- Blind mode tutorial
lichess.org
Donate

Looking at the Quality of Play in Chess 960

I think this is yet another fascinating topic for analysis. I only wish it went a bit deeper! It would help to compare the data from the chess formats on common axes, and perhaps perform some smoothing (e.g., moving averages) to better understand the trends that are present beyond the noise. The conclusion that play is less accurate in the opening in Chess960 almost goes without saying. By eye, I think there may be a far more interesting finding here, which is that the decrease in accuracy persists beyond what one would consider the opening phase. This would make sense, as players are not only playing without "book" (i.e., memorized lines), but also potentially playing without the typical middlegame plan templates that one would normally have in an opening system that one understands deeply. You could potentially analyze this not just by move number but also by other classification methods for which stage of the game one is in (e.g., based on piece development or amount of material remaining on the board).

This then gets to the question of how much transferrance is there of typical plans/structures in the middlegame from standard chess to Chess960. In certainly cases, a Chess960 game can certainly produce quite typical-looking structures once the opening has finished, but this does not always seem to be the case. I would therefore expect even GMs to potentially suffer in terms of accuracy due to less prior experience and pattern recognition within the types of middlegames that arise. A further interesting question would be whether such (hypothesized) decreases in accuracy might persist even into the ending! Probably there is not enough of a sample size here to yet answer that question though.

I think this is yet another fascinating topic for analysis. I only wish it went a bit deeper! It would help to compare the data from the chess formats on common axes, and perhaps perform some smoothing (e.g., moving averages) to better understand the trends that are present beyond the noise. The conclusion that play is less accurate in the opening in Chess960 almost goes without saying. By eye, I think there may be a far more interesting finding here, which is that the decrease in accuracy persists beyond what one would consider the opening phase. This would make sense, as players are not only playing without "book" (i.e., memorized lines), but also potentially playing without the typical middlegame plan templates that one would normally have in an opening system that one understands deeply. You could potentially analyze this not just by move number but also by other classification methods for which stage of the game one is in (e.g., based on piece development or amount of material remaining on the board). This then gets to the question of how much transferrance is there of typical plans/structures in the middlegame from standard chess to Chess960. In certainly cases, a Chess960 game can certainly produce quite typical-looking structures once the opening has finished, but this does not always seem to be the case. I would therefore expect even GMs to potentially suffer in terms of accuracy due to less prior experience and pattern recognition within the types of middlegames that arise. A further interesting question would be whether such (hypothesized) decreases in accuracy might persist even into the ending! Probably there is not enough of a sample size here to yet answer that question though.

If you want us to compare graphs, then set the scales to be the same.

If you want us to compare graphs, then set the scales to be the same.

@whatthefat said in #3:

I think this is yet another fascinating topic for analysis. I only wish it went a bit deeper! It would help to compare the data from the chess formats on common axes, and perhaps perform some smoothing (e.g., moving averages) to better understand the trends that are present beyond the noise. The conclusion that play is less accurate in the opening in Chess960 almost goes without saying. By eye, I think there may be a far more interesting finding here, which is that the decrease in accuracy persists beyond what one would consider the opening phase. This would make sense, as players are not only playing without "book" (i.e., memorized lines), but also potentially playing without the typical middlegame plan templates that one would normally have in an opening system that one understands deeply. You could potentially analyze this not just by move number but also by other classification methods for which stage of the game one is in (e.g., based on piece development or amount of material remaining on the board).

This then gets to the question of how much transferrance is there of typical plans/structures in the middlegame from standard chess to Chess960. In certainly cases, a Chess960 game can certainly produce quite typical-looking structures once the opening has finished, but this does not always seem to be the case. I would therefore expect even GMs to potentially suffer in terms of accuracy due to less prior experience and pattern recognition within the types of middlegames that arise. A further interesting question would be whether such (hypothesized) decreases in accuracy might persist even into the ending! Probably there is not enough of a sample size here to yet answer that question though.

That's an interesting point and intuitively it would make sense when people play worse absent of typical middlegame plans and ideas.
Comparing the quality of play between the classical and 960 games I used is probably not straightforward as there may be differences in the ratings of the players (not many 2700+ players were playing in Grenke) and also the games at the Grenke tournament were played with 2 rounds per day which probably makes them less accurate than games from tournaments with 1 round per day.
But this is certainly something to look into when more big 960 tournaments get played.

