lichess.org
Donate

The Shape of Time: How Openings Evolve from Blitz to Classical

The pivot point is 1500, because a player that becomes member will either drop in rating or increase in rating. So to have a true graph, number of games played at a particular rating would be important info. If a play progresses through the ratings they don't have many games in a particular rating. Since ratings have to be correlated with the time control that the speed of moving the pieces might be the common factor that would be better than saying rated 1500 but if the moves are played slowly compared to fast that it's not the same rating category. One might be blitz while the other bullet. Average speed of moves payed should replace the category names. If you premove everything than you created an ultrabullet game even if the rating is placed in the standard time control given games. Try to use the ratio between rating and average move time and see how it evolves.

The pivot point is 1500, because a player that becomes member will either drop in rating or increase in rating. So to have a true graph, number of games played at a particular rating would be important info. If a play progresses through the ratings they don't have many games in a particular rating. Since ratings have to be correlated with the time control that the speed of moving the pieces might be the common factor that would be better than saying rated 1500 but if the moves are played slowly compared to fast that it's not the same rating category. One might be blitz while the other bullet. Average speed of moves payed should replace the category names. If you premove everything than you created an ultrabullet game even if the rating is placed in the standard time control given games. Try to use the ratio between rating and average move time and see how it evolves.

@Toscani said in #2:

The pivot point is 1500, because a player that becomes member will either drop in rating or increase in rating. So to have a true graph, number of games played at a particular rating would be important info. If a play progresses through the ratings they don't have many games in a particular rating. Since ratings have to be correlated with the time control that the speed of moving the pieces might be the common factor that would be better than saying rated 1500 but if the moves are played slowly compared to fast that it's not the same rating category. One might be blitz while the other bullet. Average speed of moves payed should replace the category names. If you premove everything than you created an ultrabullet game even if the rating is placed in the standard time control given games. Try to use the ratio between rating and average move time and see how it evolves.

Your idea is interesting but this would require a very different software approach. I'm currently using the opening API so I get access to broad statistics and little filtering choices. Your idea (like ideas 1 and 2 of the "Next steps" section) requires scanning very large numbers of games to recreate the stats. But it would be interesting to see

@Toscani said in #2: > The pivot point is 1500, because a player that becomes member will either drop in rating or increase in rating. So to have a true graph, number of games played at a particular rating would be important info. If a play progresses through the ratings they don't have many games in a particular rating. Since ratings have to be correlated with the time control that the speed of moving the pieces might be the common factor that would be better than saying rated 1500 but if the moves are played slowly compared to fast that it's not the same rating category. One might be blitz while the other bullet. Average speed of moves payed should replace the category names. If you premove everything than you created an ultrabullet game even if the rating is placed in the standard time control given games. Try to use the ratio between rating and average move time and see how it evolves. Your idea is interesting but this would require a very different software approach. I'm currently using the opening API so I get access to broad statistics and little filtering choices. Your idea (like ideas 1 and 2 of the "Next steps" section) requires scanning very large numbers of games to recreate the stats. But it would be interesting to see

Very interesting. It shows numerically what is the "rating" of the opening where the peak is, or maybe the average rating of a player to achieve a level of mastery for it.

Very interesting. It shows numerically what is the "rating" of the opening where the peak is, or maybe the average rating of a player to achieve a level of mastery for it.

"I removed the 2500+ Elo bracket because it had a low amount of games (especially in classical) and the values often were non-significant statistically and extreme in value." * You can also add a 3000+ bracket long time control: ICCF. All main openings are draws. All bad openings are losses.

"a game between stronger players with less time should be somewhat similar to a game between weaker players with more time" * This makes sense.

"take a performance curve from Blitz, slide it to the left, and have it roughly match the curve for Rapid. Slide it again, and it would match Classical." * It is the other way around. A blitz game by strong players is like a rapid game between medium players and like a classical game between weak players.

"if an opening is good in practice, it gets better with more time. If it's bad, it gets worse."

  • That is logical. Bongcloud and Englund Gambit are playable in bullet, but not in classical.
    Ruy Lopez is playable in any time control and at all levels.

"It seems that from an opening position, increasing the time on the clock tends to increase the practical advantage of the player who is already better off." * It is the other way around. A bad position may be defendable with lots of time, but not in a fast time control. An early deviation with a bad opening can be thought of as a gambit: you sacrifice objective position to put pressure on the clock. You bet your opponent cannot refute it within the time limit.

