drummyfish/less_retarded_wiki
5
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Update

Browse source
This commit is contained in:
Miloslav Ciz 2024-11-13 01:37:31 +01:00
parent db7839946f
commit da00222f76
31 changed files with 2180 additions and 1926 deletions
Download patch file Download diff file Expand all files Collapse all files

213
chess.md
View file

@ -100,7 +100,7 @@ So secondly we need to implement a so called **search** algorithm -- typically s
Exhaustively searching the tree to great depths is not possible even with most powerful [hardware](hw.md) due to astronomical numbers of possible move combinations, so the engine has to limit the depth quite greatly and use various [hacks](hacking.md), [approximations](approximation.md), [heuristics](heuristic.md), [caches](cache.md) etc.. Normally it will search all moves to a small depth (e.g. 2 or 3 half moves or *plys*) and then extend the search for interesting moves such as exchanges or checks. Maybe the greatest danger of searching algorithms is so called **horizon effect** which has to be addressed somehow (e.g. by detecting quiet positions, so called *quiescence*). If not addressed, the horizon effect will make an engine misevaluate certain moves by stopping the evaluation at certain depth even if the played out situation would continue and lead to a vastly different result (imagine e.g. a queen taking a pawn which is guarded by another pawn; if the engine stops evaluating after the queen's pawn capture, it will think it's a won pawn, when in fact it's a lost queen). There are also many techniques for reducing the number of searched tree nodes and speeding up the search, for example pruning methods such as **alpha-beta** (which subsequently works best with correctly ordering moves to search), or **transposition tables** (remembering already evaluated position so that they don't have to be evaluated again when encountered by a different path in the tree). Furthermore we may try to combine many different things together, for example exhaustive search for some situations along with monte carlo in others; we may also try to employ more machine learning, e.g. make a special neural net just for suggesting which moves and to what depth should be searched etc.
**Alternative approaches**: most engines work as described above (search plus evaluation function) with some minor or bigger modifications. The simplest possible stupid AI can just make random moves, which will of course be an extremely weak opponent (though even weaker can be made, but these will actually require more complex code as to play worse than random moves requires some understanding and searching for the worst moves) -- one might perhaps try to just program a few simple rules to make it a bit less stupid and possibly a simple training opponent for complete beginners: the AI may for example pick a few "good looking" candidate moves that are "usually OK" (pushing a pawn, taking a higher value piece, castling, ...) and aren't a complete insanity, then pick one at random only from those (this randomness can further be improved and gradually controlled by scoring the moves somehow and adding a more or less random value from some range to each score, then picking the moves with highest score). One could also try to just program in a few generic rules such as: checkmate if you can, otherwise take an unprotected piece, otherwise protect your own unprotected piece etc. -- this could produce some beginner level bot. Another idea might be a "Chinese room" bot that doesn't really understand chess but has a huge database of games (which it may even be fetching from some Internet database) and then just looking up what moves good players make in positions that arise on the board, however a database of all positions will never exist, so in case the position is not found there has to be some fallback (e.g. play random move, or somehow find the "most similar position" and use that, ...). As another approach one may try to use some **non neural network [machine learning](machine_learning.md)**, for example [genetic programming](genetic_programming.md), to train the evaluation function, which will then be used in the tree search. Another idea that's being tried (e.g. in the Maia engine) is **pure neural net AI** (or another form of machine learning) which doesn't use any tree search -- not using search at all has long been thought to be impossible as analyzing a chess position completely statically without any "looking ahead" is extremely difficult, however new neural networks have shown to be extremely good at this kind of thing and pure NN AIs can now play on a master level (a human grandmaster playing ultra bullet is also just a no-calculation, pure pattern recognition play). Next, **[Monte Carlo](monte_carlo.md) tree search** (MCTS) is an alternative way of searching the game tree which may even work without any evaluation function: in it one makes many random playouts (complete games until the end making only random moves) for each checked move and based on the number of wins/losses/draws in those playouts statistically a value is assigned to the move -- the idea is that a move that most often leads to a win is likely the best. Another Monte Carlo approach may just make random playouts, stop at random depth and then use normal static evaluation function (horizon effect is a danger but hopefully its significance should get minimized in the averaging). However MCTS is pretty tricky to do well. MCTS is used e.g. in Komodo Dragon, the engine that's currently among the best. Another approach may lie in somehow using several methods and [heuristics](heuristic.md) to vote on which move would be best.
**Alternative approaches**: most engines work as described above (search plus evaluation function) with some minor or bigger modifications. The simplest possible stupid AI can just make random moves, which will of course be an extremely weak opponent (though even weaker can be made, but these will actually require more complex code as to play worse than random moves requires some understanding and searching for the worst moves) -- one might perhaps try to just program a few simple rules to make it a bit less stupid and possibly a simple training opponent for complete beginners: the AI may for example pick a few "good looking" candidate moves that are "usually OK" (pushing a pawn, taking a higher value piece, castling, ...) and aren't a complete insanity, then pick one at random only from those (this randomness can further be improved and gradually controlled by scoring the moves somehow and adding a more or less random value from some range to each score, then picking the moves with highest score). One could also try to just program in a few generic rules such as: checkmate if you can, otherwise take an unprotected piece, otherwise protect your own unprotected piece etc. -- this could produce some beginner level bot. Another idea might be a "Chinese room" bot that doesn't really understand chess but has a huge database of games (which it may even be fetching from some Internet database) and then just looking up what moves good players make in positions that arise on the board, however a database of all positions will never exist, so in case the position is not found there has to be some fallback (e.g. play random move, or somehow find the "most similar position" and use that, ...). As another approach one may try to use some **non neural network [machine learning](machine_learning.md)**, for example [genetic programming](genetic_programming.md), to train the evaluation function, which will then be used in the tree search. Another idea that's being tried (e.g. in the Maia engine) is **pure neural net AI** (or another form of machine learning) which doesn't use any tree search -- not using search at all has long been thought to be impossible as analyzing a chess position completely statically without any "looking ahead" is extremely difficult, however new neural networks have shown to be extremely good at this kind of thing and pure NN AIs can now play on a master level (a human grandmaster playing ultra bullet is also just a no-calculation, pure pattern recognition play) -- a paper called *Grandmaster-Level Chess Without Search* managed to implement pure NN engine that on Lichess achieved rating of 2895, close to the strongest engines on the site. Next, **[Monte Carlo](monte_carlo.md) tree search** (MCTS) is an alternative way of searching the game tree which may even work without any evaluation function: in it one makes many random playouts (complete games until the end making only random moves) for each checked move and based on the number of wins/losses/draws in those playouts statistically a value is assigned to the move -- the idea is that a move that most often leads to a win is likely the best. Another Monte Carlo approach may just make random playouts, stop at random depth and then use normal static evaluation function (horizon effect is a danger but hopefully its significance should get minimized in the averaging). However MCTS is pretty tricky to do well. MCTS is used e.g. in Komodo Dragon, the engine that's currently among the best. Another approach may lie in somehow using several methods and [heuristics](heuristic.md) to vote on which move would be best.
