y' //Inspection
E2' R2' F r B' //First Block (5/5)
U' M2' U M U' R U2 R2' U R' //Second Block (10/15)
F R' F' R U2 R U2' r' //(8/23)
U2 M' U2 M U M2' U M' //(8/31)
31 STM with Roux.
Next: B' U F' U2 L D' F' L' B2 U' L2 F2 U F2 U' B2 U F2 U L

The main thing when comparing ZZ-4 with Yao is the amount of regripping and rotations used in line with movecount and recognition during the reduction phase of a solve. Right now methods that do not require much thinking are the fastest, as recognition is quick, but in this case, rotations on...

There is also the minor set back that after implementing some of those longer algorithms for ZZ-C, you will end up with a 15+ move PLL Diagonal Corner Swap, probably making Last Slot Layer Layer around 30+ Moves. I think the good thing would be to learn 3 different permutation algorithms for...

x2 y' //Inspection
U F R' F //EO
R U R F2 //XEOLine
U' R' L' //BL Square
U L' U R U2 R' U //Multislot
//OCLL Skip
x' U L' D2 L U' L' D2 L2 //Cancel into CPLL
26 HTM
Next: U' L2 D U2 R2 B2 D2 R2 F2 R2 D' B' L2 R2 D U2 B' R U2 L B2

Usually I would just rotate with a "y" to get a different orientation case (Basically just partial color neutrality), but before that I would look to see if the 10 Bad Edges case is actually an easy one, where the F/B Layer has 4 Bad Edges and the other F/B Layer has 3, that are also well...

Scramble (Same as 4Chan's): F R2 F2 L2 F' U2 R2 B' D2 R2 U' B F2 R B' F' R
R' U2 B U2 F' U' F R' //XXEOLine
U' R' U R L U L' //F2L-1 + Forced TSLE Skip
U2 R' U' R' U2 R2 U R2 U R //TTLL
25 HTM with ZZ-CT

Right off the bat I would say your EOLine could be a little faster in terms of TPS (Partially from pausing in some solves).
In some solves, your F2L to F2L Transition has some pauses that add a couple seconds to your solves.
I may try to reconstruct some of your solves to see if there is...

Well, the 500 Algorithms in ZZ is if you want a 1LLL, otherwise you just need 7 O(C)LL's and 21 PLL's for the equivalent 2LLL when comparing to CFOP's 2LLL, which is 28 Algorithms (ZZ) to 78 Algorithms (CFOP).
In the Reddit Forum Post Felik's says that <R, U, L> Gen isn't very good for 2H, so...

Blockbuilding takes some time to get used to, even more from using CFOP for a while.
Usually when I solve the First Square I will have it Oriented 90 Degrees off, unless while I am looking ahead I cannot find much and I decide to solve the same Side's Pair, and having the Square offset will...

A Reconstruction of a 10.584 Solve I got: Here
After a while you should get used to Blockbuilding to the point where it's just Recognize and Execute without thinking much at all. Knowing that the Edges are Oriented help when trying to quickly find out what to do in a Speedsolve because if I...