Disclaimer: I have little experience with chess but i see good reason for that from a theory of game trees kinda angle.
Why are chess puzzles not 50% about spotting and preventing attacks?
Chess is a problem where you need to reason about all the actions your opponents can do. Since a lot of chess teaching tradition comes from a linear medium (books) they have a trouble representing how an attacker would attack differently against if different defensive move are played. This problem does not exist as attacker as one can force certain moves on the defender. Thus the branching factor of the puzzle can often be contained. If books would represent the branching reactions, the book will also spoil which moves to pick, finding a solution to a chess problem would be reduced to reading all the replies and pick the most favourable. Another problem is that "defend for 6 rounds" or "defend all material" might allow and train solutions which survive satisfy that criterion but cause the player to loose (a sacrifice would have been the only way to prevent losing) while solutions exists where this isn't the case. In a linear medium it is not possible to handle that well without just printing a solution and say that is the right one.
Any advice to train that specifically?
You need to leave the linear medium. This is possible very easily by playing certain board configurations against an AI. However generating such a board positions for which there is a few or even one unambiguous chain of moves that constitutes a valid defense is difficult. My wild guess is that they are near the border of endgame libraries for both players. I think a chain of moves would constitute a valid defense if it leads to a state which is in the players winning end game library for all actions the opponent takes however to make this puzzle difficult this chain of move needs to be hard to find and other moves must be actually bad. My proposal would be to search for board positions where the vast majority of moves including some promising moves lead to a position in the attackers end game library while there exists one chain of moves which independent of actions of the attacker lead to a position in the endgame library of the defender. Caveats: finding moves at the border of the two endgame libraries with these properties might be hard. Many positions of these are likely synthetic and would never arise in natural play.
Comments
Disclaimer: I have little experience with chess but i see good reason for that from a theory of game trees kinda angle.
Chess is a problem where you need to reason about all the actions your opponents can do. Since a lot of chess teaching tradition comes from a linear medium (books) they have a trouble representing how an attacker would attack differently against if different defensive move are played. This problem does not exist as attacker as one can force certain moves on the defender. Thus the branching factor of the puzzle can often be contained. If books would represent the branching reactions, the book will also spoil which moves to pick, finding a solution to a chess problem would be reduced to reading all the replies and pick the most favourable. Another problem is that "defend for 6 rounds" or "defend all material" might allow and train solutions which survive satisfy that criterion but cause the player to loose (a sacrifice would have been the only way to prevent losing) while solutions exists where this isn't the case. In a linear medium it is not possible to handle that well without just printing a solution and say that is the right one.
You need to leave the linear medium. This is possible very easily by playing certain board configurations against an AI. However generating such a board positions for which there is a few or even one unambiguous chain of moves that constitutes a valid defense is difficult. My wild guess is that they are near the border of endgame libraries for both players. I think a chain of moves would constitute a valid defense if it leads to a state which is in the players winning end game library for all actions the opponent takes however to make this puzzle difficult this chain of move needs to be hard to find and other moves must be actually bad. My proposal would be to search for board positions where the vast majority of moves including some promising moves lead to a position in the attackers end game library while there exists one chain of moves which independent of actions of the attacker lead to a position in the endgame library of the defender. Caveats: finding moves at the border of the two endgame libraries with these properties might be hard. Many positions of these are likely synthetic and would never arise in natural play.
Today, most professional games of standard chess are draws because of the power of defensive moves.