Many other aspects come into the AI design such as opening books (databases of best opening moves), endgame tablebases (precomputed databases of winning moves in simple endgames), clock management, pondering (thinking on opponent's move), learning from played games etc. For details see the above linked chess programming wiki.
@ -110,29 +110,29 @@ Many other aspects come into the AI design such as opening books (databases of b
Here are some notable chess engines/computers/entities, as of 2024:
- **[Stockfish](stockfish.md)** (SF): FOSS engine (written in [C++](cpp.md)), without any doubt **the strongest chess engine** that's been reliably winning all the computer tournaments for years now; its strength is far beyond any human, even if run on quite a weak device -- it actually caused some trouble because it's extremely easy to just download onto a cellphone and [cheat](cheating.md) even in OTB tournaments. Currently the engine is using a [neural network](neural_network.md) for evaluating positions but still also uses the tree search algorithm (a greatly optimized one so that it searches gigantic numbers of positions per second). Important part of the development is so called *Fishtest*, a distributed framework for testing and improving the engine's performance, it's one of the reasons why it good so strong. Stockfish's current CCRL Elo rating is 3639 (warning: this is incomparable to human Elos).
- **[Stockfish](stockfish.md)** (SF): [FOSS](foss.md) engine (written in [C++](cpp.md)), without any doubt **the strongest chess engine** that's been reliably winning all the computer tournaments for years now; its strength is far beyond any human, even if run on quite a weak device -- it actually caused some trouble because it's extremely easy to just download onto a cellphone and [cheat](cheating.md) even in OTB tournaments. Currently the engine is using a [neural network](neural_network.md) for evaluating positions but still also uses a search algorithm (a greatly optimized one so that it searches gigantic numbers of positions per second to very high depth) and offers using hand crafted evaluation as well. It's actually quite well written and only runs on the [CPU](cpu.md), it supports many different [architectures](isa.md) and is very [portable](portability.md), unlike for example LC0 which requires a bloated [GPU](gpu.md). Important part of Stockfish development is so called *Fishtest*, a distributed framework for testing and improving the engine's performance, it's one of the reasons why it good so strong. Stockfish's current CCRL Elo rating is 3639 (warning: this is incomparable to human Elos).
- **Magnus Carlsen**: Human, most likely the strongest player ever, has been quite comfortably winning every tournament he entered including the world championship until he quit, basically because he got "bored". His top FIDE Elo was 2882.
- **Komodo Dragon**: [Proprietary](proprietary.md), currently seems to be the second strongest engine, it's main feature is [Monte Carlo] ("randomized") search algorithm. Current CCRL Elo is 3624.
- **Komodo Dragon**: [Proprietary](proprietary.md), currently seems to be the second strongest engine, its main feature is [Monte Carlo] ("randomized") search algorithm. Current CCRL Elo is 3624.
- **Hikaru Nakamura**: Human, popularly considered "the second best" right after Magnus (who the mainstream likes to see as his biggest "rival") -- although this is not really so clear, he is definitely among the very top (his peak rating was 2816); We mention him here for his style of play: he likes to [troll](trolling.md) and do various lulz even on high level, many hate him for it. Some say he is a bit of a dick but his chess is definitely extremely entertaining. He also often plays strong computer engines, which is something that Magnus refuses to do (for commercial reasons Magnus only plays his shitty mobile "[app](app.md)"), so thanks to Nakamura we can really watch the best humans play the best computers.
- **[Leela Chess Zero](lc0.md)** (lc0): FOSS engine (written in C++), among top strongest engines (currently top 50 on CCRL), it is interesting mainly for how it works: it is a neural network engine that's **completely self-taught** from the ground up, i.e. it didn't learn chess by watching anyone else play, it was only allowed to learn by playing against itself. Current CCRL Elo is 3441.
- **[Deep Blue](deep_blue.md)**: A historically famous supercomputer, the first one to have beaten the human world chess champion in 1997.
- **[GNU chess](gnu_chess.md)** Free engine by [GNU](gnu.md), not among absolute top by strength but still very strong. Current CCRL Elo is 2825.
- **Maia**: FOSS engine, or rather neural network, notable by not trying to be the strongest, but rather most human-like, i.e. tries to imitate human play, even with errors. There are several versions, each trained for different strength. It is also notable by using pure neural network, i.e. it doesn't perform any search, it's a pure "pattern recognition"/static engine that still manages to play quite well.
- **Toledo Nanochess**: Seems to be the world's smallest [C](c.md) chess engine, with only 1257 non-blank characters of source code.
- **[smallchesslib](smallchesslib.md)/smolchess**: Tiny LRS [C](c.md) library/engine, very weak but is very simple, small and portable, may be [good enough](good_enough.md) in many situations.
- **[smallchesslib](smallchesslib.md)/smolchess**: Tiny [LRS](lrs.md) [C](c.md) library/engine, very weak but is quite simple, small and [portable](portability.md), may be [good enough](good_enough.md) in many situations.
- **Chessmaster**: A famous proprietary chess video games with its own engine, it was strong for a video game of its time (around 2000 Elo) but nowadays would be considered rather weak for an engine -- its significance is cultural, it's used for comparisons, many people played against it and still use it to test their engines against.
- **Turochamp**: Probably the first chess program ever, made by David Champernowne and [Alan Turing](turing.md) himself in 1948, in times when computers still couldn't execute it! It was very primitive, looking only two moves ahead, and was only ever executed manually -- of course, it got raped pretty bad the human opponent.
- ...
## Stats And Records
Chess stats are pretty [interesting](interesting.md). Thanks a lot e.g. to Lichess (and NOT thanks to fucking capitalist idiots like chess dot com) we have some great [public domain](public_domain.md) databases of billions of games played between both people and computers, and thanks to chess engines we can generate new and new on demand, so naturally many people create cool statistics, look for patterns and rarities. This can be very insightful and entertaining.
Chess stats are pretty [interesting](interesting.md). Thanks a lot e.g. to Lichess (and NOT thanks to fucking capitalist idiots like chess dot com) we have some great [public domain](public_domain.md) databases of billions of games played between both people and computers, and thanks to chess engines we can generate new and new on demand, so naturally many people create cool statistics, look for patterns and oddities. This can be very insightful and entertaining.
