To read Hartmann’s introduction to “Chess Tech University”, visit Part 1 of this article.
Many theoreticians and trainers have described protocols for analyzing one’s own games. Among them we find notable authors like Jacob Aagaard, Jesper Hall, Axel Smith, and Alex Yermolinsky. Perhaps the best discussion of analyzing one’s own games comes from Artur Yusupov, one of Dvoretsky’s finest students and his collaborator on five books. Yusupov writes in “The Analysis of One’s Own Games” that:
“… the analysis of one’s own games is the main means of self-improvement. I am convinced that, without a critical understanding of his own play, it is impossible for a player to develop. Of course, this does not mean that other forms of chess work should not be carried out. You must study the opening, the endgame and the middlegame, and it is exceptionally useful to study the games of strong players. But in general we learn best from our own examples.
Our own games are closer to us than any others. We have played them and have tried to solve the problems that were facing us. In analysis it is possible to check and clarify the evaluations by which we were guided during the play, and determine where they were incorrect, where we played inaccurately. Sometimes the opponent punishes us for our mistakes, but often they remain unnoticed and can be revealed only in analysis.” 
The point of analyzing one’s own games, as Yusupov clearly states, is to try and discover errors in our decision making, in terms of both the moves we made and the ones we didn’t. Everything else in Dvoretsky’s training philosophy depends upon this. As Dvoretsky himself puts it,
“[t]he ability to analyze your thinking, develop rational methods of planning, determine what lies behind mistakes committed and, by contrast, identify your creative successes – it is clear that all this is no less important than the mastery of purely chess subtleties.”
Here we get to the crux of the matter. What are best practices for game analysis, and what role should the computer play in the process? Yusupov mentions four key themes:
(1) “You should find the turning-points” or critical moments in the game, where mistakes were made, the nature of the position changed, etc.
(2) “Seek reasons for your own mistakes” – not just what went wrong, but why.
(3) “Seek new possibilities.” What moves did you miss in your analysis? What ideas might you have considered?
To this I would add the following: (5) make a point of tracking your time expenditure while you play. Are you thinking too long over forced moves? Are you not spending enough time in critical positions? Add this information to your annotations where pertinent. (6) The postmortem is increasingly a lost art today, but it can be very useful to discuss games with your opponents after hostilities end. Never pass up the opportunity to do one.
So you’ve played your game, done your postmortem, and put the game into ChessBase or Fritz. What now? There are two steps to proper analysis, and both are critically important.
First, take some time to recollect what you were thinking about during the game, and put as much of it as you can into your notes. What did you analyze during the game? How did you feel – yes, feel! – during the game? Did some of your opponent’s moves surprise you, and if so, why? You can end lines with Informator evaluations, but you should use plenty of words here.
Do not turn on your engines at this stage. It is essential that your first notes mirror your in-game impressions as closely as possible. What we are doing here is trying to catch glimpse of our intuition and judgment at work, and any introduction of computer analysis at this stage spoils that.
What might this initial analysis look like? Below is my most recent tournament game, played here at our local club in Omaha. My opponent, John Stepp, is someone I’ve played quite a few times, and someone whose tactical skills I’ve learned to respect the hard way! John has been the US Deaf Champion three times over the years, and he was the American entrant in the 1996 World Deaf Championship.
It appears that there are a few critical moments or turning points worth further study:
(1) closing the center with 11. …c4,
(2) moves 17-20 for both sides, particularly 20.Nc4,
(3) and my decision to exchange queens with 22. …Qc6. The opening isn’t particularly critical, as Stepp’s 7.b3 is passive but hard to refute, and my time management was (for once) not horrific.
This first draft of unaided analysis is certainly flawed, as was my over-the-board play. This is a feature and not a bug. I now have the raw material for the second stage in the process, where after a day or two (providing a bit of critical distance), I bring Stockfish and Komodo into the picture.
The point of using the engine to check our moves and analysis is not simply to find improvements, important as that may be. The computer plays the role of publication in Botvinnik’s scheme, subjecting our moves and our ideas to objective criticism. We try to broaden our understanding of the moves we played, the positions on the board, what we evaluated correctly and where our judgment failed us. While we cannot (and should not) hope to play like computers, we can try to harness their insights and sharpen our intuition.
Here is the computer-assisted ‘second draft’ of my game with Stepp.
The computer punched holes in a couple of my lines – computer insertions are marked ‘MF’ or ‘metal friend’ in the text, a habit borrowed from the books of Vladimir Tukmakov – but on the whole I did not analyze badly. Stepp’s 7.b3 was dubious but playable. His sacrificial plan was objectively unsound but could have given him good play, especially in light of my poor response (22. …Qxc6 is a lemon), and the two major errors on moves 32 and 34 were fatal.
Of greater interest are those places in the analysis where the computer challenges or confirms my in-game intuitions. My hunch about the pawn sacrifice after 18. …Bg6 19.Bxf4 Rxf4 20.Qxe6+ Kh8 was correct, but I misevaluated my options fairly dramatically on move 11. I should, on this basis, consider trusting my intuition more while also striving for greater objectivity in calculation.
The position after 14.g4 is worth special note, if for no other reason than it gives me an excuse to talk about the engine window in ChessBase / Fritz.
14…Be8 is a typical French re-routing idea, but I was dogmatic here, playing the move quickly and ignoring other ideas. Stockfish and Komodo, running at depths I trust (27+ ply for Komodo, 30+ for Stockfish), make strong cases for 14. …Qc7! and 14. …Ne8!?. What can we say about them?
First, when we set the engine to show multiple lines of analysis (‘PVs,’ or principle variations; achieved through clicking the plus and minus buttons) we are making a tradeoff. We do get to see two or three ranked moves, but we sacrifice some depth and speed in the process. I tend to toggle between one and two PVs as I work, as I prefer to maximize time to depth over having three or four PV on the screen.
Second, there are some cases – this is not really one of them, but we’ll pretend it is – where we might not understand why the engine recommends a move. What’s so good about 14. …Qc7? If you’re using ChessBase or Fritz, you can ask the engine to show you threats in a position by clicking ‘X’ while the engine is running. It pretends that the side to move gets to make another move (virtually inserting a ‘null move’) and shows you the results. Here we can see that Black threatens 15. …e5. This ‘X-function’ can be very useful indeed, particularly when the engine spits out a move that resists human understanding.
In showing this game, I am not trying to hold my play up as some model for others to follow. I did not play particularly well, and I did not see my opponent’s tactical ideas very clearly. Still I think there is value in using a game like this, warts and all, as an example of how to work on our own games.
Effective chess training, as we have seen, tries to rewire our intuition or ‘chess subconscious,’ burnishing our strengths and shoring up our weaknesses. A serious analysis of one’s own games is fundamental to this task, and used judiciously, the computer can play a critical role in this process. Having heeded the Delphic admonition, what do we do with this self-knowledge? How do we structure our training?
In my case, openings are not a problem, and my tendency to overvalue the bishops was muted here. While I’m good at converting technical advantages, my calculation was not stellar in this game, and I continue to underestimate my opponent’s passed pawns. There are, as always, things I need to do to improve.
My training regimen should take all of this into account. Come back next month to see what this might look like.