To begin, I'm assuming you mean "recognition" as recognizing a specific case with an algorithm commonly designated to it. I think the categorization of "look" in F2L is a bit murky due to lookahead. For example, if we can plan out the first one or two pairs in inspection? Shouldn't we count that planning as 1-look? In addition, what if we are able to plan our solution for an F2L pair without even having to recognizing the pair? To elaborate, once you get decent at lookahead, you should be able to track the following F2L pair as you are solving the current one. What this will do is allow you to find the pieces quicker, thus reducing pauses. What it won't necessarily do is allow you to recognize the case significantly quicker, at least in my experience. As I was learning to do lookahead, I could track the pieces, but I still paused just a bit to recognize the case I had. I had to recognize the case and execute the algorithm, your definition of a "look". However, once you get to a certain speed, you don't even need to track necessarily. You just see the positions of the pieces and work out what the next part of your solution without really having to necessarily track the pieces. Most notably, I'd say this can happen when either a corner or edge is in its right place. There's no point in tracking, you just know what's going to happen. You're not exactly "recognizing a case", so is it a "look"? In addition, what if you can predict PLL based on the OLL? Is it a look still? What if someone performed a cancellation in a transition between pairs? Is that still 2 looks, or just 1 look. There's so many algorithm/lookahead recognition strategies nowadays that it seems very difficult to really be able to categorize something as a "look".
For clarification, I'm putting your definition of "recognition" and adding "at-the-time recognition", which is what I interpreted it as. This basically means that it requires you to recognize the case when you see it. Some examples of how it could go in CFOP could be down to its greatest reduction:
1. Cross+F2L-1
2. Rest of F2L done via lookahead, no at-the-time recognition required
3. OLL
4. PLL
or even:
1. All of F2L done incredibly fluidly with lookahead. As you solve a pair, you are able to plan the next pair solution without having to do "at-the-time" recognition.
2. OLL
3. PLL
and to it's very limits:
1. All of F2L done via fluid lookahead.
2. 1LLL/ZBLL, as you said.
For elite speedcubers, I'd argue that with lookahead they are probably around the 4/3-look solve right now. Their F2L is so fluid and lacking recog pause that it seems like they just know the moves to do next, and there's no need to recognize a case, because it's just muscle memory after so much practice. I know I can do that sometimes.
Overall, my point is that what a "look" is nowadays has a ton of gray matter, and it's difficult to really discern what is considered a "look" in a solve anymore, especially on an elite level. That being said, I believe a complete 1LLL set would be interesting to develop, though I believe there'd be little following for it.
But then again, there's always Jabari...
After First Layer.
After Corners.
After L8E. Solved!
