Still a Little Wiggle Room for Humans in Correspondence Chess

     I ran across a game from the 2010 ICCF e-mail world championship that was annotated by Rafael Leitaio (a GM in both OTB and CC) and found the position after white's 12th move interesting. 
     Leitaio noted that he has worked a lot with engines both from the programming/testing side and from the practical analysis side, and he observed that it's important to keep in mind that the only objective evaluations in chess are win, loss, and draw for the side to move and everything else is a practical consideration. 
     He went on to say that what this means for engine analysis is that even in positions they play correctly their evaluations have been tuned by the authors to maximize the engine's chances against other engines and a position from which a top engine will score well against other engines may be a position a human would be advised to avoid. This is especially true OTB because in a tactical situation and engine might see all the tactics perfectly, but that is beyond the scope of humans. He added that if an engine says his position is -0.50 but he understands the position thoroughly, it's better for him to play that position than one that is evaluated at 1.00 in his favor but is so chaotic that he doesn't understand it. That is in OTB, anyway. 
     Of course, engine assisted correspondence chess (as most high level CC play is these days) is another matter. And to be successful it takes good hardware, a strong engine and painstaking preparation. Without tablebases, or a lot of pieces on the board, in the endgame engines may miss the best lines and you will rarely see them sacrificing material for the initiative. Also, since this game was played in 2010, I was curious to see if today's engines which are much stronger than they were 5 years ago could shed any new light on the game. They engines pumped out a confusing array of opinions, so if one is strong enough, there's a little wiggle room for humans in correspondence chess, but I'm not one of them. 
     In the following position after Leitaio played 12.h4 black replied with 12...h5 which was given a question mark because it was a conceptual mistake. The year before this game was played the same position appeared in a Kramnik vs. Svidler game and black played 12...f5, but was still unable to solve his problems in the opening.

     I got to wondering if today's engines were any better at finding a decent move for black. Unfortunately in his notes Leitaio didn't offer any suggestions. So, what should black have played? Was Svidler's 12...f5 OK? Was Schuster's 12...h5 an improvement? What do various engines think?