Six holes in my bucket
My previous tournament back in december lead to a clear conclusion: my weakest point was making plans for the endgame, followed by a lack of knowledge (skill) of endgame specific tactics. (promotion, sacrifices, key squares, zugzwang, fortress, stalemate, trebuchet et cetera). Expected benefit from specific endgame training: 150 points. Back in December I performed miserably, and I ended up in a lot of endings which were generally better or equal, and I screwed them all up, spilling half points all over the shop.
In May this year I made a discovery about the best method of training for any skill (discovery of The Trick). I decided to swallow my pride and gave up the idea to focus on endgame planning for the moment. Instead I wanted to prove my ideas about the Trick. Proof is more important than rating points.
The past tournament this July showed that the Trick is usable and decisive for each and every game. In all my wins (4 times), my losses (3 times) and my draws (2 times) the Trick was the most decisive ingredient. In all games except one I got the initiative from the start The losses were caused by an opponent who did the Trick better than me or by me failing at the Trick. The two draws showed a miraculous Houdini-like escape by me, followed by a better endgame which I was too lazy to screw up.
I realize that most discoveries that I find captivating are not at all engaging for my readers. So I will limit myself to the highlights.
Areas of improvement
- Tactics overall
- Planning at start of the middlegame
- Endgame planning
Tactics overall
I got a good picture how my tactics compare to equal rated opponents.I filled most holes and I notice that my opponents start to become scared of my tactical prowess. Between December and July I did the following:
- I repeated 400 tactics and 300 mates "old school"
- I learned 200 new tactics according to the new "Trick" method
This is the most important discovery. I lack knowledge and skill in this area. General tactical skill is of no use when you follow the wrong plan. The plans in this area (which exchanges to make, which pawn structure to strive for, improving piece activity, restraining the opponents pieces, provoking weaknesses, pawn breaks et cetera) have their own specific tactics which are not governed by general tactics.
Endgame planning
That I am bad at this was already clear from the penultimate tournament. A new discovery is that Capablanca is was wrong. You shouldn't start with the endgame. Because you need energy and time to play the endgame well. So you must play the previous stages of the game well, otherwise you arrive with a lack of energy and time at the endgame.
So the tournament was a great success because it provides the necessary proof and guidance how to proceed.
Furthermore I must reconsider the amount of games I want to play in a tournament. 9 games (most of them 4-5 hours) plus 4 hours travel time is pretty straining, both mentally and physically.
The openings are of the least concern. I didn't lose points due to the opening. Yet improvement are possible. For white I can look for more active responses. My openings for black are so minimalistic that they even have no name. Certain choices of the author are too much concerned with safety and simplicity. I will look for replacement of certain lines for more aggressive ones.
ReplyDeleteAs stated, the start of the middlegame needs serious improvement.
ReplyDeleteThe end of the middlegame can go in two directions:
* Mate or forcing my opponent to give up wood to prevent mate. This is definitely my forte and where I gain my points.
* Transition to the endgame. Most of the time I get a winning or better or at least not losing endgame. I would be better off when I knew what to look for.
I have a clear picture what needs to be done for a better endgame. I still put it on the backburner, though. I lose too much time and energy in the beginning of the middlegame. I must fix that first.
ReplyDeleteThe picture of the general tactics is very clear. I need to master 1800 puzzles in total and I can do 200 new problems in accordance with the Trick in 3 months. I master 900 problems already now. General tactics influence all stages of the game but especially the kingside attacks.
ReplyDeleteOFF TOPIC (Maybe):
ReplyDeletelichess.org provides the means to download their entire puzzle database. (Their entire database of complete 7.9 BILLION games can also be downloaded but it’s at least 1 TB.) To their credit, they allow using their puzzle and game databases for free with no restrictions:
Database exports are released under the Creative Commons CC0 license.
Use them for research, commercial purpose, publication, anything you like.
You can download, modify and redistribute them, without asking for permission.
Here is the request for feedback:
Did you use this database? Please share your results! contact@lichess.org
LINK: https://database.lichess.org/
I wanted to get the lichess puzzle database downloaded to my computer, but I knew the compressed file was very large. The compressed download file was about 300 MB. When I uncompressed it, it ballooned to 1.095 GB. It's a text file in CSV format; the total number of puzzles is 6,057,356. The first line of the file is a column header.