{ Some chess world records are here: https://timkr.home.xs4all.nl/records/records.htm. ~drummyfish }
**Number of possible games** is not known exactly, Shannon estimated it at 10^120 (lower bound, known as *Shannon number*). Number of possible games by plies played is 20 after 1, 400 after 2, 8902 after 3, 197281 after 4, 4865609 after 5, and 2015099950053364471960 after 15.
Similarly the **number of possibly reachable positions** (position for which so called *proof game* exists) is not known exactly, it is estimated to at least 10^40 and 10^50 at most. Numbers of possible positions by plies is 20 after 1, 400 after 2, 5362 after 3, 72078 after 4, 822518 after 5, and 726155461002 after 11.
Similarly the **number of possibly reachable positions** (position for which so called *proof game* exists) is not known exactly, some upper estimates have been made, lower bounds are much harder to set. The estimates are placed around 10^40 or 10^50 at most. [Here](https://tromp.github.io/chess/chess.html) is a site that gives a proven upper estimate of 45193640626062205213735739171550309047984050718 (2^155), also providing a more precise one of 7728772977965919677164873487685453137329736522 (~10^45.888, ~2^152) which was however proven with a program that's a bit obscure and less trustworthy. Numbers of possible positions by plies are 20 after 1, 400 after 2, 5362 after 3, 72078 after 4, 822518 after 5, and 726155461002 after 11.
**Shortest possible checkmate** is by black on ply number 4 (so called *fool's mate*); in fact there are 8 different games that can end like this. As of 2022 the **longest known forced checkmate** is in 549 moves -- it has been discovered when computing the Lomonosov Tablebases. EDIT: now it seems there is one in 584 moves. Please note this: there most likely exist much longer forced mates, these are just the KNOWN ones. Consider e.g. that if black blunders a queen in the opening, the game is very likely a theoretical win for white since then, i.e. a forced mate, and with perfect play black can probably resist for very long. However such situations are too complex to explore fully.
@ -156,29 +156,185 @@ What is **the most typical game**? We can try to construct such a game from a ga
22. Qb3 O-O-O 23. Qa3 Qc5 24. Qb3 d5 25. Bf1
```
What's **the best and worst opening move according to the engines?** Please remember that engine best move is not necessarily the best move for a human, see the note on perfect play above. Also the answer will depend on which engine (what evaluation function) you use and to what depth you search. According to Lichess cloud database (accessible via public API) that stores stockfish evaluations for various positions, e4 leads to the best evaluated position (18 centipawn, evaluated to depth 70; the move is also seen by many as best for humans), closely followed by d4 and Nf3 (both 17 centipawn, depth 47 and 56) and c4 (12 centipawn, depth 59). Worst move here is by far g4 with evaluation -96 centipawn (depth 52) -- almost a whole pawn, i.e. stockfish says that by playing this move you basically just throw away your pawn immediately. Another bad move is apparently f3 (-76, depth 40), Nh3 (-42), Na3 (-33), b4 or h4 (both -28). In general both humans and engines are basically deciding between e4 and d4, deeper evaluations keep oscillating between them.
**Note on good and BAD play**: as we'll be looking at WORST moves and games, there's a similar catch as when looking for the BEST ones (see note on perfect play above). When judging something as good or bad, we have to ask "good or bad considering WHAT kind of players?" (what skill, what goal, what kind of behavior, ...) -- best move for an engine may require precise play and so may not be best for human, and best move for a grandmaster may not be the best for average player, AND also a good move against human may be not best against a computer and vice versa. For example when looking for the worst move in a position, the first we think of is this: consider all moves and take the one which will take us to a position that has the worst evaluation by computer engine. This is quite cool, but not always and may not really be what we want, because when evaluating the position, the computer assumes GOOD play from both sides. So when we e.g. flip the rules and try to make computers play the worst moves and get themselves mated, they should rather assume the opponent to play the WORST moves, we want a different kind of estimate -- here it's not enough to offer opponent a checkmate, but also ensure he MUST give it. So these are some things to keep in mind.
How big is the **white's starting move advantage**? Based on the above evaluations of all starting moves the initial position is rated at 18 centipawn (for the best move found, e4), i.e. with this specific engine and search depth we are told white has, in material terms, an advantage of almost a fifth of a pawn.
What's **the best and worst opening move according to the engines?** With what's been said above, the answer will also depend on which engine (what evaluation function) you use and to what depth you search. The situation is basically this: both engines and humans are deciding between e4 or d4 for the best move, opinions differ and strongest engines currently oscillate between e4 and d4 as we keep analyzing the starting position deeper and deeper. According to Lichess cloud database (accessible via public API) that stores stockfish evaluations for various positions, e4 leads to the best evaluated position (18 centipawn, evaluated to depth 70), closely followed by d4 and Nf3 (both 17 centipawn, depth 47 and 56) and c4 (12 centipawn, depth 59). Lichess stockfish is currently an older version (14) also running in the web browser, so not absolutely strongest, but still very strong. On Chessbase (proprietary database) someone analyzed the starting position to depth 79 with stockfish 12, giving evaluation 0.17 and best move d4 (followed by Nf6 c4 ...). Running the strongest version of stockfish at this point (stockfish 17) to depth 65 (which took some 4 hours) also gives best move d4 (followed by Nf6), but as said, it just seems to go there and back between e4 and d4. So pick one. Worst move, as in "leading to worst evaluation in Lichess database" (also usually given by humans), is by far g4 with evaluation -96 centipawn (depth 52) -- almost a whole pawn, i.e. stockfish says that by playing this move you basically just throw away your pawn immediately. Another bad move is apparently f3 (-76, depth 40), Nh3 (-42), Na3 (-33), b4 or h4 (both -28). So g4 is likely the worst move under normal conditions, however if we play an opponent who is also trying to play the worst moves, i.e. we flip the rules and make each player try to get himself mated (in a computer engine flip the sign of the evaluation function), the engine actually elects e3 as the worst move, because that allows the white queen to immediately run out, attack the enemy king face to face and force him to take it.
What's **the perfect game according to an engine**? Again, this will vary depending on new and better versions of engines coming out, on hardware, time we spend on computing moves etc. The following draw was produced by taking the first six highly analyzed moves from the Lichess cloud database and letting the rest of the game be played by stockfish 15.1 (each player given 30 minutes plus 5 second increment, moves were usually analyzed to 50+ depth):
How big is the **white's starting move advantage**? Based on the above evaluations of all starting moves the initial position is rated at about 18 centipawn (for the best move found, e4), i.e. with this specific engine and search depth we are told white has, in material terms, an advantage of almost a fifth of a pawn.
