How to analyse your own games without letting the engine think for you

The standard way to review a game takes four minutes. Load it, press analyse, watch the little symbols appear, notice that you blundered on move twenty six, feel briefly bad about it, close the tab. Nothing about this changes how you will play next week, because it answered a question you already knew the answer to.

The engine is extraordinary at telling you what happened. It has no opinion whatsoever on why you did it, and why you did it is the only part you can fix. Here is a method that puts those two things in the right order.

01The engine goes second. Always.

This is the whole method compressed into one rule, and it is the one people skip.

The moment you see an evaluation, your memory of the game rewrites itself. A move you agonised over for eleven minutes becomes obvious. A move you played instantly becomes something you nearly saw. You lose access to the thing you came for, which is an accurate record of what you were thinking when you had no idea what was going on.

02Step one · Annotate blind

Open the game with no engine running. Walk through it and write, in plain words, at every point where you had a real decision:

• What you thought the position needed. "I wanted to trade the dark squared bishops." "I thought I was worse and wanted to bail out."
• What you considered and rejected. This is the gold. The move you almost played tells you far more about your chess than the move you played.
• How you felt. Rushed, comfortable, confused, certain. Write it. Half of all mistakes are visible in this column before they are visible on the board.

Ten to fifteen notes across a whole game is plenty. This takes about twenty minutes and produces something no engine and no coach can reconstruct later.

Notice what that note contains. Not an evaluation. A cause. The player did not lose because of the bishop move, they lost because they had no plan and used moves to fill the gap. That is a fixable habit, and no amount of engine output would have surfaced it.

03Step two · Find the turning points, not the blunders

A blunder is where the evaluation moved. A turning point is where the game changed character, and the two are usually several moves apart.

Before touching the engine, mark three positions in the game:

1. 01The last position you understoodThe final moment where you knew what you were doing and why. Everything after this is where your preparation and your understanding ended.
2. 02The moment the game got difficultUsually a structural change: a pawn break, a big trade, an open file appearing. This is where the plan should have changed and often did not.
3. 03The decision you would take backBy instinct, before you check. You are usually right, and when you are wrong that is itself worth knowing.

Positions like the one below are typical turning points. Nothing is hanging, no tactic is available, and the entire game hinges on whether you know what this structure wants.

Games are lost in positions like this far more often than they are lost to one move hallucinations, and the engine is close to useless at flagging them, because at the moment of the mistake nothing measurable has happened yet.

04Step three · Now switch on the engine

With your notes written and your three positions marked, run the analysis. You are no longer looking for mistakes. You are checking a hypothesis, which is a completely different task and much faster.

Work through the output in this order:

1. 01Check your three turning points first. Were you right about where the game changed? If the engine agrees, your positional instincts are working. If it points somewhere else entirely, that gap is your next month of study.
2. 02Look at the moves you rejected. The candidate you almost played is more informative than the best move. If your rejected move was good, the problem is confidence. If it was bad, the problem is evaluation, and those need different fixes.
3. 03Only then look at the big evaluation swings. These are the obvious errors. They matter, but they are the easiest part and they are usually symptoms of something upstream.

The engine is a witness, not a coach. It will confirm your account or contradict it. It will never tell you what to work on.

05Step four · Classify by cause, not by size

This is the step that turns review into improvement. Every mistake you found gets exactly one label, describing why it happened rather than how much it cost.

Do this for twenty games and a pattern appears that no single game could show you. Most players find that two categories account for the large majority of their losses, and that at least one of them is not a chess problem at all.

Notice what this does to your study plan. The player above should not be doing more tactics puzzles, which is what almost everyone does by default. They should be playing through annotated master games and fixing their clock habits.

06Step five · Turn findings into something you will meet again

An insight that stays in a notebook is a hobby. Each review should produce at most two concrete artefacts:

• A position to train. If the mistake was knowledge or evaluation, save the position and come back to it in a few days, then a few weeks. Recognition is built by repetition with gaps, not by understanding something once.
• A repertoire edit. If you left your preparation and disliked what followed, that is a hole in the repertoire, and it should be patched while the discomfort is fresh.

Two artefacts per game, one review a week, and after three months you have a personal training set built entirely from your own failures. It is the most efficient study material that exists, because every position in it is one you have already proved you get wrong.

07What one review actually produces

It helps to see the output rather than the process. Here is what a single properly reviewed loss looks like when it is finished, written out in full, because the whole thing fits in a paragraph.

Four sentences. No variations, no evaluations copied out, nothing that requires the software to reread. Notice that the labelled cause sits eight moves before the move the engine flagged, which is the normal distance and the reason engine first review teaches so little.

That review took thirty five minutes and produced two positions you will meet again. Do it forty times in a year and you have eighty positions that were each chosen by your own failures. No purchased course can match that hit rate, because no course knows what you specifically do not understand.

08Keep an error book

One document, appended to after every review, containing nothing but one line per game. Date, colour, cause label, and the sentence describing what actually happened. That is the entire format.

Its value is not in any single line, it is in reading the whole thing every three months. Three things surface reliably:

1. 01The repeat offender. A phrase you have written five times in slightly different words. That is not five mistakes, it is one habit with five instances, and it can be worked on directly.
2. 02The false priority. The thing you have been studying all year that never appears in the book at all. Almost everyone finds one, and it is usually the topic they enjoy studying most.
3. 03The improvement you did not notice. A cause that used to appear constantly and has quietly stopped. Ratings move slowly and hide this; the error book shows it clearly, and it is the best reason to keep one.

The book also solves the motivation problem that kills most review habits. Reviewing a loss is unpleasant, and the unpleasantness is what stops people. Writing one line into a document that is visibly making you better is a different experience from reopening a game you would rather forget.

09What to do with the games you won

Review one in three. Wins hide more errors than losses do, because nothing forces you to look at them, and a game you won by a bad plan teaches you to repeat the bad plan.

The question for a win is narrower: at what point was I actually winning, and was it for the reason I thought? Players routinely discover that the game was decided ten moves before they believed, by something they were not paying attention to. That is a more useful discovery than anything a loss will hand you.

10How much, how often

Depth beats coverage by a wide margin. One game reviewed properly is worth twenty games run through an engine.

1. 01One deep review a weekAround forty minutes. Blind annotation, turning points, engine, classification, artefacts.
2. 02A two minute triage on everything elseWhere did I leave preparation, and was there one moment I would take back. Log it. That is enough.
3. 03A pattern pass every twenty gamesCount your cause labels. Let the count decide what you study next, rather than what you feel like studying.

11Four ways this goes wrong

• Reviewing with the engine open in the background. Even a glimpse of the evaluation contaminates the annotation. Close it.
• Collecting long engine lines. A fourteen move variation you cannot reproduce tomorrow is entertainment. Write the idea in a sentence or do not write it.
• Only reviewing disasters. The games where you were confused throughout teach more than the ones where you hung a rook.
• Reviewing immediately after playing. Wait a day if you can. You want to see the position, not relive the emotion.

The same discipline applied to somebody else’s games becomes opponent preparation, where you are hunting for the habits behind their mistakes rather than your own.