I normally use LibreOffice Calc to look at CSV files. By default, LibreOffice Calc supports up to 1,048,576 rows per sheet by default, but this can be extended to 16,777,216 rows by enabling very large spreadsheets in the options. Instead of playing with the options, I asked Copilot to generate an HTML/JavaScript program to split a large file into smaller parts. The user specifies the input file and designates where each individual split file will be stored. It worked like a charm!
Now I’ve got all of the lichess puzzles, tags and solutions in 7 smaller files (900,000 puzzles per file except for the last file) on my laptop. I have some vague ideas for developing a training program using those puzzles. Time will tell. . .
I had Copilot generate two programs. The first one asks for a file and then counts the number of records based on newlines (\n) found. The second one can split an arbitrarily large file (again, based on newlines) into parts, labeled with the large filename appended with “Part “ and the sequential smaller file number. (This was so I could get the small file size number of records down to what LibreOffice Calc could handle by default.) Total time to develop: less than 5 minutes plus testing. AI programming can be amazingly productive!
If you want the JavaScript files, send me an email at “robert p coble AT hotmail.com” (remove the blanks and change “AT” to “@”) with the subject: “Please send me the two JavaScript files” (or something similar). It only costs you a little time; otherwise, it’s FREE!
Aagaard has an interesting idea about how to improve the chess area you are worst at. For positional play, take the games of a positional grandmaster Botvinnik, Carlsen and play guess the move for each move on a real chessboard recording your answer on paper
ReplyDeleteAn excellent suggestion. Welcome back to the commentary, Jim!
DeleteIs there software that supports "guess the move" for your own pgn's?
DeleteChess Tempo has Guess The Move but only for a preset list of several hundred master games. (“It’s a big club, but you ain’t in it.” – George Carlin). I could not find any way to simulate it using the PGN Explorer and the Analysis board.
DeleteSo, I asked Copilot for help:
Does lichess provide a function to "guess the move" or play your own games blindfold, covering up the moves and allowing guesses?
[Copilot]
Playing “Guess the Move” on Lichess
Lichess does not have a built‑in “blindfold” or “cover the board” mode, but you can simulate a “guess the move” experience using its Study and Interactive Lesson features.
How to set it up
Create a new study and set it to Interactive Lesson mode.
Import the PGN of the game you want to study. This can be a master game, your own game, or any position you want to guess.
Save the chapter and click the Preview button below the board to see the game move by move.
[One has to select "Interactive lesson" in the Chapter settings, and then the study will automatically appear as intended to other users.]
On each move, you can play your guess on the board. If you’re correct, you’ll get points; if not, you can see the correct move and comments.
You can also access the study from another account so it launches in interactive mode without needing the preview button.
Tips for “blindfold”‑style play
Use the circle button on the analysis board (to the right of the opening book icon) to toggle the board visibility. Press E to enable/disable the opening book, then click the circle to hide/show the board.
This lets you “cover” the board and only see the position, forcing you to think and play without visual cues.
Related features
Guess the Move is a popular study type on Lichess, often used by streamers to replay master games and guess moves.
You can also use ChessMemory or similar tools to import Lichess games and play them as “solitaire chess” or blindfold drills [chessmemory.app].
In short: Lichess doesn’t have a native blindfold mode, but you can create an Interactive Lesson study with a PGN and use the preview + circle button toggle to simulate “guess the move” play. This is the closest built‑in way to achieve that effect.
[END Copilot]
Here’s a link to an example Study:
LINK: https://lichess.org/study/DmduoPE7/4ymCQ474
HTH
I found a couple more options for Guess-The-Move. The second one looks very promising!
Delete1. Chess PGN Master (Android)
LINK: https://pgnmaster.kalab.com/docs/index.html
The free version of Chess PGN Master has a dedicated Guess the Move mode. In this mode:
The move list is hidden.
You tap a piece of the player to move and then a target square to guess the next move.
If your guess is correct, the move is applied, and the engine automatically plays the opponent’s next move.
You can repeat this until the game ends.
You can open PGN files from your device, Google Drive, or a web link, and the app supports annotations, coloring, and arrow drawing for study chrome-stats.com.
