mechanics like those you noted, that can be appealed to in moves as desired by their prototypers.
Just-enough-consistency, is very far from consistency, when it comes to engineering it. That's easy to overlook. In fact, humans so naturally work with just-enough-consistency that we barely notice the difference
until we try to engineer it. A CRPG has no choice but be consistent, even in its special cases. A CRPG move Q can't one day do X and another Y... it
must always do X and only X. We can maximise its expressiveness by being very complete about X's meaning to all other parts of the system. A human can easily do X from Q one day, X and Y another, just Z a third, while keeping X, Y and Z
sufficiently consistent that other humans will grasp what is going on and see the threads.
Tying back to my OP, I speculate that 6e might keep 5e's combat system (with just a few refinements that have come out of
extensive use), and where great work could be done is to on the explore and social pillars. To produce a sufficiently consistent magical game system, with good meta-rules and a cadre of mechanics to appeal to (like hold, forward etc). Paying attention to the interfacing of that system with combat. Spells versus skills needs to be looked at and choices made about their distinct jobs. I might imagine spells continuing to be narrower-but-stronger than skills... a set of special cases with their own meta-rules and cadre of mechanics. I think there is a lot of scope for spells to do more work as buffs and riders on skill and combat moves, maybe that is the space they should own?