How to Analyze Your Own Chess Games: The Loop That Actually Teaches You Something

Annotate the game before you switch the engine on. That single ordering is what separates a review that teaches you something from twenty minutes of watching a bar move. The engine knows what was best; it has no idea what you were thinking, and what you were thinking is the thing that needs fixing. The loop that works is four steps and takes half an hour a game: write your own notes from the scoresheet, then check them against an engine, then classify every error into a category, then, once a quarter, read the categories back and let the tally choose what you train next. The output of a good review is never "I blundered on move 23." It is "I have stopped calculating one move too early in six of my last ten losses."

The loop

1. Recover the game. From your scoresheet if it was over the board, from the site if it was online.
2. Annotate blind. No engine, no database. Your notes, your reasoning, your alternatives.
3. Check. Now the engine, used to test your notes rather than to replace them.
4. Classify. Every error gets a category, and the categories accumulate.

Half an hour per game is enough. Fifteen minutes is enough for a blitz game and blitz games are worth reviewing less often than people think. What is not enough is the thirty-second version, where you open the computer analysis, read the number of blunders, and close it.

Pass one: annotate blind

Sit down with the score and no assistance and write four things.

The critical moments. Not every move. The three to six positions where the game could have gone more than one way. Mark them, and if you cannot find at least three, look again, because a game where nothing was ever unclear is a game you have not looked at properly.

What you were actually thinking. At each critical moment, write the candidate moves you saw and why you rejected them. This is the most valuable data in the entire exercise and it decays within about a day, which is why the review should happen the same evening.

Where the time went. Note the moves where you spent more than a few minutes. Long thinks on quiet positions and instant replies in sharp ones are both errors, and both are invisible in an engine report.

Your assessment. At two or three points, write down in words who you thought was better and why, before the engine tells you.

For over-the-board games this pass is only possible if the moves exist, which is what a scorebook is for. Write legibly, record your opponent's clock times at a few points, and use the margin for the one-line assessments while the game is fresh. A larger notation notebook is the better object if you want games, notes, and your running error tally in one place. If you play a lot of serious games at home on a sensor board such as the Chessnut Air, the file arrives automatically, which removes the transcription step but not the annotation one; a game that gets recorded and never annotated has been backed up, not reviewed.

Pass two: their side

Before the engine, spend five minutes on the game from your opponent's chair. What were they trying to do? Which of their moves did you not understand at the time, and does it make sense now? Where did they go wrong, and did you see it?

This pass is short and disproportionately useful, because a large share of club losses are not blunders but a failure to notice that the opponent had a plan. Naming their plan in one sentence is a skill, and it is trainable.

Pass three: the engine, used properly

Now run it, and impose three rules.

Only investigate swings. Ignore evaluation differences of a few tenths of a pawn. Those are engine preferences, not human mistakes, and chasing them is the commonest way to waste an evening. Look at the moves where the assessment moves by a decisive margin, roughly a pawn and a half or more, and at any point where the engine's verdict contradicts an assessment you wrote in pass one.

Ask why, not what. The engine's move is the start of the question. Play out its line until you can say in a sentence what idea you missed. If you cannot, the line is above your level and you should note the position and move on rather than memorise moves you do not understand.

Never memorise engine output. Deep computer lines in a middlegame are not reproducible by humans and are not what you are training. What transfers is the idea, stated in words.

The disagreements between your pass-one notes and the engine are the entire yield of the exercise. A move you thought was fine that turns out to lose is a lesson. A move the engine dislikes that you had already flagged as suspicious is confirmation your judgment is working.

Classify every error

Use a fixed, short list, because a taxonomy that keeps growing cannot be tallied. Something like:

• Hung a piece or pawn outright
• Missed a tactic available to me
• Missed a tactic available to my opponent
• Calculation stopped one move too early
• Left the opening badly, or out of preparation
• Wrong plan, correct tactics
• Endgame technique
• Time management

Write the category next to the move in your notes and keep a running count. Ten games in, the shape is usually unmistakable and often not what the player expected. Players convinced they have an opening problem discover eight endgame entries and one opening entry, and that discovery is worth more than any course.

Four ways review goes wrong

Chasing the accuracy score. Online review tools produce a percentage, and percentages are addictive. Accuracy is heavily dependent on how sharp the position was: a quiet game where nobody had a decision can score higher than a complicated one you played well. Treat it as a curiosity, never as a target.

Leaving the engine on from move one. Once you have seen the evaluation, you cannot un-see it, and your pass-one notes become a reconstruction of what the computer already told you. If you review online, work from the plain game first and open the analysis tab only when your own notes are written.

Blaming the opening. It is the most comfortable diagnosis available, because it implies the fix is a purchase. Before accepting it, check where the evaluation actually turned. In most club games the position after the opening was fine and the game was decided twenty moves later, and the tally will say so.

Annotating for an audience. Notes written to look intelligent are useless. The valuable entry is the embarrassing one: I saw the rook was hanging, decided it was defended, and did not check. Nobody else is reading this.

One more that is less a mistake than a missed opportunity. If you have a stronger player available, hand them a game with your pass-one notes attached and ask which of your assessments were wrong. That question gets a far more useful answer than "can you look at my game," because it points them at your reasoning instead of your moves.

What to look for by rating

Under 1200. Almost exclusively the first three categories. Review the game by asking one question at every move: was anything of mine attacked and undefended, and was anything of theirs? Do not spend time on plans yet. Fixing the hanging-piece rate is the whole job, and the review's purpose is to prove to yourself how often it happens.

1200 to 1600. The tactical categories still dominate but calculation depth and opening transitions start to matter. Add a specific question at each critical moment: what did my opponent threaten, and did I check for it before deciding? Most errors in this band are not failures of ability but failures to ask.

1600 to 2000. The interesting entries are wrong plan, correct tactics, and endgame technique. This is where you need vocabulary for what a position wants, which is what Silman's imbalance method supplies: material, minor pieces, pawn structure, space, initiative, king safety, development. Working through the Reassess Your Chess Workbook gives you both the checklist and practice applying it, and its solutions are long and opinionated in a way that is either helpful or exhausting depending on temperament. Use the imbalance list as the format for your pass-one assessments and the "wrong plan" entries stop being vague.

Which games, and how often

Losses first, always. Then draws you should have won or lost. Then wins where you were worse at some point, which are the most commonly skipped and among the most instructive, because a win hides its lessons.

Every serious long game gets the full treatment. Blitz gets a batch treatment at most: run a dozen games, look only for repeated opening problems, and do not annotate them individually. And review the same evening or the next morning, because the record of what you were thinking is the perishable ingredient.

The quarterly read-back

Once every few months, ignore the games and read only the tally. Which category has the most entries? That is your next training block, and it is chosen by evidence rather than by whatever is being recommended online this month.

Two honest cautions. The categories only mean something if you were strict; a review where every error is labelled "missed tactic" is not a diagnosis. And the tally lags reality, so if you fixed something two months ago the old entries will still outnumber the new. Weight the recent games heavily and be willing to rerun the count on the last ten games only.