What's **the perfect game according to an engine**? Again, this will vary depending on new and better versions of engines coming out, on hardware, time we spend on computing moves etc. The following annotated draw was produced by taking a few first highly analyzed moves from the Lichess cloud database and letting the rest of the game be played by stockfish 17, the strongest available engine at the time, at a reasonably powerful desktop PC, giving each player 90 minutes plus 10 second increment, with endgame tablebases proving that since 7 men on the board the game is really a theoretical draw:
```
1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. O-O Nxe4 5. Re1 Nd6 6. Nxe5 Nxe5 7. Rxe5+ Be7
8. Bf1 O-O 9. d4 Bf6 10. Re1 Re8 11. c3 Rxe1 12. Qxe1 Ne8 13. Bf4 d5 14. Nd2 Bf5
15. Qe2 Nd6 16. Nb3 b6 17. Re1 c6 18. Nd2 h6 19. h3 Qd7 20. Qd1 Rd8 21. Nf3 Nc4
22. b3 Nd6 23. Qc1 c5 24. Ba6 Re8 25. Rxe8+ Nxe8 26. Qe3 Nc7 27. Be2 Ne6 28. Be5
Be7 29. Bd3 Bxd3 30. Qxd3 Ng5 31. Nxg5 hxg5 32. a4 f6 33. Bg3 Bd6 34. dxc5 Bxg3
35. Qxg3 bxc5 36. Qe3 d4 37. cxd4 cxd4 38. Qd3 Kf7 39. Kf1 Qd6 40. Qc4+ Kf8 41.
Ke2 Qe5+ 42. Kd1 Qe4 43. Qc5+ Kg8 44. Qc4+ Kf8 45. b4 Qb1+ 46. Kd2 Qb2+ 47. Ke1
d3 48. Qc5+ Kf7 49. Qd5+ Ke7 50. Qb7+ Kd6 51. Qa6+ Ke5 52. Qb5+ Kf4 53. Qc4+ Ke5
54. Qb5+ Ke4 55. Qc6+ Ke5 56. Qc5+ Ke6 57. Qc4+ Ke5 58. Qc5+ Ke6 59. Qc6+ Ke5
60. Qc7+ Ke4 61. Qb7+ Kd4 62. Qxa7+ Kc3 63. Qc5+ Kb3 64. Qd5+ Kc3 65. Qc6+ Kb3
66. Qd5+ Kxa4 67. Qxd3 Kxb4 68. Qe4+ Kc3 69. Qc6+ Kd3 70. Qd6+ Qd4 71. Qa6+ Qc4
72. Qxc4+ Kxc4 73. Ke2 Kd4 74. Kf3 Kd3 75. g3 f5 76. h4 gxh4 77. gxh4 Kd2 78. h5
Ke1 79. Kg2 f4 80. Kf3 Kf1 81. Kxf4 Kxf2 82. Kf5 Kg3 83. Kg6 Kg4 84. h6 gxh6 85.
Kxh6 1/2-1/2
1. e4 {Lichess } e5 {Lichess } 2. Nf3 {Lichess } Nc6 {Lichess }
3. Bb5 {Lichess } Nf6 {Lichess } 4. O-O {+0.10/49} Nxe4 {-0.06/49}
5. Re1 {+0.12/46} Nd6 {-0.06/45} 6. Nxe5 {+0.09/49} Be7 {-0.07/46}
7. Bf1 {+0.09/47} Nxe5 {-0.06/42} 8. Rxe5 {+0.15/44} O-O {-0.10/43}
9. Nc3 {+0.20/58} Bf6 {-0.09/43} 10. Re1 {+0.13/43} Re8 {-0.04/41}
11. Nd5 {+0.08/46} Rxe1 {-0.10/42} 12. Qxe1 {+0.12/49} b6 {-0.12/45}
13. Nxf6+ {+0.13/44} Qxf6 {-0.16/45} 14. c3 {+0.16/56} Bb7 {-0.14/40}
15. d3 {+0.12/42} Re8 {-0.13/42} 16. Qd1 {+0.12/43} c5 {-0.11/40}
17. Qg4 {+0.18/41} Bc6 {-0.12/42} 18. Bd2 {+0.10/43} g6 {-0.10/39}
19. Rc1 {+0.12/42} h5 {-0.10/46} 20. Qg3 {+0.13/39} b5 {-0.12/47}
21. b3 {+0.02/48} a5 {-0.02/40} 22. Bg5 {+0.03/47} Qe6 {-0.02/45}
23. Bd2 {+0.00/45} Qf6 {-0.01/44} 24. a3 {+0.08/48} Kh7 {+0.00/42}
25. Qf4 {+0.03/46} Qxf4 {+0.00/48} 26. Bxf4 {+0.00/38} Nf5 {+0.00/52}
27. f3 {+0.00/43} f6 {+0.00/54} 28. Kf2 {+0.00/47} Ra8 {+0.00/48}
29. Bc7 {+0.00/52} b4 {+0.00/63} 30. a4 {+0.00/56} Bd5 {+0.00/53}
31. d4 {+0.00/49} Bxb3 {+0.00/59} 32. dxc5 {+0.00/60} Bxa4 {+0.00/60}
33. Ra1 {+0.00/76} Bc6 {+0.00/63} 34. cxb4 {+0.00/53} axb4 {+0.00/47}
35. Rxa8 {+0.00/48} Bxa8 {+0.00/55} 36. Ba5 {+0.00/51} Nd4 {+0.00/56}
37. Bxb4 {+0.00/47} Kg7 {+0.00/50} 38. Bd2 {+0.00/54} h4 {+0.00/69}
39. Bf4 {+0.00/64} Bc6 {+0.00/61} 40. Bd3 {+0.00/50} Ne6 {+0.00/78}
41. Bd6 {+0.00/54} Nd4 {+0.00/63} 42. Bf4 {+0.00/69} Ne6 {+0.00/65}
43. Bd6 {+0.00/57} Ng5 {+0.00/56} 44. Bf4 {+0.00/68} Nf7 {+0.00/56}
45. Ke3 {+0.00/70} Nd8 {+0.00/56} 46. Bc7 {+0.00/58} Ne6 {+0.00/62}
47. Bd6 {+0.00/69} Nd8 {+0.00/62} 48. Kf4 {+0.00/63} Nf7 {+0.00/80}
49. Kg4 {+0.00/58} g5 {+0.00/60} 50. Kh3 {+0.00/46} Nd8 {+0.00/58}
51. Kg4 {+0.00/54} Ne6 {+0.00/66} 52. f4 {+0.00/50} Bxg2 {+0.00/66}