@whatthefat said in #3: > I think this is yet another fascinating topic for analysis. I only wish it went a bit deeper! It would help to compare the data from the chess formats on common axes, and perhaps perform some smoothing (e.g., moving averages) to better understand the trends that are present beyond the noise. The conclusion that play is less accurate in the opening in Chess960 almost goes without saying. By eye, I think there may be a far more interesting finding here, which is that the decrease in accuracy persists beyond what one would consider the opening phase. This would make sense, as players are not only playing without "book" (i.e., memorized lines), but also potentially playing without the typical middlegame plan templates that one would normally have in an opening system that one understands deeply. You could potentially analyze this not just by move number but also by other classification methods for which stage of the game one is in (e.g., based on piece development or amount of material remaining on the board). > > This then gets to the question of how much transferrance is there of typical plans/structures in the middlegame from standard chess to Chess960. In certainly cases, a Chess960 game can certainly produce quite typical-looking structures once the opening has finished, but this does not always seem to be the case. I would therefore expect even GMs to potentially suffer in terms of accuracy due to less prior experience and pattern recognition within the types of middlegames that arise. A further interesting question would be whether such (hypothesized) decreases in accuracy might persist even into the ending! Probably there is not enough of a sample size here to yet answer that question though. That's an interesting point and intuitively it would make sense when people play worse absent of typical middlegame plans and ideas. Comparing the quality of play between the classical and 960 games I used is probably not straightforward as there may be differences in the ratings of the players (not many 2700+ players were playing in Grenke) and also the games at the Grenke tournament were played with 2 rounds per day which probably makes them less accurate than games from tournaments with 1 round per day. But this is certainly something to look into when more big 960 tournaments get played.

so how much of the plans from "opening" theory transder to 960 non standard middlegames..

it amounts to how much of the opening ideas are determining of the middle game features in standard itself.

I am not yet learned well enough in chess to answer that. But I think one could use 960 data to keep studying the world of chess ideas not just accuracy by engine.

The geometric (and emergent consequences) of the board itself as the initial condition loses its "imprint" influence. Does the center object still matter in 960 middle games.. what of other "principles" how are they attributable to the initial conditions specifics and how deep can we keep making such relatin. that would be chess theory to me.. that is not "opening theory" (i.e. of the dept first study program and heavy imitation learning tenet).

so how much of the plans from "opening" theory transder to 960 non standard middlegames.. it amounts to how much of the opening ideas are determining of the middle game features in standard itself. I am not yet learned well enough in chess to answer that. But I think one could use 960 data to keep studying the world of chess ideas not just accuracy by engine. The geometric (and emergent consequences) of the board itself as the initial condition loses its "imprint" influence. Does the center object still matter in 960 middle games.. what of other "principles" how are they attributable to the initial conditions specifics and how deep can we keep making such relatin. that would be chess theory to me.. that is not "opening theory" (i.e. of the dept first study program and heavy imitation learning tenet).

The primary focus in chess should be on harmonious development, getting the pieces to work together. Early pawn moves should aim to increase mobility for your pieces while limiting the opponent's options. Central control is a key principle, but it’s also important to recognise that the timing of establishing that control can vary based on the situation and the opponent's moves.

Opportunities often arise when your opponent deviates from solid play. These unconventional moves can present tactical advantages if you can identify the weaknesses they create. As the game transitions to the middle game, complexity increases, especially with asymmetrical pawn structures and the positioning of material. This is where tactics play a central role in your strategic plans.

The central tensions are dynamic, and as the middle game unfolds, the board opens up, increasing entropy for both attack and defence. It’s crucial to maintain a clear horizon, especially as the game nears the endgame. Each phase of exchanges sharpens the focus on that horizon. Ultimately, in the endgame, your primary concern should be the opponent’s king, reducing the entropy in that critical area to deliver checkmate.

What matters most in chess is the ability to read the position and anticipate the next move, much like predicting the weather by observing the horizon. It's about recognising patterns and understanding how the game will unfold, enabling you to make the right moves at the right time.

The primary focus in chess should be on harmonious development, getting the pieces to work together. Early pawn moves should aim to increase mobility for your pieces while limiting the opponent's options. Central control is a key principle, but it’s also important to recognise that the timing of establishing that control can vary based on the situation and the opponent's moves. Opportunities often arise when your opponent deviates from solid play. These unconventional moves can present tactical advantages if you can identify the weaknesses they create. As the game transitions to the middle game, complexity increases, especially with asymmetrical pawn structures and the positioning of material. This is where tactics play a central role in your strategic plans. The central tensions are dynamic, and as the middle game unfolds, the board opens up, increasing entropy for both attack and defence. It’s crucial to maintain a clear horizon, especially as the game nears the endgame. Each phase of exchanges sharpens the focus on that horizon. Ultimately, in the endgame, your primary concern should be the opponent’s king, reducing the entropy in that critical area to deliver checkmate. What matters most in chess is the ability to read the position and anticipate the next move, much like predicting the weather by observing the horizon. It's about recognising patterns and understanding how the game will unfold, enabling you to make the right moves at the right time.

