First, I would like to say that ‘500 Insanely Hard Mate in Two Chess Puzzles’ will soon be out. Most likely in the first half of April. Since I’m only 4 paid subscribers away from reaching 100 wonderful supporters I would like to offer the next 5 who sign up for a yearly paid subscription a free paperback copy of the book.
You just have to sign up for a yearly subscription and live in a country with an Amazon store. Just click and upgrade to paid and I will email you if you are one of the first 5.
Secondly, The Chess Gym is hosting a training session with Dr. Can Kabadayi on his calculation philosophy: goal-setting, candidate move generation, and anticipating the opponent's resources. The session is on Saturday at 20.30 CET / 3.30 PM EDT.
Normalized Popularity Scores of Selected Chess Puzzle Themes on Lichess
The other day I entered into a debate with GM Jacob Aagaard and Andras Toth on X about whether tactics trainers are worse compared to author-selected puzzle books. I see benefits for both the curated selection of puzzles, but also for the algorithmically selected puzzle you are served online. One downside of the human-curated selection is that you sometimes can guess the solution because the material is very thematic.
This made me think about how the Lichess puzzle themes are “curated” and at what rating bands you see the most of a specific theme. The above plot is a try to dive into the Lichess puzzle database. I have divided the puzzles into 100-point rating bands. Then a percentage representation for each theme is calculated within the rating band. The specific theme percentages across the rating bands are then compared and transformed into a 0-100 score. 100 represents the band where the specific theme is shown the most often.
If we look at the fork theme it peaks at the 800-900 rating range and then slowly drops. Skewers also peak at around 900. Deflections and X-ray attacks peak around 1400, and a trapped piece at 1600.
We also see that the long solution puzzles (brown) slowly rise as we progress in difficulty only to go down when the very long solutions begin to take over.
The Lichess rating system in a way curates the themes for the users for when they are ready to work on this specific area. That is also something The Tactics Ladder books will show you. In the first book, there will be a good share of back-rank tactics. When you have learned this pattern you advance to learn new themes not decided by me, but by the rating system created by people solving and failing to solve the puzzles.
I’m not saying puzzle trainers are better than curated puzzle books, but they do offer something else and I hope the plot helps show this a little better and maybe can give future debates about what is the best resource a little more nuance.
/Martin
Interesting data! Do you think the fact some themes are user added could bias the patterns there a bit? For instance, a beginner will only know the very basic tatics, even when something more complex shows up. Conversely, a more advanced player might add only the main theme for the puzzle even when the more basic tactics show up as well...