How to review your own chess games and what to look for
A review is not an engine reading out better moves. It is you finding the two or three moments where the game actually changed, recovering what you were thinking at each one, and writing down what the position was asking for instead. The engine is there to keep you honest about which moments those were.
Write your version down before the engine speaks
Open the game and, before any analysis, write two or three sentences: where you thought the game turned, what you were trying to do in the opening, and the move you already suspect was wrong. This takes a minute and it is most of the value of the exercise. Without it you will read the engine's answer and feel that you knew it, which teaches you nothing.
What you are testing is not your chess. It is your sense of your own chess - whether the moment you remember as decisive is the moment that actually decided it. Most of the time it is not, and the gap between the two is more useful than any single best move.
Find the moments the game turned
A game is not thirty mistakes; it is usually two or three. Look for the moves where the assessment jumped - equal to clearly worse, winning to merely better, lost to drawn. Those are the turning points, and they are the only places worth real time. Every other inaccuracy is noise you cannot act on.
Mikhail's review runs an engine pass over every move and marks the ones where the loss was real, so the turning points are picked out for you. It also names the phase each one lives in, which matters more than it sounds: three errors on move 12 and three errors on move 45 are two different problems with two different fixes.
Ask three questions at each turning point
First: what did I think was happening? Reconstruct the plan honestly, including the wrong version. "I thought his knight was defended." "I was playing for the attack and never looked at the queenside." The answer is often embarrassing and always informative.
Second: what was actually happening? Not the best move yet - the description. Whose pieces were better placed, which side of the board had the play, what was hanging, what the opponent was threatening. Most errors are description errors before they are calculation errors.
Third: what would I have needed to notice, in time, to play the better move? This is the only question whose answer can change your next game. "I would have needed to check every capture" is trainable. "I should have played Rd1" is not, because that position will never occur again.
Name the opening you actually played
Almost every review skips this and then discusses the middlegame as though it arrived from nowhere. It is worth knowing which named line you were in, how far you stayed in it, and the exact move where one of you left it. That move is a fact about your preparation, and it is usually much earlier than players expect.
Mikhail traces your moves through its named-line catalog and reports the line, its ECO code, how many plies stayed in theory and the exact move that left it - including whether it was you or your opponent who left first. When the game did not start from the standard position it names nothing at all, because a non-standard start has no book to be out of.
Leave with one sentence, not a list
A review that ends in fourteen resolutions changes nothing. End with one sentence in the form "when X happens, I will do Y" - when the position opens, I will list every check before moving; when I am up material, I will spend thirty seconds on the opponent's threats. One sentence per game, kept somewhere you will read it, beats a page of general intentions.
Then check it against your other reviews. If the same sentence keeps appearing, you have found a habit rather than an accident, and that is the thing to train. If every game produces a different sentence, your errors are still scattered and the useful next step is volume: more games, reviewed the same way.
What a review cannot tell you
An engine can say a move lost half a pawn. It cannot say whether you were tired, whether you had eleven seconds left, or whether you had decided the plan four moves earlier and stopped looking. Those are the causes, and only you have access to them. The measurement narrows the search; the explanation is still your work.
Questions
How long should a review take?
Fifteen to twenty minutes for a serious game is plenty if you spend it on two or three positions instead of spreading it over sixty moves. A review that takes an hour usually means you analysed the whole game rather than the moments that decided it.
Should I use an engine at all?
Yes, but second. Write your own version first, then let the engine check which moments actually mattered. Used first, an engine replaces your judgement; used second, it corrects it.
What if the engine's move makes no sense to me?
Then the useful question is not the move but the position. Play the engine's line on a few moves and ask what changed - which piece became good, which square stopped being available. If it still makes no sense, note it and move on; a move you cannot understand is not yet a move you can learn.
Can I review a game I played over the board?
Yes. Enter the moves as a PGN and paste it in. A PGN pasted for a single review is analysed and not stored - saved game history exists only once you turn on game storage.
Try it on your games
The parts of Mikhail this guide describes are Game review and Paste a PGN, which need a free account — your games and your counts have to live somewhere.