Tagging Your Losses By Cause, Not By Opening

Ask a 1400 why they lost on Saturday night and you will get an ECO code. “I don’t really know the Caro-Kann Advance.” “My Najdorf falls apart around move 12.” “I need a proper answer to the London.” The diagnosis arrives instantly, fully formed, and with a shopping list attached: a Chessable course, a YouTube playlist, forty minutes a night for six weeks.

It is almost never true.

Openings absorb the blame because openings have names. Calculation hygiene does not have a name, it has no ECO code, nobody sells a 4,800-move repertoire for “checking your opponent’s checks before you move.” Chess.com’s Game Review makes this worse by handing you an Opening score out of 100, which quietly implies the opening is a component that can be broken. Meanwhile the thing that actually lost you the game happened on move 23, took you six seconds, and you have already forgotten it, because you were calm on move 8 and panicking on move 23 and memory rewards calm.

So here is a different postmortem. Instead of naming the position where the game started, name the cause of the move where the game ended. Do it across twenty games, tally the tags, and read the tally. Almost everyone in the 1000 to 1900 band finds the same two words at the top of the list, and neither of them is an opening.

Six categories, and nothing else

The whole point of a taxonomy is that it forces a choice. If you allow yourself seventeen categories you will invent a new one every game and learn nothing. Six is enough to be honest and few enough to be countable.

CALC. Calculation hygiene. You had the information. You did not process it. Three sub-tags, because the distinction matters for training:

• CALC-THEM: you missed a forcing move for your opponent. Check, capture, or threat you never looked at.
• CALC-ME: you missed a forcing move for yourself. The tactic was there, you played something solid instead.
• CALC-STOP: you saw the line, stopped two moves early, and evaluated the wrong position as the end of it.

CLOCK. Time use. The move was bad because of when and how fast it was played, not because the position was hard. Either you had time and did not spend it, or you had already spent it somewhere it did not matter.

JUDGE. Positional judgement. The right move was on your candidate list and you rejected it. You traded the wrong bishop, opened the wrong file, accepted a structure you could not play. This is a real category and it is smaller than you think.

PLAN. Drift. Three or four moves in a row with no target. Rook to e1, queen to d2, king to h1, and the eval slides from +0.4 to -0.9 without a single tactical error. You were not trying to do anything.

BOOK. Opening knowledge. Strict definition: the engine eval is worse than -0.80 for you by move 12, and the cause is a move you played because you did not know the line. Not “I felt uncomfortable.” Not “my position looked passive on move 18.” A concrete number, before move 12.

TECH. Endgame technique. A known theoretical result, botched. Rook and pawn versus rook, Lucena, wrong-coloured bishop, king and pawn races. You reached a position with a name and did not know its rules.

The tie-break rule

CALC and CLOCK overlap constantly, and if you do not settle the overlap in advance you will tag whichever one you feel worse about. Use this: if the move took under 10 seconds and you had more than five minutes on the clock, it is CLOCK. If you thought for 45 seconds or more and still played it, it is CALC. The question being answered is “what would have had to change to prevent this move,” and for a six-second move in a nine-minute position, the answer is not more tactics puzzles.

BOOK and PLAN also collide. Resolve it with the move-12 eval. If you were at -0.20 on move 12 and -1.10 on move 20 with no tactics in between, the opening did its job and you drifted. That is PLAN.

Getting the numbers out

You need three things per game: the engine eval before and after each move, the time you spent on each move, and the move number. Lichess gives you all three for free, which is why this is a Lichess workflow even if you play on Chess.com.

Request a computer analysis on each game (server-side Stockfish with NNUE, typically depth 20 and up, which is far beyond what you need to identify a 1.5-pawn swing). Then pull your PGNs with clock data attached:

https://lichess.org/api/games/user/YOURNAME?max=20&perfType=rapid,classical&clocks=true&evals=true&opening=true

That returns PGN with both annotations inline:

19. Rfe1 { [%eval 1.42] [%clk 0:06:19] }
    Bxh2+ { [%eval -1.88] [%clk 0:04:02] }
20. Kxh2 { [%eval -1.91] [%clk 0:06:12] }

Seven seconds between 0:06:19 and 0:06:12. An eval swing of 330 centipawns. A queen and a bishop already pointing at h2. That is not an opening problem, and the Chessable course will not touch it.