2. Guess the Move (chessimprovementtools.com)
LINK: https://chessimprovementtools.com/tools
This is a browser-based tool that lets you:
- Load a PGN file or paste PGN text.
- Select a game, tournament, or player.
- Try to guess moves interactively.
- Save your correct and incorrect guesses with notes, and export them as a PDF or PGN with comments.
INSTRUCTIONS:
Position your mouse over the Decision Training – Guess the move block. Click on OPEN TOOL →.
You can drag and drop a .pgn file into the first block, or you can copy the PGN text and paste it into the second block, or you can select game from a list. Regardless, followup by clicking on the Load Game button.
Select Which side will you guess?: WHITE BLACK BOTH
Set Start guessing from move: Enter the starting move number.
Click on the Start guessing button.
The board will be displayed and you can start guessing. There are some neat capture features to the right of the board.
Everything is stored only in your browser tab and cleared when you close it.
PART I:
ReplyDelete1. Why is it so difficult to learn that there IS a Trick?
2. Why is it so difficult to learn to apply the Trick?
1. A fundamental assumption is that formal KNOWLEDGE is the prerequisite for SKILL. Pedagogical methods are oriented toward transmission of KNOWLEDGE from “teacher” to “student,” not toward acquisition of SKILL such as would occur in a “master – apprentice” relationship.
2. Applying the Trick requires focused attention on every aspect (the context) of a position and the necessary elementary components required for a “solution.” An essential part of the process is “connecting the dots” between what has previously been acquired as SKILL and what is to be added. There is no way to readily formally “teach” the abstraction/generalization process that System 1 must use to store SKILL in long-term memory.
Could Dr. Lasker have been correct in speculating that a teacher armed with an appropriate method could, in 200 hours, ingrain chess SKILL into a young person with no natural talent for chess? If so, what would that process look like? I think the emphasis would be on learning the Trick and how to use it effectively.
PART II:
ReplyDeleteHere’s an example from GM Rowson. The “lesson” is implicitly there; you have to “read between the lines” in order to ferret it out.
“The only thing is that most sources have concrete definite answers. These are still useful for improving concentration, but ideal sources for this kind of work [Stoyko exercises] are positions where the solution is not completely concrete, as is the case with real games. However, the main thing is to get into the habit of sitting down, selecting a problem, setting it up, setting the clock [for about 20 minutes], thinking, stopping, comparing your analysis with the source analysis … and keep doing this. It will be hard, and you will not really be “learning” [KNOW THAT] anything through this. But, as I’ve said before, chess is about SKILL — what you need is not ‘know-that’ but ‘know-how’. One exercise completed with full application for 20 minutes is easily worth an hour of playing through games.”
“I stand by the advice I gave at the end of that lesson [previous paragraph], and now I believe that the main function of chess trainers should be to guide the training of their students, rather than to teach them directly. The best thing you can do for a student is to select interesting positions for them, and analyze [their responses] carefully so you can see the kinds of things that the student is missing. The trouble with this approach is prosaic, but real. Chess tutors are normally paid by the hour, and this tends to make you think that you should be showing the student certain things, imparting KNOWLEDGE, giving hints and tips, etc. This is understandable, but much more useful, I think, is to give the student difficult positions to solve [at or slightly above their current SKILL level], to be there in a supportive role as they solve them, and then carefully consider what the student missed and why [– then provide that information as feedback].”
“I say this based on my experience as a chess tutor, but also my experience of being “taught” by [GM] Artur Yusupov. This consisted almost entirely of attempting to solve exercises that he knew well. The training exercises had definite answers, but they also resembled real positions so if I deviated from the answers, Yusupov would put down his book of answers and I had to deal with the man himself! This took place over the course of five days [NOT 200 hours!] at Artur’s house in Germany. I enjoyed the hospitality of the Yusupov family, but my ego has never taken such a systematic pounding before or since. I imagine that we looked at about 30 different difficult positions, and in most cases I got the first few moves right only to slip up towards the end. Almost never did I get the solution right from start to finish. It made me feel like a very weak player. Yusupov is generous in spirit, and didn’t want me to suffer, but he pointed out, and I think rightly, that I did not calculate like a grandmaster. On the one hand this was discouraging, but it was good to know what I needed to work on, and given that I was almost 2550 at the time I consoled myself with the thought that I must be very good at other aspects of the game. Perhaps, I thought, if I could sort this part (admittedly a fairly major part) of my game out, then I could still make significant strides.”