53. fxg5 {+0.00/47} fxg5 {+0.00/65} 54. Bf5 {+0.00/70} Kf6 {+0.00/85}
55. Bxe6 {+0.00/87} dxe6 {+0.00/78} 56. Bc7 {+0.00/80} h3 {+0.00/90}
57. Ba5 {+0.00/59} Ke5 {+0.00/73} 58. Kxg5 {+0.00/67} Kd5 {+0.00/85}
59. Bb6 {+0.00/65} e5 {+0.00/74} 60. Ba7 {+0.00/84} Bf1 {+0.00/69}
61. Kg4 {+0.00/72} e4 {+0.00/89} 62. Kf4 {+0.00/95} Bb5 {+0.00/89}
63. Bb6 {+0.00/65} Bc6 {+0.00/84} 64. Kg3 {+0.00/69} e3 {+0.00/66}
65. Kxh3 {+0/80,draw} e2 {+0.00/84} 66. Ba5 {+0.00/70} Kxc5 {+0.00/93}
67. Kg3 {+0.00/78} Kd4 {+0.00/85} 68. Kf2 {+0.00/89} Kd3 {+0.00/82}
69. Bb4 {+0.00/88} Be4 {+0.00/88} 70. Ba5 {+0.00/85} Bh7 {+0.00/91}
71. h4 {+0.00/77} Bg6 {+0.00/72} 72. Bb4 {+0.00/72} Bh5 {+0.00/81}
73. Be1 {+0.00/65} Bg6 {+0.00/85} 74. Kf3 {+0.00/83} Bh5+ {+0.00/69}
75. Kf2 {+0.00/87} Kc2 {+0.00/70} 76. Ke3 {+0.00/77} Kd1 {+0.00/76}
77. Kf2 {+0.00/69} Bf3 {+0.00/70} 78. Bb4 {+0.00/81} Bh5 {+0.00/77}
79. Ke3 {+0.00/65} e1=R+ {+0.00/76} 80. Bxe1 {+0.00/58} Kxe1 {+0.00/72}
81. Kf4 {+0.00/80} Kf2 {+0.00/87} 82. Kg5 {+0.00/84} Bf7 {+0.00/83}
83. h5 {+0.00/98} Bxh5 {+0.00/82} 1/2-1/2 {insufficient material}
```
What's the **theoretically worst game possible**, and how to find out? This is easy: just sit two [women](woman.md) at a chessboard and watch :D OK, [jokes](jokes.md) aside -- like with the perfect game we will probably never know, plus there are the pecularities mentioned above about how we really define "bad play". Anyway we may try this: take the best engine and just revert its evaluation function, i.e. literally flip the sign of evaluation (in practice we usually have to handle some additional stuff in the code that relied on normal evaluation) -- this should basically internally revert the rules of chess to trying to get mated, AND also make sure we assume the opponent is trying to do the same etc. This game will represent the serious effort to really force your opponent to beat you. Doing this with the current best engine, stockfish 17, giving both players 30 minutes plus 10 second increment, leaves us with the following beautifully terrible, excruciatingly long abomination of a game:
```
1. e3 e6 2. Qh5 Qg5 3. Qxf7+ Kd8 4. Qe7+ Kxe7 5. f4 Qg3+ 6. Ke2 Qf3+ 7. Kd3 Qd5+ 8. Ke2 Qd3+
9. Kf3 Qe2+ 10. Bxe2 Kf6 11. Bd3 Nc6 12. Kg4 h5+ 13. Kh3 Rh7 14. Bf5 Ba3 15. Nc3 Ke7
16. Bg4 Nd4 17. Ne4 hxg4+ 18. Kg3 Rh3+ 19. gxh3 Nf3 20. Nd6 c6 21. Ne2 Nf6 22. Nxb7 Ne4+
23. Kg2 Nh4+ 24. Kg1 Ng2 25. Nc3 Ke8 26. Nd5 Ng3 27. Kf2 Ne1 28. Nd8 Bb7 29. h4 a6
30. Ne7 Nf1 31. Kg1 d5 32. Nxe6 Nf3+ 33. Kf2 Ne1 34. Ke2 Nd3 35. Kd1 Nxb2+ 36. Ke2 Ng3+
37. Kf2 Nf1 38. Nc7+ Kf7 39. Ne8 Bc5 40. Nc8 Bxe3+ 41. Kg2 Bb6 42. Na7 Bd8 43. Nf6 Rb8
44. a3 Ke7 45. d4 Rc8 46. Be3 Bc7 47. Nb5 Be5 48. Rd1 Nd3 49. Rc1 Nb4 50. a4 Rg8
51. Nd7 Ke6 52. Nb6 Nd2 53. Bg1 Nf1 54. Kf2 Nd3+ 55. Ke2 Rc8 56. Nc4 Rb8 57. Nca3 Rd8
58. f5+ Ke7 59. h3 Ne3 60. f6+ Ke6 61. Nc7+ Kf5 62. Rf1+ Kg6 63. Na8 Nc1+ 64. Kd2 Bc7
65. Nb5 Ba5+ 66. c3 Nc2 67. Rf3 Rc8 68. h5+ Kh7 69. Re3 Rc7 70. h4 g6 71. Re4 Rg7
72. Kd1 Bxc3 73. f7 Ne3+ 74. Kxc1 Bb2+ 75. Kd2 Nc4+ 76. Ke2 Ne5 77. f8=N+ Kg8 78. Nd6 Rh7
79. Rh3 Bc1 80. Rf3 Bf4 81. Nf7 Rh6 82. Kf2 Nd7 83. Rh3 Rh7 84. Ke2 Nb8 85. Nh8 a5
86. Rh2 Ba6+ 87. Kf2 Be3+ 88. Ke1 Bd2+ 89. Kd1 Re7 90. Kc2 Kg7 91. Rh1 Re5 92. Nh7 Bb7
93. Nf7 Be1 94. Kd3 g5 95. Nh8 Ba6+ 96. Ke3 Bd2+ 97. Kf2 Be1+ 98. Ke3 Be2 99. h6+ Kg8
100. Nb6 Bd2+ 101. Kf2 g3+ 102. Kg2 Bf1+ 103. Kf3 Be2+ 104. Kg2 Bf1+ 105. Kf3 dxe4+
