Computers Still Quite Don't Get It

While following the French League game Shirov-Kramnik, the following position was reached after Black's 19th move:

Shirov-Kramnik 2005, White to move

It is an interesting position. Black has pushed his f-pawn aggressively and perhaps overextended a little, but the pawn is overprotected and there is no immediate counterplay. I fired up Crafty to see what it thought about the position. Imagine my surprise when the verdict after a few seconds of thinking was "about equal" (-0.03). But a closer look at the analysis window shows something else:



Crafty has analyzed the position up to 10 ply and decided it's a draw. Except for the small fact that White mates on move 27! Either the display window lags behind the actual analysis for some reason, or then it's a case of the well-known horizon effect. Basically if one side is down considerable material for no other compensation than a forced mate looming in 15 ply, the engine will crunch numbers for a minute before deciding that, yes, in fact one side has a forced mate. Thankfully letting Crafty run for longer produces more sensible results and the variation given is corrected, but it's still an interesting case of computers producing "credible rubbish". Closer inspection of the variation indicates it to be total guff, but I'm still mystified over why Crafty reports a mate but fails to score the line appropriately.

Incidentally, the game was agreed drawn in the position given. Maybe both players had analyzed this position at home with their engines and decided it's "just a draw (!?)".