"If a position holds a genuine, objective advantage, more time allows a player to calculate more precisely and maximize it"

  • It is the other way around. A objectively disadvantageous position cannot be refuted at a fast time control.

"In Blitz, once a player has an advantage, the defender doesn't have enough time to dismantle it."

  • It is the other way around. In blitz both players make so many mistakes, that the objective disadvantage after the first moves plays no significant role.

"converting a practical advantage is fundamentally harder (requires more time) than stabilizing the game?"

  • It is hard to refute 1 g4 or 1 d4 e5 over the board at fast time control.

"We know that draw rates increase significantly at higher levels" * Yes, near 100% in ICCF correspondence.
That is both due to increased accuracy and selection of sound opening lines.

"These findings are about the nature of humans playing chess, not necessarily about the objective nature of the game itself."

  • The objective nature of the game itself shows in ICCF correspondence and in TCEC engine versus engine.
    ICCF is now nearly 100% draw. TCEC would be 100% draw if there were no imposed openings. The bad openings get busted in both games engines play with both colors.

"the Experience Factor of Blitz players is the single most important explanation for these effects" * There are no blitz players. Most chess players play all time controls. Experience from one time control transfers to the other.

"A bad performance might be due to a subtle strategic weakness in the opening, or simply a player blundering into a well-known trap." * It is the other way around. Any subtle strategic weakness can only be exploited by very strong players at long time controls. In any game between weaker players or faster time controls both players make so many mistakes that the objective weakness of an opening plays no role.

"Analyze a large sample of games for a given opening." * This has already been done, look at ICCF or TCEC.

"how time affects play, independent of preparation" * You cannot disconnect both.

"We could then plot performance not just against Elo, but against an Experience Score for that line."

  • Experience may also stem from playing the same with the other color, or from experience in over the board classical play.

"a complex middle-game position that is reached a significant number of times through different transpositions"

  • You should track transpositions anyway. 1 e4 e6 2 d4 is exactly the same as 1 d4 e6 2 e4

"while reducing the impact of specific opening theory"

  • Opening theory does not stop at the last move of some line. Plans and ideas of how to play, where pieces belong, what to do and avoid stems from experience in games and from study of annotated grandmaster games.

"If anyone has a suggestion for a suitable FEN string for such a position, I would love to test it."

  • The Berlin endgame maybe
    r1bk1b1r/ppp2ppp/2p5/4Pn2/8/5N2/PPP2PPP/RNB2RK1 w - - 0 9
"I removed the 2500+ Elo bracket because it had a low amount of games (especially in classical) and the values often were non-significant statistically and extreme in value." * You can also add a 3000+ bracket long time control: ICCF. All main openings are draws. All bad openings are losses. "a game between stronger players with less time should be somewhat similar to a game between weaker players with more time" * This makes sense. "take a performance curve from Blitz, slide it to the left, and have it roughly match the curve for Rapid. Slide it again, and it would match Classical." * It is the other way around. A blitz game by strong players is like a rapid game between medium players and like a classical game between weak players. "if an opening is good in practice, it gets better with more time. If it's bad, it gets worse." * That is logical. Bongcloud and Englund Gambit are playable in bullet, but not in classical. Ruy Lopez is playable in any time control and at all levels. "It seems that from an opening position, increasing the time on the clock tends to increase the practical advantage of the player who is already better off." * It is the other way around. A bad position may be defendable with lots of time, but not in a fast time control. An early deviation with a bad opening can be thought of as a gambit: you sacrifice objective position to put pressure on the clock. You bet your opponent cannot refute it within the time limit. "If a position holds a genuine, objective advantage, more time allows a player to calculate more precisely and maximize it" * It is the other way around. A objectively disadvantageous position cannot be refuted at a fast time control. "In Blitz, once a player has an advantage, the defender doesn't have enough time to dismantle it." * It is the other way around. In blitz both players make so many mistakes, that the objective disadvantage after the first moves plays no significant role. "converting a practical advantage is fundamentally harder (requires more time) than stabilizing the game?" * It is hard to refute 1 g4 or 1 d4 e5 over the board at fast time control. "We know that draw rates increase significantly at higher levels" * Yes, near 100% in ICCF correspondence. That is both due to increased accuracy and selection of sound opening lines. "These findings are about the nature of humans playing chess, not necessarily about the objective nature of the game itself." * The objective nature of the game itself shows in ICCF correspondence and in TCEC engine versus engine. ICCF is now nearly 100% draw. TCEC would be 100% draw if there were no imposed openings. The bad openings get busted in both games engines play with both colors. "the Experience Factor of Blitz players is the single most important explanation for these effects" * There are no blitz players. Most chess players play all time controls. Experience from one time control transfers to the other. "A bad performance might be due to a subtle strategic weakness in the opening, or simply a player blundering into a well-known trap." * It is the other way around. Any subtle strategic weakness can only be exploited by very strong players at long time controls. In any game between weaker players or faster time controls both players make so many mistakes that the objective weakness of an opening plays no role. "Analyze a large sample of games for a given opening." * This has already been done, look at ICCF or TCEC. "how time affects play, independent of preparation" * You cannot disconnect both. "We could then plot performance not just against Elo, but against an Experience Score for that line." * Experience may also stem from playing the same with the other color, or from experience in over the board classical play. "a complex middle-game position that is reached a significant number of times through different transpositions" * You should track transpositions anyway. 1 e4 e6 2 d4 is exactly the same as 1 d4 e6 2 e4 "while reducing the impact of specific opening theory" * Opening theory does not stop at the last move of some line. Plans and ideas of how to play, where pieces belong, what to do and avoid stems from experience in games and from study of annotated grandmaster games. "If anyone has a suggestion for a suitable FEN string for such a position, I would love to test it." * The Berlin endgame maybe r1bk1b1r/ppp2ppp/2p5/4Pn2/8/5N2/PPP2PPP/RNB2RK1 w - - 0 9

