Computer thinks my winning move is a blunder!

Computer thinks my winning move is a blunder!

Nov 21, 2015, 5:27 AM |

I recently played one of my stronger games. I felt I completely outplayed my lower ranked opponents and, for once, made no mistakes. As I do with all my blitz games, I review it carefully with some computer aid - the Android app, which comes with an excellent analyzer.

To my minor disappointment, I apparently did make one blunder so figured I had something to learn. But it was not what I thought.

First of all, here is the game and my final analysis of the whole thing, with the help of ideas from 3 computer engines. (You'll later see why I employed 3). This is a 5|5 blitz game. I managed the time well although by the time I made the g4 break, I 1 minute left with my opponent over 2 minutes so I'm more concerned about my play before then rather than after then.

As you can see, I overall am very happy with how I played this and that I learned a few things on the way. If people are still looking, I'd appreciate comments on

1- Was my reasoning for going for a king side attack correct? Some computer analysis suggests I might have had better objective success using the open Q-side lines to attack.
2 - What could I have done better instead of placing Ng5/Nf3? Was there a way to get g4 earlier? Should I have rook lifted with Rh3-Rg3
3 - Any other ideas I missed!

So onto the topic of the blog. According to the Android app maximum analysis, I made one blunder and one mistake (which is smaller than a blunder).So did I have something to learn from these? The one mistake was on move 23, which was me playing safe in time trouble - I'd spotted a winning continuation and didn't want to spend time finding better.

So what was the blunder?


So instead of playing g4 (+4.68), I should have played the boring g3 (+1.82)? Obviously a glitch in the engine! Just to confirm this, I put the position into self analysis mode and it refused to consider g4 and the rating of the position was certainly no higher than +2. Weird! I guess the Android app is no Stockfish so I threw it at the online analyser.

Here are it's rated 2000 analysis:

What is also interesting here is that just before 21. g4, I rates b5 as an "okay" move (even if it didn't lose to g4, I'd consider it pretty bad) but it is interesting how the analysis jumps from 2.0 after black's move to 3.79 after g4, and then to 4.17 after black plays the best move next move. It seems the online analyser has a similar blind spot to this basic tactic. As for move 20, it prefers the Rh3 rook lift, and doesn't like my Bb1 - which disappointed me because I thought Bb1 was strong.

So finally, I had to get hold of Stockfish, the best free engine in the world to make sure I hadn't gone made. I was sure that g4 was the obvious play and winning, certainly not a blunder.

Not only does Stockfish confirm that 21 g4 is winning and that 20 ... b5 is bad but I also learn that I should have played 20 g4 one move earlier! Both other engines simply missed this. In fact, Stockfish says my chosen 20 Bb1 was the second strongest move and that g4 was only a shade better (3.1 vs 3.0) so I was clearly on the right idea. Stockfish Rh3 wasn't worth much consideration!

So what am I to learn from all this? It seems that even this 1400 player can make better moves than a 2000 rated engine from time to time and that if one really wants to perform serious computer analysis for learning then Stockfish is the way to go. You can't beat 3000 analysis.

And I did indeed play an excellent game today, if I may say so!

Comments appreciated both on the game analysis and what may have happened with the engines!

(p.s. Apologies for the poor formatting. It seems my nicely formatted text in the editor hasn't come through properly)