Building A Blunder Log That Actually Changes Your Play

Most improvers who keep a chess blunder log are keeping the wrong thing. They screenshot the position, note “missed Nxe5 forking queen and rook,” tag it “tactics,” and move on. Six months later they have 140 entries and the same rating. The log grew; the play didn’t.

Here’s why. A position plus a correct move is a puzzle. You already know thousands of puzzles. What you don’t know is the specific mental event that happened at the board: the thing you believed that wasn’t true, and the signal on the board that would have told you so. That’s the part your log needs to capture, because that’s the part that repeats.

The three-column log

Strip everything else out. Three columns:

That’s it. No engine lines, no annotations, no “interesting alternative was.” Those go in your analysis file. The log is for diagnosis.

Column two is the hard one and the one everybody skips. A false belief is a sentence you can be wrong about, stated in the words you actually used in your head. “I miscalculated” is not a false belief. “I thought the knight was defended” is. “I didn’t see it” is not a false belief. “I assumed he’d recapture on d5 because that’s what the pawn structure wanted” is.

Column three is the cue you failed to register. Not the refutation, the signal. There is a difference between “Qh4+ wins the rook” and “two of his pieces were aimed at h2 and my h-pawn had moved.” The second one is transferable to games you haven’t played yet.

Extracting the raw material

You need the moments where you went wrong, and you need them with the reasoning still attached, which means you have to do this within a day or two of playing. Memory for your own thought process decays fast.

Start on Lichess. Go to your game, hit “Request a computer analysis,” and wait for the server-side Stockfish pass. What you want from that screen isn’t the arrows, it’s the list on the right. Lichess flags moves as Inaccuracy (roughly 50-100 centipawns lost), Mistake (100-300), and Blunder (300+). Those thresholds are worth knowing because they’re not symmetric in importance: dropping from +0.4 to -0.3 is a 70cp inaccuracy that changed the game’s character entirely, while going from +7.2 to +5.1 is a 210cp “mistake” that changed nothing at all. The tag is a starting point, not a verdict.

Then interrogate. On the Lichess analysis board, turn on the local engine (the toggle at the top of the analysis panel runs Stockfish 16 in your browser via WASM) and let it sit on the critical position. Crank the multi-PV setting to 3 lines. This is the single most useful setting most club players never touch. One line tells you the answer. Three lines tell you the shape of the position:

Depth 28
1. +1.84  22. Bd3 Qb6 23. Be4 Rad8 24. c4
2. +1.71  22. Rfe1 Rad8 23. Bd3 g6 24. Qh4
3. +0.15  22. Rfd1 Qxh2+ 23. Kf1 Rad8

Look at that gap. Lines one and two are within 13 centipawns of each other, which for practical purposes means they’re the same move: both play Bd3, just in a different order. Line three is what you played, and it drops 169cp. So the position had one idea (get the bishop to the b1-h7 diagonal, cover h2) and you missed it. That’s very different from a position where the top three lines are +1.80, +0.90 and +0.10, where there’s a single narrow path and you missed a hard one.

Chess.com’s Game Review gives you the same raw data through a different door. The Analysis board there lets you set multiple lines too, and its “Key Moments” list is decent at finding turning points. What it’s worse at is letting you sit in a position and poke, so I’d run detection there and interrogation on Lichess. If you want the full workflow for the engine side of this, including depth settings and how to avoid the classic trap of treating a +0.6 as meaningfully better than a +0.4, that’s covered in analysing your own games with an engine.

Filling in column two honestly

Now the part no software helps with. Go back to the position before you see the engine line again, and answer one question: what did I think was true here?

This is uncomfortable, and most people’s first attempt is a dodge. Watch the difference:

• Dodge: “Blundered a pawn.”
• Dodge: “Needed to calculate more carefully.”
• Dodge: “Time pressure.”
• Real: “I counted the defenders of e5 as two and there was one, because I’d stopped tracking the knight after it moved to d7 on move 14.”

That last one is a diagnosis. It names a mechanism: your mental board went stale after a piece moved, and you kept using the old picture. That mechanism will appear in twenty more games. The word “blundered” will appear in zero.

If you genuinely can’t reconstruct what you believed, write “no recall” and move on. Don’t invent. A log with 18 honest entries and 6 “no recall” is worth more than one with 24 confabulations, and the fact that you have no recall in blitz and clear recall in rapid is itself a finding about which time control is actually teaching you anything.

Time pressure deserves a word, since it’s the most common dodge. If you played the move with 25 seconds left in a 15+10, the false belief might be procedural rather than positional: “I believed I still had time to work it out, so I spent 4 minutes on move 18 and had nothing left for the critical move 22.” Log that. It’s real, it’s fixable, and it’s invisible if your log only records positions.

Reading the log

Twenty-five entries is roughly where patterns get legible. Sort by column two, not by opening or by phase, and read the belief statements as a block. You are looking for repeated kinds of wrongness, and there are usually only three or four running your whole game.

A log I’d expect from a 1500 might cluster like this:

Stale board picture (piece moved, I kept using old position)   7
Opponent's plan assumed passive ("he's not doing anything")    6
Counted a trade as favourable on material, not on function     5
Committed to a plan formed on move 12, never re-checked        4
One-off / no recall                                            3

Seven instances of “stale board picture” out of 25 is not a tactics problem and no amount of Puzzle Rush will move it. That’s a board-visualisation-and-update problem, and it has a specific drill: before every move you play, name every enemy piece and what it currently attacks, out loud if you’re alone. Tedious for a week. Then automatic.

The second cluster, six entries of “he’s not doing anything,” is a different animal with a different fix. Every move, one question: what does he want to play next? Just the next move, not a plan. Most club players never ask it, which is why the b8-h2 diagonal keeps swallowing their kings.

Notice what the clusters gave you that the puzzle collection couldn’t: a ranked list of your actual failure modes, with counts. That’s a training plan. You now know that five hours on visualisation drills is a better spend than five hours on tactics, and you know it from your own games rather than from a coaching article.

Running it as a habit

Cadence beats completeness. Three entries per game, maximum. If a game gave you eight blunders, take the three where the belief is clearest and leave the rest. The value is in the reading, and a log you’re behind on is a log you never read.

I’d use a plain spreadsheet, Google Sheets or Excel, with one row per entry plus a date, a FEN, and the time control. Four extra characters of setup and you can filter. Lichess Studies are excellent for storing the positions themselves (chapter per game, arrows and text comments), but they’re bad at the sort-and-count operation that makes the log useful, so keep both: Study for the boards, sheet for the beliefs.

Review on a schedule, not on a whim. Every Sunday, read the last four weeks’ belief column top to bottom. It takes eight minutes. Retire a cluster when it hasn’t appeared in three weeks; if “stale board picture” was your top cluster in March and it’s gone by May, the drill worked and you get to stop doing it. That retirement is the only proof you’ll ever get that a training choice paid off, and it’s why the log has to name mechanisms instead of moves.

One warning. The log will tell you unflattering things, and the temptation is to soften column two into something more forgivable. “I underestimated the position’s sharpness” reads better than “I was bored and wanted the game to be over.” Write the second one. The bored entries cluster too, usually around move 25 in longer games, and the fix (a deliberate 90-second stop at move 25) is one of the cheapest rating points available to an adult improver who mostly plays in the evening after work.

Start tonight, with one game and three rows.