Blunder, Mistake, Inaccuracy: Where the Lines Are

Game reviews sort your errors into three buckets. The labels sound like judgements about the kind of error you made. They are not. Each is a range of centipawn loss, and the boundaries were chosen by whoever built the report.

Typical boundaries

Different sites use different numbers, and some scale the thresholds by position. Losing three pawns when you are up twelve is not the same event as losing three pawns in a level game, and the better implementations account for that. Others do not, which is why a game can show a blunder at the point where you were converting a winning position perfectly well.

The labels are ranges, not diagnoses

A move labelled a blunder might be any of these:

• Hanging a piece to a one-move tactic.
• A plausible plan refuted six moves later by a resource nobody would find.
• The only move that avoided a worse outcome, which the engine rates poorly because the alternative was already lost.
• A perfectly reasonable practical decision in a position the engine evaluates by different criteria than a human under a clock.

All four look identical in the report. Only the first is what people mean when they say blunder, and only the first two teach you anything.

Which errors are worth your time

Sort by size and start at the top. Then, for each one, ask a different question than "what should I have played". Ask what you would have needed to notice.

A hanging piece means your check for undefended pieces failed. That is a process problem, not a knowledge problem, and it is fixed by changing what you do before moving rather than by studying the position.

A tactic three moves out is a pattern you have not internalised yet. Look at the refutation and name the motif. Naming it is what makes it recognisable next time.

A slow positional drift across several inaccuracies with no single large loss is the hardest to see and the most valuable. It usually means a plan was wrong from the start, and the engine's alternative plan at the first inaccuracy is the thing to understand.

Ignore the count

Reports often lead with totals: two blunders, four mistakes, nine inaccuracies. That number is close to meaningless, because it depends entirely on the thresholds and the analysis depth. Deepen the analysis and the count rises without your play having changed.

What does not change is which of your moves lost the most, and why. Two blunders in a game is two positions to understand. That is the whole exercise, and it takes about ten minutes.

Scaled thresholds, and why they matter

The better game reviews do not use a flat centipawn threshold. They convert the evaluation into a winning probability first, then measure how much of that probability the move gave away.

The difference shows up at the extremes. Going from +9.0 to +6.0 is a three-pawn loss and a flat threshold calls it a blunder. In winning-probability terms you went from 99% to 98%, which is not a blunder by any useful definition. Going from +0.2 to −0.6 is a smaller centipawn change and a much larger swing in who is likely to win.

If your review flags blunders in positions you were comfortably winning, it is using flat thresholds. That is worth knowing so you can discount them rather than trying to learn from them.

A practical review order

1. Find your single largest loss and understand it fully, including the line that refutes your move.
2. Check whether any other loss in the game shares its cause. Two instances of the same oversight is a pattern.
3. Look at the first move where the evaluation left equality, whatever its label. Games are usually decided before the move that looks decisive.
4. Stop. Three positions is a productive review; forty is a way of avoiding one.

The labels are useful for step one and nearly useless after it. What matters is the size of the loss and the reason behind it, and both are visible without reading a single word of the classification.