106. Kxg3 Be1+ 107. Bf2 Bh3 108. Nc8 Re7 109. Nf6+ Kf8 110. Ng6+ Kf7 111. Nd6+ Kxf6
112. Ne8+ Kf7 113. Ne5+ Kf8 114. Nd7+ Kg8 115. Nc7 Re5 116. Nd5 Bg2 117. Nb4 Bf1
118. Rh3 Bb5 119. Kg4 Ba6 120. Rf3 Rf5 121. Nc5 Kf8 122. d5 Ke7 123. Nd7 Rf4+ 124. Kh3 Rf6
125. Nc5 Bf1+ 126. Kg3 Rd6 127. Nxc6+ Ke8 128. Rf8+ Kxf8 129. Ne6+ Kf7 130. Nb4 Bh3
131. Nxg5+ Ke7 132. Ne6 Rb6 133. Kf4 Bf5 134. Bc5+ Rd6 135. Bg1 Bd2+ 136. Ke5 Rxd5+
137. Nxd5+ Kd7 138. Nb6+ Ke8 139. Bc5 Bf4+ 140. Kd4 Bd6 141. Bb4 Bg4 142. Kc3 Be2
143. Nc5 Na6 144. Nd3 Nc5 145. Nc8 Bd1 146. Kc4 Bb3+ 147. Kc3 Ne6 148. Kd2 Bd1 149. Ne5 Ng7
150. Nd3 Bc5 151. Nf2 Kd7 152. Kc3 Bd4+ 153. Kd2 Ke6 154. Ng4 Bc3+ 155. Ke3 Nf5+
156. Kf4 Bd2+ 157. Ne3 Bxe3+ 158. Kxe4 Be2 159. Bd6 Nd4 160. Ne7 Bd3+ 161. Kxe3 Ne2
162. Bb4 Be4 163. h5 Nf4 164. Nf5 Kd5 165. Ne7+ Ke6 166. Nf5 Ke5 167. Bc3+ Kd5 168. Bb4 Bf3
169. Kf2 Ke5 170. Kg3 Ne2+ 171. Kh4 Ng3 172. Ne3 Bg4 173. Bd6+ Ke4 174. Kg5 Kf3
175. Bb4 Ne4+ 176. Kg6 Bd7 177. Kf7 Nc3 178. Kf6 Kf4 179. Bd6+ Ke4 180. Bb4 Nd5+
181. Kg5 Bb5 182. Nc4 Ne3 183. Nd2+ Kd5 184. Nc4 Kd4 185. Kf4 Nd5+ 186. Kg5 Be8
187. Bc3+ Kd3 188. Bb4 Bd7 189. Nd6 Bb5 190. Nc4 Be8 191. Kf5 Ne3+ 192. Ke5 Ke2
193. Kf4 Ng2+ 194. Ke4 Bc6+ 195. Ke5 Ne3 196. Kf4 Kd3 197. Ke5 Nd5 198. Nb2+ Ke2
199. Nd3 Nf4 200. Ne1 Nd3+ 201. Kd4 Nc5 202. Nd3 Nb3+ 203. Ke5 Nd4 204. Nf4+ Kf3
205. Nd3 Be8 206. Ne1+ Ke3 207. Bd2+ Ke2 208. Kf4 Bxa4 209. Nd3 Kd1 210. Bb4 Ne2+
211. Kf5 Bd7+ 212. Ke5 Nc1 213. Ke4 Kc2 214. Nb2 Bc6+ 215. Ke3 Ne2 216. h7 Nc3
217. h8=N Nd1+ 218. Kf4 Ne3 219. Nf7 Nf1 220. Ng5 Bf3 221. Ne6 Kc1 222. Nc5 Bb7
223. Nb3+ Kb1 224. Nc1 Kc2 225. Nb3 Ba8 226. Kg4 Bb7 227. Bd2 Ng3 228. Bb4 Bf3+
229. Kh3 Bh1 230. Nc1 Bg2+ 231. Kh4 Bh3 232. Kg5 Ne2 233. Nb3 Bd7 234. Bc3 Nf4 235. Kh4 Ne2
236. Nd2 Ng3 237. Kg5 Ne2 238. Kh4 Be6 239. Nb3 Bg4 240. Na1+ Kb1 241. Nd3 Ng3 242. Kg5 Ne2
243. Bb2 Nf4 244. Kh4 Ne2 245. Nb4 Kxb2 246. Na2 Nf4 247. Kg3 Nh3 248. h6 Ka3 249. h7 Ka4
250. h8=N Kb5 251. Nb4 Ng5 252. Nf7 Ne4+ 253. Kh4 Kc4 254. Nb3 Kc3 255. Nd6 Ng5 256. Nc4 Bh3
257. Kh5 Bf5 258. Kh6 Bg6 259. Kg7 Nh7 260. Nd4 Nf6 261. Nb5+ Kb3 262. Nb2 Ng8 263. Nc2 Bf7
264. Kh7 Bg6+ 265. Kh8 Bh5 266. Nca3 Bf7 267. Na4 Kxa4 268. Nc3+ Kb4 269. Na4 Bd5
270. Kh7 Bf7 271. Kh8 Nf6 272. Kg7 Ng8 273. Kxf7 Kb3 274. Ke6 Ne7 275. Ke5 Nc6+
276. Kd5 Nb4+ 277. Kc5 Nc2 278. Nb5 Ka2 279. Nb2 Ne3 280. Nd4 Nc4 281. Nd3 Nd2 282. Nb3 Kb1
283. Na1 Ne4+ 284. Kd4 Nd6 285. Nc2 Ka2 286. Na1 Kb1 287. Nc2 Ka2 288. Nc5 Kb2 289. Kd3 Nc4
290. Na4+ Kc1 291. Nb2 Nxb2+ 292. Kc3 Na4+ 293. Kd3 Nc5+ 294. Kc3 Na4+ 295. Kd3 Nc5+
296. Kc3 Ne4+ 297. Kd3 Nf2+ 298. Kc3 Ne4+ 299. Kd3 Nf2+ 300. Kc3 Nd1+ 301. Kd3 Kb2
302. Ne1 a4 303. Kd2 Nf2 304. Ke2 Nh1 305. Nd3+ Kb1 306. Kf3 a3 307. Nb4 a2 308. Nxa2 Kxa2
```
Lichess analysis seems to only handle the first 150 moves, the evaluation graph explodes up and down and almost jumps out of the roof. The following are the analysis results (for the first 150 moves). White: 15 inaccuracies, 15 mistakes, 97 blunders, 581 average centipawn loss, accuracy: 21%. Black: 11 inaccuracies, 17 mistakes, 97 blunders, 587 average centipawn loss, accuracy: 21%. That doesn't seem that bad, why aren't all moves blunders? Well, firstly the analysis is relatively quick (takes like 10 seconds for whole game), it likely doesn't see as deep as the engines who were given hours to play, but secondly we changed the rules of the game: the analyzing engine still assumes the players will be playing good moves, which is not the case.
