What are the estimated average move counts of the four most popular methods? (any estimates for other methods or variations beyond that are welcome, of course) Speed solving Wiki has numbers, but they don't add up:

CFOP: ~55

Petrus: ~50

Roux: 48 (For "speed". For FM it has "28-", but that's not likely to be a fair comparison to these methods, as I want to compare apples to apples: so only estimates which assume some sort of system which can be recalled and performed in a speed solve.)

ZZ: 44 (Assuming ZBLL, though it seems to me that COLL/EPLL would be a better assumption, or just put both. Or Phasing + ZZLL while you're at it.)

- General wisdom and these numbers agree that CFOP has a higher movecount.

- Wiki says that Petrus has a lower movecount than most if not all blockbuilding methods, but these numbers list it as having more moves than either Roux or ZZ, the two other most popular block-building methods. Even if ZZ doesn't count, Roux certainly does. Maybe the "~50" means the half-open interval "[45,55)" and ZZ doesn't count? That could explain it.

- I see people in the forums regularly saying that Roux has a higher potential for speed due to fewer moves than ZZ or CFOP, but this shows ZZ as beating Roux by 4 moves per solve, on average. Is this just people defending their favorite method, or are these numbers wrong? It doesn't say where the 48 and 28 came from, so there's no way to tell if 48 is the best practical average movecount you can achieve, or if that's just what you get when you use only CMLL and intuitive LSE and no other algs, nor what the equivalent number to ZZ's 44 (with ZBLL) is. That is, the average movecount assuming you take the method to it's limit. But if any of this is correct, then it just completely torpedoes Petrus' claim to have the lowest movecount of almost all blockbuilding methods.

- These numbers put ZZ at the lowest move count, but general wisdom says that ZZ compromises between low move count and ability to alg-spam. It shouldn't be the best at either if it's a compromise, or else it would just completely replace methods like Roux that rely on a low move count to make up for fewer opportunities for alg-spam than ZZ or CFOP.

So the only one of these numbers from the Wiki which makes sense is the CFOP one. What's going on here? Are these numbers outdated? Am I misinterpreting them? Are the comments about the method pros and cons outdated? Are there any forum threads I missed which compare and contrast average move counts amongst different methods? It seems like the sort of thing for which a pretty accurate answer could be achieved with a program running a bunch of simulated solves using a hardcoded method such as (for CFOP)

1. Use monte carlo, or brute force, or whatever to solve the quickest cross.

2. Use monte carlo, or brute force, or whatever, to solve F2L. If this is producing inhuman solutions to F2L, add in a restriction like only solving the quickest F2L pair, and then again thrice more. This could be optimized further if necessary by allowing the program to favor F2L pair solutions which reduced the number of moves for the next pair. Just whatever seems to be producing human-like F2L solutions after half a dozen solves or so. Doesn't have to be exact.

3. Apply the appropriate OLL (+ AUF as a human will probably only solve from one angle)

4. Apply the appropriate PLL (+ AUF on algs a human will probably only solve from one angle)

5. AUF if necessary

for example. Then dump the total move count for the solve plus a "," to a .txt file and repeat. You could also save data on movecounts for individual steps while you're at it; may as well.Then load that .txt up in Excel, average the data, grab some other fun stuff like standard deviation or some confidence intervals or whatever you feel like, and do it again with a different method hard-coded. Someone has to have done this already, right? This approach should be able to provide highly accurate answers to this, no? I can't imagine noone has ever done it.