Reviewing your own games

This game has been analysed for more than a century and a half. Paris, 1858: Paul Morphy plays in a box at the opera house against the Duke Karl of Brunswick and Count Isouard, who consult each other. Press “Review the game”: the engine will go through all the moves, judge the accuracy of both sides and show the key moments.

Press a key moment: the board goes to the position before the move, an orange arrow shows the move played, a green one the engine's choice. You can clear the field with the score and paste your own game in PGN. The engine takes longer over a long game, and has less time for each move, so for a serious review the analysis board is more convenient: there you can also ask the engine about any position.

This chapter is about how to work with such a review. The engine will quickly tell you where the mistake was. It is harder to understand why you made it and what to change so that it does not happen again. Without that, a review turns into a list of someone else's moves that is forgotten the next day.

First on your own, then with the engine

A review starts right after the game, while you remember what you were thinking. Write down where you thought for a long time, which of the opponent's moves you feared, which lines you calculated and what you rejected. Mark the moments where you felt you stood better or worse. This is the most valuable part of a review: no engine knows what was in your head.

The scoresheet helps too. In many tournaments you may write the time on the clock next to a move, and afterwards you see where the extra twenty minutes went. If writing the time is not the custom, mark after the game at least the two or three moves you spent the most time on. Often that is exactly where the key moment was: you felt the position demanded a decision but did not find it.

Then go through the game on your own, without the engine. Find the places where, as you felt it, the evaluation of the position changed. Try to work out whose move changed it. Look at the opening: where the game left the moves you know, and what you did next. If the game was against a person, it is useful to review it together with your opponent: they will tell you what they feared themselves and which of your threats they saw.

And only then switch the engine on. Now you have something to compare: your evaluations and your lines against its own. The most useful places are where you misjudged the position or missed the opponent's move. If you switch the engine on first, you will see the answer but will not learn how you yourself thought in that position. And fixing a mistake in your thinking matters more than fixing a mistake in one move.

How to ask the engine

The engine shows the best move, but a move on its own teaches little. Ask why. Follow the engine's line a few moves ahead and see what position it leads to. Play your own move on the board and see how the engine answers it: the refutation is often clearer than the best move. On the analysis board you can press a line, and the engine will explain its first move step by step: what threats it creates and what changes in the evaluation.

Beware of moves you do not understand. If the engine suggests a move whose point does not become clear even after the line, there is no use learning it by heart: you will not repeat it in a similar position. Better find a move the engine considers almost as good but with a clear idea. A difference of a few hundredths of a pawn decides nothing in a game. A clear plan, on the other hand, you will play again, in another game too.

Do not try to review everything. Two or three key moments per game are enough. For each one, put the conclusion into one sentence: “before capturing, check the replies with check”, “in this structure, do not trade on f3”, “in time trouble, play a safe move”. Such conclusions are the result of a review.

Accuracy and move classes

The engine evaluates a position in pawns, but for a review chances are more convenient: how likely a White win is from this position, on a scale from 0 to 100%. Chances are computed from the evaluation: about 50% when the position is equal, closer to 100% with a big advantage. The review compares each move with the best one and looks at how much of their chances the side to move lost. By that loss the move gets a class.

• Best move: the engine's move or a loss under 1%.
• Good move: a loss under 5%.
• Inaccuracy: 5 to 10%.
• Mistake: 10 to 20%.
• Blunder: over 20%, and also a missed or allowed forced mate.

The accuracy of a single move is computed by a formula: the bigger the loss, the lower the accuracy, and it falls fast. A move with no loss gives 100%, a loss of 10% of the chances about 64%, a loss of 30% about 25%. A side's accuracy is the mean over all its moves. Key moments are the moves on which one of the sides lost the most, plus every blunder involving mate.

These numbers have limits. First: accuracy is a mean, and in a short game a single blunder gets lost among good moves. Second: in a position that is already lost there is almost nothing left to lose, so a move that speeds up the defeat can get a high class. The exception is mate: if a move allows a forced mate, the engine calls it a blunder whatever the evaluation. Third: the engine spends limited time on each move of a review, and with a different budget the classes of debatable moves can change.

