How to calculate in chess

Most advice on calculation is about depth. The counts here say depth is the smaller half of the problem, and that the harder half is which moves you agree to look at in the first place.

• three-move answers under 800: 1 of 45
• between 1800 and 2100: 87 of 149
• 40% of answers start with a capture
• 399 one-move answers, all of them mate

How deep is deep enough

Depth scales with rating and it is easy to overstate how much. In the rating-drawn pools, answers needing three moves from you go from 1 of 45 below 800 to 87 of 149 between 1800 and 2100 -- 2% to 58%.

Going the other way, one-move answers fall from 12 to 1. And there is a detail in those worth having: across the whole corpus, all 399 single-move answers are a mate in one. Not one of them is simply taking a free piece, because a free piece is not a puzzle -- it is a scan.

Why candidate moves matter more than depth

Here is the number that reframes the whole subject. The winning move is a capture in 4,189 of the 10,440 positions -- 40% -- and that figure barely moves with rating. In the rating-drawn bands it runs from 31% to 47% and never trends anywhere.

So "check the captures first" is good advice at every rating and sufficient at none. Three times out of five the move you need takes nothing at all, which means the failure is usually not that you stopped calculating too early. It is that the right move was never on the list you were calculating.

A method that fits what the positions ask for

1. Ask what the position is offering before you look at any move. An undefended piece, a king with no escape square, two pieces on one line, a piece with no squares. This is pattern recognition, and it is what decides which candidates get written down.
2. List three or four candidate moves, and make one of them quiet. Every forcing move you can see, plus at least one that takes nothing and gives no check. That last one is the habit that separates this from a capture scan.
3. Take one candidate at a time and go to the end of it. Not two moves down every branch. Jumping between lines is how you calculate the same position four times and finish unsure.
4. For each of your moves, find your opponent's best reply, not their worst. Most lines that fall apart do so because the reply that was calculated was the one that loses.
5. Stop when you reach a position you can judge. The end of a line is where you can count material and say who stands better -- not a fixed number of moves.
6. Before you play it, check it once more from the current position. Blunder checks fail because they run inside the line rather than at the board you are actually sitting at.

How to practise it, rather than read about it

Calculation is trained by two things this site can give you: positions where the motif is not named, and positions at your level rather than at somebody's idea of a good level.

• No WarningRating-drawn, motif deliberately withheld
• Mate in twoForcing sequences, exactly two deep
• Mate in threeThe same, one move further out
• Quiet movesWhere the answer takes nothing

Where your band sits

• At 1000 Elo20% of answers run three moves
• At 1500 Elo45% do
• How to stop blunderingThe check at the end of the calculation