@Toscani
how can data be further making sense of this post content? which seems to apply to all chess 960 initial conditions, as they do have a king as ulitmate mobility target to control ability to further abide by obligation to move. (if not a lot of 2 person competitive "zero-sum" games). This seems like very agreable descriptive commenting but I wonder how we can further develop and make hypotheses and then use experimental data to shed light on such hypotheses.

Also "complexity" as single word needs development. I have seen it used in many fields to the point where alone, I find it a fill in the blanks word... so is the fate of word going very widespread usage.. they need more qualifiers to become useful in discussions.

Also about middle game plan idea in 960. we should recall that castling rules there kind of make the propable geometric discrimant like kingside and queen side.. still converging to standard chess.. the target of castling brings back a lot of the standard..

we would need more than one dimension score at the base of each tool of data exploration though.. we should be able to assigne quantitative metrics to board content features that human could eventually learn to see.. and SF single number oracle score, even if distributed over modes within games (which might give the impression of more metric "discriminant" power) is not telling of the foresight complexity. (well, some engine sharing more dimensions do exist, but even those are only about oracling and the board features correlation or learnable or quantifiable are absent). so we can do like in quant finance and built castles of metrics trying to extract juice from a trickle of information by poor bottom metrics and build some theoretical castle, but it seems to be something to be revisiting from bottom often... as it might lead us into overcondidence in the math. consequences based on too much faith in the bottom information feeder).

@Toscani how can data be further making sense of this post content? which seems to apply to all chess 960 initial conditions, as they do have a king as ulitmate mobility target to control ability to further abide by obligation to move. (if not a lot of 2 person competitive "zero-sum" games). This seems like very agreable descriptive commenting but I wonder how we can further develop and make hypotheses and then use experimental data to shed light on such hypotheses. Also "complexity" as single word needs development. I have seen it used in many fields to the point where alone, I find it a fill in the blanks word... so is the fate of word going very widespread usage.. they need more qualifiers to become useful in discussions. Also about middle game plan idea in 960. we should recall that castling rules there kind of make the propable geometric discrimant like kingside and queen side.. still converging to standard chess.. the target of castling brings back a lot of the standard.. we would need more than one dimension score at the base of each tool of data exploration though.. we should be able to assigne quantitative metrics to board content features that human could eventually learn to see.. and SF single number oracle score, even if distributed over modes within games (which might give the impression of more metric "discriminant" power) is not telling of the foresight complexity. (well, some engine sharing more dimensions do exist, but even those are only about oracling and the board features correlation or learnable or quantifiable are absent). so we can do like in quant finance and built castles of metrics trying to extract juice from a trickle of information by poor bottom metrics and build some theoretical castle, but it seems to be something to be revisiting from bottom often... as it might lead us into overcondidence in the math. consequences based on too much faith in the bottom information feeder).

Complexity in Chess can be measured using chess engines. It probably comes from a composite of multiple factors:
Each could have weights and than combined to produce a final complexity percentage.

Number of Legal Moves: Count the legal moves using a perft command, which shows how many possible moves a player can make. More legal moves suggest higher complexity.

Tactical Threats/Possibilities: Use engine analysis to check each legal move and compare the tactical threats or opportunities they present. Positions with multiple threats are more complex.

Strategic Uncertainty (PV Line Length): Examine the length and number of significant variations in engine analysis (PV lines). More branching in the PV indicates greater strategic uncertainty and complexity.

Pieces' Common Moves: Use browser extensions that show the PV moves in different colors. Observe the most common moves for pieces across variations, which can give insight into the complexity of the position.

King Safety: Evaluate king safety through static engine evaluations. A vulnerable king increases complexity, as it requires careful planning and defense.

Pawn Structure Complexity: Perform a static evaluation, then remove pawns one at a time to see how the evaluation changes. This helps determine which pawns are most critical and contribute to the complexity of the position.

I use Lucas Chess to analyse a chess game and I look at the analysis index. It has these terms:

Result of analysis
Average lost scores
Domination
Complexity
Efficient mobility
Narrowness
Pieces activity
Exchange tendency
Accuracy

They are the same terms whether its 960 or standard.