+~+~+~+~+~+~+~+~+~+~+~+~+~+~+~+~+~+~+
What would “calculate like a grandmaster” look like? It would be based almost exclusively on the Trick, using logical patterns. Admittedly, there are times when a grandmaster has to calculate in an unexperienced position, but most of the time, he is “SEEing” logical patterns and evaluating the consequences based on his expectancies derived from familiarity with similar positions. Most of the “thinking” time is spent on verification of the pattern(s) and the corresponding typical actions, not on finding novel solutions to totally unfamiliar problems.
PART III:
ReplyDeleteIt is fascinating to think that a GM-level [2550] player apparently did not know the Trick!
1. Why is it so difficult to learn that there IS a Trick?
ReplyDelete2. Why is it so difficult to learn to apply the Trick?
In 1999 I suspected that there HAD to be a trick. That was the only way that a 15 year old girl could become a grandmaster. And so the Polgars became my hero's. Why did it take me 23 years to theoretically FORMULATE the trick? There are a few reasons.
*First there is an abundance of advice that is incomplete or plain wrong.
*The second problem is that my mind was easily overwhelmed. I complained a lot about that I wasn't able to apply my daily logic to chess. There were too many parameters to consider, and I had no method to weed out the nonsensical parameters.
*The third problem is that System 1 and System 2 don't talk the same language. It took an awful lot of experimenting to find a method that would at least THEORETICALLY work (23 years).
*The fourth problem was to solve the TRANSFER problem.
*The fifth problem was to experiment with the THEORETICAL solution to make it PRACTICAL (that took another 3-4 years).
*The sixth problem was to make it plausible that the Trick applies to ALL stages of the game and to ALL levels from novice to super grandmaster.
Currently I'm experimenting with the tiniest logical elements. Like mate in 1. It turns out that there are three types: mates I see immediately (20%). Mates I solve fast (70%) and mates that take me time (10%).
The hypothesis is that those 10% are holes in my bucket which I try to plug. And I want to know whether I can improve on the 70% from fast to immediate.
I'm curious:
Delete1. What is/are the mini-skill(s) associated with mate-in-1? The number of moves to mate is not a mini-skill, AFAIK.
2. What is/are the mini-skill(s) that you do not SEE immediately in the mate-in-1 problems that you could train with a targeted problem set?
There are 3 types of mate in 1:
Delete* the ones that I see immediately
* the ones that I see fast
* the ones that take me some time
I'm interested in the latter. What skill do I miss. Why do I need System 2 to solve these? What do I not SEE?
Answer: the mates in 1 with a lot of empty squares around the enemy king, where the aura of my pieces create the walls of the killbox are difficult to see.
I have done a lot of exercises in the past to try to learn to see the auras. But apparently none of these attempts worked. So I want to see whether the Trick does a better job.
The idea to get some clarity about the importance of mini-skills. Are they paramount or just additional?
ReplyDeleteFrom a logical point of view you can't build macro skills without mini skills. But how well first is a 1750 player in mini skills? How does it differ from a 2000 player?
ReplyDeleteThink of a mini-skill as a flashlight, shining on a small area of a path. The more flashlights you can bring to bear on the path as it twists and turns, the easier it is to find your way forward in the dark. I think of mini-skills being represented by a small circle of light around a simple concept.
DeleteFor example, the close Opposition and how it works in K+P vs K endgames covers a distinctly small area of pawn endgames. The Square of the Pawn is another such mini-skill. [Trébuchet? Another mini-skill, but don’t get me started on THAT! ;-)] The Fox in the Chicken Coop is a collection of mini-skills. A collection of mini-skills related to a particular concept would constitute a macro-skill.
IMHO, a 1750-level player would have a smaller, more disparate collection of mini-skills than a 2000-level player. The difference in the set of mini-skills between 1750-level players is large, whereas the set of mini-skills would be more uniform between 2000-level players—but not always. The 1750-level mini-skills would be close to a proper subset of the 2000-level player’s mini-skills. In short, the higher the rating, the more overlap there will be in the set of acquired mini-skills between any two players at that rating level. As skill level increases, the set of common mini-skills tends to converge.
