I finally had a chance to look at this method, and it doesn't sound very good at all.
Let's start out with what I consider the most obvious flaw: the constant M moves. For every M move, you need to do a grip shift, which, if you plan to use them a lot, as you seem to think, then F2L TPS will be a lot lower than without M moves. Roux only gains an efficiency advantage because the M slice is fully unrestricted for both blocks; having only one edge unsolved removes that benefit while still imposing restrictions. I don't know how many moves you think it would save, but I guarantee it isn't more than psuedo-slotting would save, as evidenced by Xuanyi's solves, which he averages nearly sub-50 moves. Even for a more average solver, it wouldn't have any advantage, because lookahead would be worse, and with the grip shifts for M moves, it would not flow well compared to just doing a few rotations for standard F2L. And having to end the solve with M' U2 M on most solves just isn't worth the time you lose to grip shift. Just look at H perms, they may be one of the fastest PLLs, but it is only 7 MU moves, while an E perm can be done in the same amount of time at over double the TPS, as per
this video. M moves are just not ergonomic unless the entire method is built with them in mind, as Roux is, and I can tell this method hasn't had the same foresight.
Now, let's look at another obvious downside to this method, reducing to T ZBLL. I can tell you already, T is NOT the best ZBLL subset to reduce to; there are some really bad algorithms and recognitions. If anything, I would consider Sune ZBLL the best single subset of ZBLL. Based on my own personal experience, I know full Sune, U, 4/6 Pi, 3/6 Anti-Sune, full 2GLL, and a handful of other subsets and cases. And out of U and Sune, the 2 sets I know fully, I am nearly sub-3 for recog-exec-AUF for Sune, while I am closer to mid-3 to 4 seconds for U. Full Sune has been much better optimised than any other OCLL case because it has to compete with just how good doing Sune-PLL is, and this has made it one of the most ergonomic sets. Even recognition is easy: just learn any good way for 3-sticker CP recog, then you only have a single 2x2 block to recog BH with, which makes recognition a lot easier than T, which has a 2x2x3 block to look at that makes doing BH slightly more difficult.
Now, as for actually reducing to a single ZBLL subset, it just adds more moves and complexity, with no upside. The EOFLL algs are not optimised, and they make it difficult to lookahead into SZBLL, which means you NEED 2 looks to solve both, which is just not optimal. AND there are even more M moves, which NEED another grip shift, which lowers TPS and flow. If anything, combining the two steps into one would be better, which is basically a worse OLL since you're not even orienting all the pieces, AND you still need to solve the last cross edge. Then, at that point, it might be better to create algs to solve both steps in one while ALSO solving the last cross edge, but that would probably require more algs than you want. But this is the only way I see this method having any viability, though that is extremely unlikely even with this optimisation.
The last point I have to make, and a point I have been advocating for in recent times, is how many looks are required for any given solve. First, let's look at some popular methods. CFOP requires at least 3 looks: full F2L in inspection, then OLL, then PLL. This assumes the solver can plan full F2L every time, which is a generous assumption, so let's say they can plan Cross+3, which bumps it up to 4 looks consistently. ZBLL is the best method we currently have in this category, requiring 2-3 looks. Cross+3 and/or ZBLS in inspection, which Xuanyi is already capable of doing; if not done in inspection, then ZBLS, then ZBLL, giving consistent 3-look solves at the highest level. Now, let's look at what your method can offer. First, planning a full F2L in inspection is possible, but it would not be as efficient or easy to trace because of the M moves you say are required. Followed by another 2 looks for EOFLL+SZBLL, then another look for ZBLL. At BEST, your method is only on par with CFOP's average at 4-look, but realistically, planning full F2L with your idea of using M moves would make it difficult to see deeper into the solve, which might reduce it to only Cross+3, adding one more look and bringing the average up to 5-look.
All of this just makes this method unviable. Reducing ZBLL to only one subset + PLL isn't worth adding multiple extra steps and convoluting F2L for no gain; CFOP will ALWAYS be better at every level. Even as a transition method between CFOP and ZB, it doesn't work, since just slowly learning ZBLL sets and incorporating them into CFOP with edge control during F2L to increase the chances of a cross on top will be more beneficial.
I seriously doubt this method is more efficient than optimised standard CFOP at sub-55 moves, but if you want to create 5 example solves, send them here; that would be very beneficial for getting an actual feel for the method instead of just text and taking your word for it. That would allow sceptics, like me, and others to see whether this method has any potential, however unlikely it currently sounds.
, but still like I said full zbll isn't very practical , if you have learned full zbll(I think yes)sz isn't even needed for you , like I said sz is for the guys who don't want to sell their time to alg sheets