For local work where you want multiple candidate lines at once, En Croissant or Nibbler running your own Stockfish binary will show you multipv output, which is what you need for JUDGE calls specifically: if the top three moves are within 20 centipawns of each other, the position was not a test and your “mistake” was noise. The general method for interrogating the engine rather than just reading its verdict is covered in Analysing Your Own Games With An Engine; this post assumes you can already get an eval on screen and want to know what to do with thirty-one of them.

Tag threshold, and stick to it: any move where the eval swings 150 centipawns or more against you. Maximum two tags per game, the first two chronologically. Ignore everything once the eval passes -4.00, because a 1400 losing a lost position is not a leak, it is arithmetic.

What twenty games actually looks like

Here is the tally from a real 1480-rated rapid player’s last twenty games (15+10 and 10+0, mixed colours, no bullet). Thirty-one tagged errors across the twenty games: eleven games had two, nine had one.

Before this exercise, the player’s own stated weakness was “my openings with Black, especially against 1.e4.” Two tags out of thirty-one. Both in the same line of the Alapin, both a single move, both fixable in about forty minutes with a Lichess Study and no repetition. Not forty hours.

The TECH zero is worth a second look, because it is not good news. Fourteen of the twenty games ended before move 35. You cannot have an endgame leak if you never reach an endgame, and the reason the endgames never arrive is sitting in the first two rows of the table.

The clock game

One of those eight CLOCK tags deserves spelling out, because it is the pattern nobody self-diagnoses.

A 15+10 game, Black, quiet Queen’s Gambit Declined structure. On move 14 the player spent 4 minutes 41 seconds choosing between Rc8, Qb6, and h6. Stockfish 17.1 at depth 28 rates those three at -0.11, -0.14, and -0.19. The position had no test in it. All that thinking bought fourteen hundredths of a pawn.

Consequence arrived on move 31. Rook endgame, eval +3.10, a completely won position, 1:12 remaining plus increment. Moves 31 through 36 took 8, 11, 6, 9, 7 and 5 seconds. The eval line reads +3.10, +2.95, +1.40, +0.30, 0.00, and then the draw. Two tags for this game, both CLOCK, both pointing at move 14 rather than move 33. The technique was fine. The budget was spent on a position that did not need it.

Do the arithmetic on your own time control once and you will never un-see it. A 15+10 game lasting 40 moves gives you 900 + 400 = 1,300 seconds, about 32 seconds a move. Spending 281 of them on move 14 means the remaining 26 moves average under 25 seconds each, and three or four of those moves genuinely needed two minutes.

Turning the tally into a week

Fourteen CALC tags, seven of them CALC-THEM, means the intervention is a ritual, not a curriculum. Before every move in a slow game: opponent’s checks, opponent’s captures, opponent’s threats, in that order, out loud if you are alone. It takes eight seconds. Seven of the thirty-one errors above die to eight seconds of discipline, and five more (the CALC-ME set) die to running the same sweep for your own moves.

Eight CLOCK tags means a written rule taped next to the monitor. Two candidates: no move under 10 seconds between move 12 and move 35 unless it is a forced recapture, and no single move over 90 seconds unless the pawn structure is about to change permanently or material is being offered. Then audit compliance from the [%clk] tags at the end of each session, because a rule you do not measure is a mood.

Four JUDGE tags across twenty games is roughly one every five games, and it is the category improvers most want to be their problem, because it sounds like the interesting kind of weakness. It is not your problem yet. Come back to it when CALC is under six.

The next time you finish a loss and hear yourself reaching for the name of the opening, stop and open the game with clock tags visible instead. Find the first 150-centipawn swing. Look at how many seconds it took and how many minutes you still had. Write the tag in a spreadsheet with four columns (date, game URL, move number, tag) and do not analyse the tally until you have twenty rows. The twenty-first row is where you find out what you have actually been practising all year.