That all sounds very logical. The question is, how complete is the mini skill set of a super grandmaster. Is it close to 100%? Or can he come away with 35% because the average grandmaster has only about 30%?
DeleteGiven the fact that there are two enormous holes in chess theory, and nobody has even the inclination to plug them, it might well be closer to 35%.
The two gaping holes: My System of Nimzowitsch and the Art of Attack in Chess of Vukovic are two bodies of theory where the authors only scratched the surface. Which they themselves were the first to admit.
In Secrets of Modern Chess Strategy: Advances Since Nimzowitsch it seems that GM John Watson throws in the towel because concrete calculation is enough to come away with in modern practice.
IDLE SPECULATION that sounds very logical
Delete[None of this will improve our capability to get to 2000 Elo.]
I used to think that super grandmasters (above 2700 Elo) were as close to perfection as possible for a human player. There are currently 10 GMs above 2700 Elo, and only one who has consistently stayed above 2800: Magnus Carlsen at 2823 (Standard), 2803 (Rapid), and 2860 (Blitz). Obviously, there must still be areas (very SMALL areas!) where their skills are relatively lacking. Otherwise, they would not be able to occasionally beat each other. Or, it could be that one of them was just having a “bad day” (maybe one got up on the wrong side of the bed), having nothing to do with their respective set of mini-skills, which I conjecture are as close to complete as is possible for humans. There cannot be significant gaps, because that is detectable from game scores and would be exploited as soon as it was found.
Now consider the number of chess programs that have far exceeded Magnus Carlsen’s rating. In 2026, the following engines are confirmed to be above 3400 Elo:
Stockfish 18 – 3651 Elo (undisputed leader, combining deep search with NNUE evaluation)
Reckless 0.9.0 – 3648 Elo (aggressive, high win rate)
PlentyChess 7.0.0 – 3645 Elo (balanced positional and tactical play)
Torch v4d – 3641 Elo (extremely stable, high draw rate)
Obsidian 16.0 – 3638 Elo (clean, accurate play)
Alexandria 9.0.0 – 3636 Elo (stable, balanced performance)
Other engines in the top tier (e.g., Komodo, Leela Chess Zero, Houdini, Fritz) also exceed 3400 Elo, though their exact CCRL ratings vary by list and test conditions.
CAVEATS:
1. Elo ratings for engines are not human Elo; they depend on the rating list, hardware, time control, and opening suite.
2. The “3400+” threshold is relatively low for top-tier engines; in 2026, most of the strongest public engines are in the 3600–3700 Elo range.
It is certainly questionable if any human is capable of reaching 3000 Elo, but records are made to be broken.
Even though engine ratings are NOT directly comparable to human ratings, I think it rather obvious (from human-computer matches) that the engines have reached a much higher skill level than humans, yet there is no way to directly identify the skill set possessed by the machines and compare it to human skill sets. Heck, we can't even identify the human skill set! Regardless, there is certainly an implication that the engines possess a much more robust and complete skill set than humans.
FWIW (not much), John Watson is an IM, not GM, and currently rated 2178 (Standard) Elo. He and GM Jacob Aagaard [2429 (Standard), 2502 (Rapid), 2483 (Blitz)] have an ongoing professional “disagreement” as to the validity of Watson’s confident assertion that modern chess is based on concrete calculation and is NOT rules-based (heuristics, laws, principles, whatever).
There are definitely exceptions to heuristics, simply because a heuristic is a generalized “rule of thumb,” not an iron-clad law.
"The exception that proves the rule" is another heuristic (“rule”). We humans simply can’t get along without them.
I have only had a deeper peek in the evaluation function of Stockfish. It is rather meager. The only way that it can beat humans (who make use of a much better evaluation function) is because it can look into the future. Not even the BEST future (that would require brute force), but a GOOD ENOUGH future.
DeleteWhich means that somewhere down the line, we humans make a mistake DESPITE being better at evaluation. For that very reason, an engine doesn't need a perfect evaluation function in order to beat humans. It only needs to look further in the future.
first should read versed.
ReplyDelete