For comparison here is another bad game in which we just take regular stockfish 17 and make moves like this: from all possible moves, minus the ones that draw, choose the one that leads to the position with worst evaluation for us. 3 seconds are given for evaluating each possible move, so we get something around a minute to make a move. For "mate in N" we take the move that gets us mated sooner as better, and to decide between several "mate in N" moves with same N we try to estimate the worst by taking an average static evaluation of the board to depth 3 (for technical reasons we use [smallchesslib](smallchesslib.md)'s evaluation) -- this should help us prefer positions in which there are more ways to get ourselves mated or in which we at least lost most material and other advantage on average. This game embodies the effort to make the worst blunder in each move in a regular game of chess -- as such we won't see too many "forced blunders", just great many generous offers that keep being turned down. In result this produced another terribly long game:
{ My computer basically spent the whole day computing this game instead of mining Monero, so please enjoy :D ~drummyfish }
```
1. g4 f5 2. f3 g5 3. Kf2 Kf7 4. Ke3 Ke6 5. Kd4 Qe8 6. b4 Qh5 7. f4 Kf6 8. Ke3 Qh3+ 9. Kd4 Qc3+
10. Kd5 Qb2 11. h4 h5 12. Nf3 Bh6 13. Kc5 a6 14. d4 Qxd4+ 15. Nxd4 Ra7 16. c4 Kg7 17. Bg2 Kf6
18. gxh5 c6 19. Qa4 Rh7 20. Qa5 Ra8 21. Qb6 Ra7 22. Qa5 Ra8 23. Qb6 Rf7 24. Bb2 d5 25. Nd2 Ra7
26. Rh3 a5 27. Qa6 e6 28. b5 g4 29. Rf3 Bd7 30. b6 Bg7 31. Kd6 Rf8 32. Kc7 Rf7 33. Kd6 Rf8
34. Kc7 Rd8 35. c5 g3 36. Kd6 Rf8 37. Kc7 Rf7 38. Kd6 Rf8 39. Kc7 Rf7 40. Kd8 Nxa6 41. Rc3 Ne7
42. Ne4+ fxe4 43. a4 Nb8 44. Nxe6 Kf5 45. Bf3 Rf6 46. Rd1 e3 47. Nf8 Kxf4 48. Be4 Rd6
49. Nxd7 Bf6 50. Rdd3 Kxe4 51. Rc2 Bd4 52. Rb3 Rf6 53. Nf8 Rf3 54. Rd2 Rf5 55. Ba1 Ng6
56. Rb5 g2 57. Bc3 Ne5 58. Bb2 Nbd7 59. Kc8 Nf6 60. Kd8 Ne8 61. h6 Nf6 62. Bc1 g1=B
63. Ke7 Rg5 64. Rbb2 Kf4 65. Nh7 Ne8 66. h5 Rg4 67. Kxe8 Nf7 68. Rdc2 Bf6 69. Kf8 Ke4
70. Rc3 Rg7 71. Ng5+ Kd4 72. Ke8 Bd8 73. Ne4 Bh2 74. h7 Rg8+ 75. Kd7 Bd6 76. h6 B8e7
77. Ng3 Ke5 78. Rc4 Bc7 79. Rd4 Kf6 80. h8=N Kg5 81. Rc4 Nd6 82. Rd2 Raa8 83. Nh1 Ra7
84. Ke6 Rg7 85. Ke5 Bcd8 86. bxa7 b6 87. Ng6 Nc8 88. Nf8 Nd6 89. a8=N b5 90. Ng6 Bb6
91. Bb2 Nc8 92. Nf8 Na7 93. Nc7 Bbxc5 94. h7 Bd4+ 95. Ke6 Ba3 96. Ng3 Bc3 97. Rxc6 b4
98. h8=N Nb5 99. Rd6 Kh6 100. Kf5+ Bf6 101. Nh7 d4 102. Na8 Rxh7 103. Nh5 b3 104. Ba1 Na7
105. Kf4 Rf7 106. Ng6 Re7 107. Nf8 Rb7 108. Ng7 d3 109. Kg3 Rb4 110. Ng6 Bg5 111. Rf6 Bf4+
112. Kh4 Rxa4 113. Rd6 dxe2 114. Be5 Nc8 115. Nb6 Ne7 116. Nh5 Bb2 117. Bd4 e1=R 118. Na8 Nc6
119. Bb6 Na7 120. Rc6 Nc8 121. Rc4 Ra3 122. Kg4 Kh7 123. Bd8 Rb1 124. Nh8 Bf6 125. Ng7 Kxg7
126. Rc5 Bh6 127. Rc4 Kf8 128. Nf7 Kg8 129. Rb4 Rd1 130. Nh8 Kg7 131. Kh5 Bf4 132. Be7 a4
133. Rf2 Bb8 134. Bd8 Rf1 135. Bb6 Bd4 136. Rf5 Rg1 137. Rf6 Na7 138. Rb5 Rc1 139. Kg5 Nc8
140. Ra5 Rc7 141. Rc5 Ba7 142. Rc4 Bc5 143. Rf8 Bd4 144. Rf6 Bc5 145. Rf8 Rf7 146. Rxa4 Bd4
147. Kh4 Bb8 148. Rd8 Ra1 149. Kh3 Rc7 150. Kh2 Rg1 151. Rg8+ Kf6 152. Rg2 Na7 153. Kh3 e2
154. Ba5 Bf2 155. Be1 Bh4 156. Ra5 Rf1 157. Rg7 Re7 158. Ra3 Rf7 159. Bf2 Ra1 160. Nb6 Ke7
161. Kg2 Re1 162. Nd7 Bf6 163. Ra1 Bh2 164. Bd4 Bb8 165. Ra2 Ra1 166. Rd2 Ra2 167. Nxf6 Ra1
168. Bb6 e1=R 169. Bd4 Re2+ 170. Bf2 Re3 171. Ne4 Ra6 172. Kg1 Re6 173. Nd6 Kf8 174. Rd1 R6e4