Can't comment on higher ELOs, but I don't think lower lichess elos are directly comparable between blitz and classical. Anyone can play on a slow clock, blitz (and especially bullet) is tough. The ratings are different. Opening theory is particularly relevant. There are plenty of trap positions in theory that are difficult to solve for the first time over the board with limited time. Some of them are quite easy to get.

Can't comment on higher ELOs, but I don't think lower lichess elos are directly comparable between blitz and classical. Anyone can play on a slow clock, blitz (and especially bullet) is tough. The ratings are different. Opening theory is particularly relevant. There are plenty of trap positions in theory that are difficult to solve for the first time over the board with limited time. Some of them are quite easy to get.

It is known that a large part of the low ELO on lichess or chess.com is due to a large number of cheaters who have nothing else to do in life other than bother those who want to play for real, but anyway, that's part of it, right?

It is known that a large part of the low ELO on lichess or chess.com is due to a large number of cheaters who have nothing else to do in life other than bother those who want to play for real, but anyway, that's part of it, right?

Did you test it out on your lichess insights (Opening repertoire)?

https://lichess.org/insights/Toscani/result/openingFamily/variant:blitz/color:white/phase:1
https://lichess.org/training/openings
https://github.com/RemiFabre/WickedLines/blob/main/README.md

Knowing what works for others is great to know, but what is most important to me is knowing what works for me.
My opening repertoire is in the lichess link. It shows white pieces only, the generalized opening family names, and since we are talking about openings, I press phase 1 to remove the rest of the phases. I pressed results to see which opening I win the most with and the ones I lose the most with.

Openings I played the most in rated games:
Sicilian Defense
English Opening
Queen's Pawn Game
Zukertort Opening
French Defense
King's Pawn Game
Scandinavian Defense
Philidor Defense
Petrov's Defense
Indian Defense
Caro-Kann Defense
Benoni Defense

How can the WickedLines help me more that what I see in my lichess insights?

Did you test it out on your lichess insights (Opening repertoire)? https://lichess.org/insights/Toscani/result/openingFamily/variant:blitz/color:white/phase:1 https://lichess.org/training/openings https://github.com/RemiFabre/WickedLines/blob/main/README.md Knowing what works for others is great to know, but what is most important to me is knowing what works for me. My opening repertoire is in the lichess link. It shows white pieces only, the generalized opening family names, and since we are talking about openings, I press phase 1 to remove the rest of the phases. I pressed results to see which opening I win the most with and the ones I lose the most with. Openings I played the most in rated games: Sicilian Defense English Opening Queen's Pawn Game Zukertort Opening French Defense King's Pawn Game Scandinavian Defense Philidor Defense Petrov's Defense Indian Defense Caro-Kann Defense Benoni Defense How can the WickedLines help me more that what I see in my lichess insights?