Chess with 64 squares with the same pieces whether mixed all over the chessboard or sorted or not in the first 2 rows is still chess. Stop imagining their is a phase a tree or any other terms in chess and just move pieces according to your own ideas or plans. Place the pieces where you want to place them. If your goal is to grab all the pieces and place them in the first row then go for it. If it's to capture a rook no mater the cost, than go for it. If chess is a clock race, than go for that aim. Chess is what you make of it.

The quality is the quality of your plan. If you discover there was a better way of reaching your goal, than the quality of your plan increased.

Not every word needs to be confusing or complex. A sentence needs more than nouns and verbs. The rest gives value percentage size an feels or more or less in the sentence. It can sound robotic or polite. It mixes in with your feelings and understanding of the words. So words are a mood swing in the direction you let them affect you. Chess is the same thing.

Complexity in Chess can be measured using chess engines. It probably comes from a composite of multiple factors: Each could have weights and than combined to produce a final complexity percentage. Number of Legal Moves: Count the legal moves using a perft command, which shows how many possible moves a player can make. More legal moves suggest higher complexity. Tactical Threats/Possibilities: Use engine analysis to check each legal move and compare the tactical threats or opportunities they present. Positions with multiple threats are more complex. Strategic Uncertainty (PV Line Length): Examine the length and number of significant variations in engine analysis (PV lines). More branching in the PV indicates greater strategic uncertainty and complexity. Pieces' Common Moves: Use browser extensions that show the PV moves in different colors. Observe the most common moves for pieces across variations, which can give insight into the complexity of the position. King Safety: Evaluate king safety through static engine evaluations. A vulnerable king increases complexity, as it requires careful planning and defense. Pawn Structure Complexity: Perform a static evaluation, then remove pawns one at a time to see how the evaluation changes. This helps determine which pawns are most critical and contribute to the complexity of the position. I use Lucas Chess to analyse a chess game and I look at the analysis index. It has these terms: Result of analysis Average lost scores Domination Complexity Efficient mobility Narrowness Pieces activity Exchange tendency Accuracy They are the same terms whether its 960 or standard. Chess with 64 squares with the same pieces whether mixed all over the chessboard or sorted or not in the first 2 rows is still chess. Stop imagining their is a phase a tree or any other terms in chess and just move pieces according to your own ideas or plans. Place the pieces where you want to place them. If your goal is to grab all the pieces and place them in the first row then go for it. If it's to capture a rook no mater the cost, than go for it. If chess is a clock race, than go for that aim. Chess is what you make of it. The quality is the quality of your plan. If you discover there was a better way of reaching your goal, than the quality of your plan increased. Not every word needs to be confusing or complex. A sentence needs more than nouns and verbs. The rest gives value percentage size an feels or more or less in the sentence. It can sound robotic or polite. It mixes in with your feelings and understanding of the words. So words are a mood swing in the direction you let them affect you. Chess is the same thing.

I've been thinking about how we view the "quality of play" in chess. Chess960 randomises piece positions, but what if we took that idea even further? Instead of rigid opening theory, standard piece placement, what if chess was just about creating your own plan, placing pieces where you want on your side of the chessboard and setting goals that aren’t tied to tradition? Maybe it can be about capturing a specific piece, even the king, or just testing how long you can survive with a weird setup. The quality of play isn’t just about following established engine best shortest paths or patterns. It’s about creativity, adaptability, and defining your own challenge. The game becomes less about winning or losing or racing and more about individual expression, risk-taking, and exploring new ways to approach the game. The middle game is already there when you moved the pieces to the middle game or maybe the middle game needs to have lost some material to cal it a middle game. Chess does not have to have all that structural complexity or competition. It can be played even more than freestyle.

I've been thinking about how we view the "quality of play" in chess. Chess960 randomises piece positions, but what if we took that idea even further? Instead of rigid opening theory, standard piece placement, what if chess was just about creating your own plan, placing pieces where you want on your side of the chessboard and setting goals that aren’t tied to tradition? Maybe it can be about capturing a specific piece, even the king, or just testing how long you can survive with a weird setup. The quality of play isn’t just about following established engine best shortest paths or patterns. It’s about creativity, adaptability, and defining your own challenge. The game becomes less about winning or losing or racing and more about individual expression, risk-taking, and exploring new ways to approach the game. The middle game is already there when you moved the pieces to the middle game or maybe the middle game needs to have lost some material to cal it a middle game. Chess does not have to have all that structural complexity or competition. It can be played even more than freestyle.