So do not chase accuracy for the sake of the number. It is more useful to look at missed wins: moves after which a won position stopped being won. These are errors of technique, and they hurt the most. If such moves repeat, you know how to get an advantage but not how to convert it, and what needs training is the endgame and calm play in a better position.

The Opera game

Let us go through Morphy's game move by move. It is famous for how fast White develops the pieces and opens lines while the opponents' bishop on f8 and rook on h8 stand idle.

The finish of the game is a classic puzzle. It is worth solving on your own, as if you were sitting at the board with White.

What does this game teach in a review? The losers made few outright bad moves, but each of their moves was a little more passive than it needed to be, and the f8-bishop and h8-rook never came into play. White, by contrast, brought a new piece into the fight with every move. This is what a mistake looks like that the engine does not always mark as a blunder: falling behind in development builds up gradually, and the price is paid in one move.

There is a lesson for the winner too. On move eight Morphy could have won a pawn but chose development. The engine considers the capture stronger, and it is hard to argue. But at the board Morphy's decision was practical: after Qxb7 Qb4+ the queens came off, and Morphy apparently preferred to keep the queen for the attack while the opponents lagged in development. When reviewing your own games, it is useful to mark such places separately: where you chose a clear move over the best one, and whether the choice paid off.

What to do with mistakes

It is useful to assign a mistake you find to one of several kinds. Tactical: you did not see a blow, your own or the opponent's. Calculation: you saw the idea but went wrong in the line, stopped too early or missed a reply. Evaluation: you calculated correctly but misjudged the final position. Planning: the pieces worked without a goal, or you chose the wrong plan. Finally, time mistakes: a move in time trouble, or a long think where the move was obvious.

After a dozen games the kinds of mistakes add up to a picture. If tactical slips repeat most often, solve puzzles in the trainer. If it is calculation, train blindfold calculation from the chapter “Calculation and judgement”. If you regularly come out of the opening with a bad position, go back to your repertoire and learn the moves where the game left it. If mistakes come in time trouble, look again at where your time goes early in the game.

Review not only your own mistakes but also your opponent's missed chances. If they could have won and you did not notice, you misjudged the danger. And do not forget the good things: mark the moves you are pleased with and the decisions that paid off. A review that shows nothing but mistakes soon becomes tiresome, and confidence in your strengths is also part of preparation.

One more habit is to turn your mistakes into puzzles. When a review finds a position where you missed a blow, save it and solve it again a week later without looking at the answer. Your own positions stick better than other people's, because you remember how you failed to see the solution at the board. Strong players keep a notebook or a file for this, where they collect their mistakes and the conclusions from them.

Where to get the score

Games played online are saved by themselves: every site lets you download the score in PGN or copy it with one button. A game over the board is written down on a scoresheet, as we did in the chapter “Writing the game down”, and then the moves are transferred to the analysis board. If a move in the score is illegible, restore it from the position: the analysis board itself will point out where the score stopped being possible.

For a game with the bot on the “Play” page, the review is offered right after the last move. This is a convenient way to start: the game is still fresh in your memory, and the opponent will not be offended if you find mistakes in its play.

How much time should a review take? A good rule: no less than the game itself lasted, for games with a long time control. For fast games it is enough to look at the key moments and write down one conclusion. It is better to review one game a week thoroughly than to leaf through twenty.

The opening part is easy to check against a database of games. On lichess.org there is an open database where for every position you can see which moves were played in it and with what result. Find the move where your game left the familiar lines, and see what strong players play in that position. This way opening mistakes are fixed in a few minutes, and next time you will meet the same position prepared.

And if you have the chance, show your games to someone stronger: a coach or a club partner. The engine will tell you which move is better, and a person will explain why you chose a different one and how to think in such positions. A good review with a coach is worth a dozen reviews with the engine, especially early on.

What's next

One chapter of the course is left: what kinds of chess and tournaments there are, how ratings and time controls work, and how to plan your training from here. That is the next chapter.