I also looked at rating increase and I copy pasted the results all into a spreadsheet and sorted by opening name.
https://lichess.org/insights/Toscani/ratingDiff/openingFamily/variant:bullet/color:white/phase:1
https://lichess.org/insights/Toscani/ratingDiff/openingFamily/variant:blitz/color:white/phase:1
https://lichess.org/insights/Toscani/ratingDiff/openingFamily/variant:rapid/color:white/phase:1
https://lichess.org/insights/Toscani/ratingDiff/openingFamily/variant:classical/color:white/phase:1

I discovered that only the Caro-Kann Defense made me increase in rating in all time categories.
Opening Name, Rating gain, Quantity, Time
Caro-Kann Defense 0.73 11 Bullet
Caro-Kann Defense 0.61 109 Blitz
Caro-Kann Defense 0.82 33 Rapid
Caro-Kann Defense 0.44 9 Classical

The other openings were good in one time category but not in another.

I also looked at rating increase and I copy pasted the results all into a spreadsheet and sorted by opening name. https://lichess.org/insights/Toscani/ratingDiff/openingFamily/variant:bullet/color:white/phase:1 https://lichess.org/insights/Toscani/ratingDiff/openingFamily/variant:blitz/color:white/phase:1 https://lichess.org/insights/Toscani/ratingDiff/openingFamily/variant:rapid/color:white/phase:1 https://lichess.org/insights/Toscani/ratingDiff/openingFamily/variant:classical/color:white/phase:1 I discovered that only the Caro-Kann Defense made me increase in rating in all time categories. Opening Name, Rating gain, Quantity, Time Caro-Kann Defense 0.73 11 Bullet Caro-Kann Defense 0.61 109 Blitz Caro-Kann Defense 0.82 33 Rapid Caro-Kann Defense 0.44 9 Classical The other openings were good in one time category but not in another.

I tried out the fantastic script and by the way the blog was great.
The steps I took to start the script. I already had python installed.

  1. Downloaded the script;
    https://github.com/RemiFabre/WickedLines/tree/main?tab=readme-ov-file
  2. Moved the script to a new directory;
  3. Started my virtual environment in that new directory: pipenv shell

Ran the script and discovered the requirements I needed.

  1. Installed the script requirements I needed: pip install python-chess requests scipy tabulate matplotlib

I did not find the requirements file to run the script. I actually installed mode, which I uninstalled once I realized it was not a requirement.

  1. AI told me this usage: wickedlines.py [-h] {line,hunt,plot,compare,batchplot}
  2. I ran the following modes:
    python3 wickedlines.py line e4 e5
    python3 wickedlines.py plot e4 e5
    python3 wickedlines.py compare e4 d4
    python wickedlines.py batchplot --speeds blitz
    python3 wickedlines.py hunt -h

I think I'll test my present opening repertoire in the script to see the results.
I have a local database of my games. I wonder if the script could use a local databases instead of going on lichess.
The batchplot -- speeds blitz caused a message: [INFO] Rate limit hit. Waiting 60s...

This could be avoided if we use a database on our own local drive.
https://encroissant.org/

Thanks again for the blog and the links. @LKama

I tried out the fantastic script and by the way the blog was great. The steps I took to start the script. I already had python installed. 1) Downloaded the script; https://github.com/RemiFabre/WickedLines/tree/main?tab=readme-ov-file 2) Moved the script to a new directory; 3) Started my virtual environment in that new directory: pipenv shell Ran the script and discovered the requirements I needed. 4) Installed the script requirements I needed: pip install python-chess requests scipy tabulate matplotlib I did not find the requirements file to run the script. I actually installed mode, which I uninstalled once I realized it was not a requirement. 6) AI told me this usage: wickedlines.py [-h] {line,hunt,plot,compare,batchplot} 7) I ran the following modes: python3 wickedlines.py line e4 e5 python3 wickedlines.py plot e4 e5 python3 wickedlines.py compare e4 d4 python wickedlines.py batchplot --speeds blitz python3 wickedlines.py hunt -h I think I'll test my present opening repertoire in the script to see the results. I have a local database of my games. I wonder if the script could use a local databases instead of going on lichess. The batchplot -- speeds blitz caused a message: [INFO] Rate limit hit. Waiting 60s... This could be avoided if we use a database on our own local drive. https://encroissant.org/ Thanks again for the blog and the links. @LKama