My Great Idea

Fixing Tactics Training

When you play in a competition, nobody tells you if there is a tactical shot on your board. My solution to train tactical alertness is Puzzle Inception. Here is how it works.

Martin Bennedik is a freelance .NET developer based in Frankfurt am Main, Germany, club player, International Master in correspondence chess and founder of Chesspuzzle.net.
Martin Bennedik is a freelance .NET developer based in Frankfurt am Main, Germany, club player, International Master in correspondence chess and founder of Chesspuzzle.net.

I have been running ChessPuzzle.net since 2016, and in that time it grew from a hobby to one of the biggest chess websites. ChessPuzzle.net offers chess puzzles for players of all levels, from beginner to grandmaster. With the new Puzzle Inception mode, I am addressing one of the most common criticisms of training with chess puzzles: That you always know that a tactical solution is possible. Well, now you don’t.

Puzzle Inception is a new mode, where you get to solve a position without any prior knowledge. First you have to evaluate the position and select the correct evaluation, for example “White win”, or “Black edge”. Some positions have a tactic, some don’t. This makes Puzzle Inception a more accurate representation of real game situations. It is also more exciting to play! If the evaluated position has a tactic and it’s the only way to win, you will be asked to solve it right after you have given your evaluation.

Take this position – it is an example which is more difficult if you don’t know the evaluation:

Black to move: How would you evaluate this? Is there a tactic?
Black to move: How would you evaluate this? Is there a tactic?

At first glance, this position might look about equal. But then you will realize that both your pieces are attacked. For a moment 1…Qxc3 2.Nxc3 Nxh3 3.axb5 might look like a forced variation. You might then realize that the black knight is out of play and White is winning the endgame.

However, the correct solution is that Black is winning! In the following tactical puzzle you will have to find the correct solution 1…Nxe2! 2.Qxc6 Nxd4+ 3.Kf4 Nxc6. This is a typical example of a position that is much more difficult if you don’t know in advance that a tactical solution is available. It is also a demonstration that what goes through your mind when you play Puzzle Inception is much more like during a real chess game.

Tactical alertness is the ability to be aware of and quickly recognize tactical opportunities in a game. These can be anything from a simple fork to a complex combination that involves multiple pieces. The outcome of a chess game often depends on these tactical moments, and being able to see them quickly and accurately converts into decisive advantages or even wins. Puzzle Inception challenges you to evaluate positions for tactics and non-tactical advantages, making it a great way to improve your tactical alertness.

I have been fiddling with this idea for several years, and even tested with several different engines and positions. But it was never quite working. At first the engines were not strong enough in some strategic positions. That changed with the engines using neural nets for their evaluation, but it still didn’t work because their evaluations were sometimes too optimistic. Then came Stockfish 15.1, which is much more consistent with its normalized evaluation function and which is the engine on which Puzzle Inception is based.

Having many human solvers will help me to make an improved next beta version that adjusts the difficulty of the puzzles to the strength of the solver.

I also did many tests to determine which positions are suitable for this, and are evaluated consistently by humans and the modern engines. The first beta version includes about 1000 positions with tactics, and 1000 without. After a few days of the beta test it is already transpiring that some positions with tactics are much harder when you don’t know if there is a tactical solution available or not. There are also positions where you can quite easily recognize that one side is winning, but then it is still very difficult to find the precise moves to convert.

The next position is easy to evaluate but difficult to solve:

White to move. How do you evaluate this? And what would you play?
White to move. How do you evaluate this? And what would you play?

Black’s king is in the center and White has several options to attack it, so it is easy to guess that White is winning. But playing the precise moves is much more difficult, because Black also threatens mate in one by 1…Ra1#.

If many human solvers try Puzzle Inception this will help me to make an improved next beta version. I want it to be able to adjust the difficulty of the puzzles to the strength of the player, so you don’t get puzzles that are way too easy or too difficult for you.