The blunder you keep making isn't in your last game

The short answer

A blunder checker tells you which moves lost material or evaluation in one game. That is a report, not a diagnosis. To find the blunder you keep making, you need twenty games at once, each blunder tagged with a short reason in your own words, and a count of how often each reason appears.

The reason has to be written by you, not by the engine. Stockfish will tell you that 24.Rc1 dropped 380 centipawns. It will not tell you that you played it because you were following a plan on the c-file and stopped checking what the rook on e1 was doing. That second sentence is the thing that repeats.

• Pull your last 20 games in one time control.
• For every move flagged as a mistake or blunder, write one sentence: what you were doing instead of checking.
• Group the sentences. Anything that appears four or more times is your pattern.
• Ignore the one-offs. They are noise and you will not reproduce them.

Two different problems that look identical on the board

A one-off blunder is a failure of attention. You saw the fork was possible, you were down to ninety seconds, you moved. Nothing about your chess is wrong; you ran out of clock or focus. Reviewing it teaches you almost nothing, because the next game will not present the same position.

A recurring blunder is a failure of habit. It has a trigger — a type of position, a type of move by your opponent, a point in the game — and the trigger will show up again this week. This is the one worth an hour.

The engine's output is identical in both cases. Same red icon, same centipawn drop. The difference only appears when you stack games on top of each other.

What the numbers say about where to look

Across 140,377 engine-classified moves from 4,493 games analysed on Mated, 3.8% of moves played are outright blunders and another 4.2% are mistakes. That is roughly one real error every 13 moves, before you count the 18.3% classed as inaccuracies. In a 40-move game you are making three errors that matter. Three per game, twenty games, and you have sixty tagged sentences to sort — enough that a repeat pattern is obvious and a one-off is visibly a one-off.

In the same set, errors are not spread evenly across the game. The middlegame is where they cluster: 9.0% of middlegame moves are a mistake or worse, against 2.3% in the endgame. If you are auditing your own games and you only have an hour, the middlegame is where the material is. That is also the phase where a blunder checker's per-move list is least useful, because middlegame positions are the ones where you were following a plan — and the plan, not the move, is what needs the correction.

The third number is the one that changes how you read your own report. The average move in that set gives away 94 centipawns against the engine's best, but the median is far lower, because a small number of very bad moves carry most of the damage. Your accuracy percentage is being set by a handful of moves per game. Fixing your average move is not a project. Finding the four moves per twenty games that share a cause is.

A worked example

Take a real shape. You are 1250, you play 10+0 rapid, and your last twenty games contain nine blunders that a checker flagged. You write a sentence for each. Five of them come out roughly the same:

Game 3: rook on e1 was holding the e4 pawn, I played Rc1 to double on the c-file, ...Nxe4. Game 7: knight on f3 was covering h4, I played Ne5 to hit c6, ...Qh4. Game 11: bishop on d2 was holding b4, I played Bg5 to pin, ...Nb4 hitting c2. Game 14: same rook-leaves-the-file idea, different pawn. Game 19: queen was the only defender of f2, I brought her to the queenside.

That is not five blunders. That is one blunder played five times. The tag is: when I start a plan on one side of the board, I stop tracking what the moving piece was already doing. The fix is not tactics puzzles — you would have found every one of those tactics in a puzzle trainer, because the puzzle tells you a tactic exists. The fix is a check you run before a piece leaves a square: what is this piece defending, and who was looking at it.

Notice that four of the five are middlegame moves made while executing a plan, which is exactly where the 9.0% figure says errors live. A per-game blunder checker would have shown you five unrelated tactical oversights across five different openings and you would have shrugged at each one.

Tags that tend to earn their place

Write your own sentences first. But if you want a starting vocabulary, these are the categories that repeat often enough to be worth separating. Keep the list short — if you have fourteen tags, you have descriptions, not categories.

Nobody has measured well which of these is the most common cause of blunders at club level. Engines classify moves; they do not classify intentions, and there is no public dataset of players' own explanations for their errors. Treat any confident claim about the distribution, including mine, as opinion.

• Moved a piece that was defending something.
• Missed a forcing move for the opponent: a check, a capture, or a threat I never listed.
• Recaptured automatically and missed an in-between move.
• Kept attacking after the attack had stopped working.
• Blundered within three moves of winning material, while still feeling good about it.
• Made a positionally reasonable move that hangs to a two-move tactic.
• Clock: under 30 seconds, moved without looking.

Record the clock, not just the move

Next to every tagged blunder, write how much time you had. This separates the two problems faster than anything else. If eight of your nine blunders happened under a minute, you do not have a pattern in your chess — you have a pattern in your time management, and the fix is picking your move earlier in positions you have already understood.

If your blunders are spread across the clock, including moves where you had six minutes, the cause is in how you look at positions. That is coachable, and it is the case where the tagging exercise pays for itself.

The mixed case is the common one, in my experience: two clock blunders, four of one type, three unrelated. Fix the four.

When you want this done for you

The audit above takes about ninety minutes by hand for twenty games, and you have to keep the notes somewhere you will actually reopen. Plenty of people do it in a spreadsheet and it works fine. If that is you, you do not need software for it.

Mated exists because most people do not reopen the spreadsheet. It reads your recent games through the Chess.com and Lichess APIs, runs Stockfish over every position in your browser, and scores you across five categories so the grouping step happens automatically — then builds a fifteen-minute daily session from the positions where you actually went wrong, rather than from generic puzzles. The tagging is the valuable part either way; the tool just makes the counting survive past week two.

The blunders that never cost you anything

Here is the part the audit misses. Your opponents at 1200 miss things too. If you move a defender and your opponent does not take, no material changes hands, your evaluation barely moves, and the checker may not even flag it. The habit that produced the move is unchanged and unrecorded.

So the count you build from flagged blunders is a floor, not a total. The same is true in the other direction: some of your wins are not your play, they are somebody else's Game 3.

One way to catch the invisible ones is to review a game you won comfortably. Not the loss — the 25-move win where you were never worse. Look for the moves where the engine's evaluation stayed flat but your reason for the move was wrong. Those are the ones that will cost you a piece against someone 200 points higher, and they are the reason a rating climb sometimes stalls for a month with no obvious blunder to point at.