Computer Analysis Question

Can anyone see why the computer analysis considers my opponent's move here as a blunder worthy of a -22 evaluation? My following move is not marked as a blunder but it then evaluates the position as -2 ish.

Am I missing something or is the computer evaluation on this site just rubbish?
 

Interesting. By reading the computer's anticipated line with the suggested 18. a4, and Bd2, it looks like White was in danger of 18...Bh4. Although I don't see the winning line, Black has various threats: ...Qd5 threatening ...Qxg2#, whereupon Black's earlier ...Bh4 would have prevented Rf2 to defend. Also, after ...Bh4 if White tries g3, ...Nf3+ could be nasty. There must be a winning combination in there somewhere.

(Please leave 15 minutes after this post for me to find and correct errors, as usual!)

Maybe. But I played Bh4 which dropped the evaluation 20 points - that's why I can only suppose the -22 evaluation is just plain wrong, otherwise my move should have also been marked as a blunder.

Looks like a bug.  Pocket Shredder (2600 strength) also picks Nh5 as White's best move, not a blunder.  The resulting position is evaluated at around -.9.

Just using the human brain, I have difficulty getting remotely excited about a4, which appears to do relatively little, offensively or defensively.  

The position is so visibly "yucky" for white that I did search for a win for a while before turning to the silicon, but apparently its not quite there yet.  

Ah ok, thanks.

Yes, just another case where a computer failed at correct analysis. Unless a program sees a forced mate in x moves, I'd rather it just shut up and stop trying to be intelligent, which it is not.

My first initial thought is since black has so much more queenside space, you should have played something like a4.