175. Rc1 Rc4 176. Be1 Rec3 177. Nf5 Rf4 178. Ne7 Rb4 179. Nf5 Rcc4 180. Nd6 Rb6 181. Rh7 Nc6
182. Ra1 Rc3 183. Bg3 Rc4 184. Be5 Ne7 185. Bf6 Ba7 186. Ra2 Rc2 187. Rh2 Ng6 188. Rd2 Re7
189. Kf2 Rc1 190. Rac2 Rg1 191. Bd4 b2 192. Nhf7 Rd7 193. Bxb2 Bb8 194. Ba1 Ra6 195. Nc8 Kg8
196. Rc6 Rd8 197. Rd3 Nh8 198. Bxh8 Rc1 199. Rc2 Ra8 200. Rd5 Rg1 201. Rd1 Bc7 202. Rc3 Ra6
203. Nb6 Rd7 204. Rc4 Bf4 205. Rc7 Bh6 206. Rc5 Rc7 207. Na8 Rb6 208. Rc2 Bf8 209. Rd4 Rg4
210. Rd6 Bh6 211. Ng5 Rc8 212. Ke3 Rd8 213. Rc1 Rc6 214. Rcd1 Rh4 215. Bd4 Bg7 216. Rh6 Bf8
217. Ne6 Rc4 218. Bf6 Rc5 219. Be5 Be7 220. Rc1 Rf8 221. Nb6 Bd8 222. Rb1 Rc6 223. Ng5 Rc7
224. Bd6 Rc2 225. Bc5 Rb2 226. Bd6 Rc2 227. Nd7 Rf5 228. Rb5 Rg2 229. Rb7 Rc2 230. Nc5 Rf7
231. Bc7 Bf6 232. Bh2 Rc1 233. Rc7 Be5 234. Rb6 Rcc4 235. Rb5 Rh8 236. Nh3 Rf2 237. Rcb7 Bb8
238. Rf7 Rd4 239. Rb2 Rf3+ 240. Ke2 Rd7 241. Ng1 Rb7 242. Kd2 Re3 243. Kc2 Rd3 244. Kb1 Rc3
245. Rb6 Rch3 246. Rb5 Rd7 247. Rb3 Rxh2 248. Ka1 R2h4 249. Na4 R4h5 250. Rd3 Re7 251. Re3 Bg3
252. Rf4 Re4 253. Ref3 Rf5 254. Rf1 Re2 255. R4f2 Kf8 256. Nf3 Rd5 257. Nd2+ Bf4 258. Kb2 Rd3
259. Rh2 Re4 260. Re2 Re8 261. Ka2 Rde3 262. Rb1 R3e4 263. Rbe1 Rc4 264. Ka3 Rh6 265. Nf1 Rb4
266. Rb1 Bb8 267. Rb3 Kg8 268. Re1 Kh8 269. Nb6 Rh2 270. Ng3 Rc4 271. Ne2 Rh3 272. Rf1 Ba7
273. Nc8 Bg1 274. Rf5 Rf4 275. Rg3 Re5 276. Nb6 Rd4 277. Rh5+ Rhxh5 278. Nc1 Rh3 279. Rb3 Re2
280. Na4 Rb4 281. Rd3 Bd4 282. Rf3 Kh7 283. Rg3 Kh8 284. Nc3 Bg1 285. Rg4 Rg3 286. Rf4 Rb5
287. Nd3 Rg4 288. Rb4 Rh4 289. Nf4 Rb8 290. Nd3 Re5 291. Nxe5 Re4 292. Na4 Rc4 293. Nc3 Re4
294. Na4 Rc4 295. Nc5 Bh2 296. Ra4 Ra8 297. Ncd7 Rb4 298. Ka2 Rb3 299. Nb8 Bg1 300. Ng4 Bb6
301. Ka1 Bd8 302. Ka2 Bb6 303. Ka1 Bd8 304. Ra2 Rd3 305. Ra6 Kg8 306. Nf6+ Kh8 307. Ng4 Rh3
308. Nh6 Rc3 309. Ng8 Rh3 310. Nh6 Ra7 311. Ra2 Re3 312. Kb2 Rc3 313. Ra5 Rf7 314. Nf5 Bg5
315. Ng7 Rd7 316. Nc6 Rd5 317. Ra3 Rb5+ 318. Rb3 Rb8 319. Rb6 Be7 320. Ne8 Kg8 321. Rb5 Bf6
322. Ka2 Re3 323. Nd8 Rb7 324. Ne6 Kh7 325. Rb4 Rd7 326. Ng5+ Kh8 327. Re4 Rf7 328. Kb1 Re2
329. Rb4 Ra2 330. Rf4 Bb2 331. Kc2 Rh7 332. Rf6 Ra6 333. Rc6 Rf7 334. Nf6 Rg7 335. Kd2 Ba3
336. Nfh7 Kg8 337. Nf8 Rf7 338. Ke2 Kh8 339. Ne4 Kg7 340. Nd6 Kxf8 341. Rc2 Rf5 342. Nf7 Rh6
343. Nh8 Rg6 344. Rd2 Rf4 345. Rd7 Rd4 346. Re7 Rd7 347. Kf2 Rc7 348. Re2 Rc2 349. Kf1 Bb2
350. Rd2 Bc1 351. Re2 Bb2 352. Ke1 Bc1 353. Nf7 Kg8 354. Nd8 Rf6 355. Re6 Rc4 356. Kd1 Rg6
357. Re5 Rc2 358. Re6 Bb2 359. Re2 Kh8 360. Re5 Rg5 361. Re6 Rg1+ 362. Re1 Kg8 363. Nf7 Re2
364. Rf1 Bf6 365. Ne5 Bh4 366. Ng4 Be1 367. Rf2 Rg3 368. Rf4 Rxg4 369. Kxe2 Kg7 370. Ke3 Bh4
371. Rd4 Bg3 372. Rc4 Kg6 373. Rd4 Kg7 374. Rd6 Rd4 375. Rh6 Kf8 376. Rd6 Rd1 377. Rd2 Kf7
378. Rd4 Ke6 379. Rd2 Bd6 380. Rxd6+ Ke5 381. Re6+ Kf5 382. Re7 Rf1 383. Re6 Kg5 384. Rh6 Rh1
385. Kd3 Kf4 386. Kd4 Rh3 387. Kc5 Kf3 388. Kd6 Rh5 389. Kc6 Rb5 390. Kd6 Rb3 391. Rg6 Ke3
392. Rh6 Ra3 393. Rg6 Rb3 394. Rh6 Ra3 395. Ke6 Ke4 396. Rh3 Kd4 397. Rh6 Rh3 398. Kf6 Kd5
399. Kf5 Kc4 400. Kf6 Kd4 401. Rh7 Kd3 402. Rh4 Kc3 403. Rc4+ Kb3 404. Ra4 Kc3 405. Ra6 Rg3
406. Ke6 Rf3 407. Kd6 Rg3 408. Ke6 Rf3 409. Kd6 Rh3 410. Ke7 Kd3 411. Ra7 Kc3 412. Ke8 Rf3
413. Kd7 Rg3 414. Ke7 Kd3 415. Rb7 Rh3 416. Ra7 Rg3 417. Rb7 Rg8 418. Rb8 Ke3 419. Rb7 Rd8
420. Rb8 Re8+ 421. Kd7 Re7+ 422. Kd8 Re8+ 423. Kd7 Re7+ 424. Kd8 Rb7 425. Ke8 Rb3 426. Kf7 Kf4
427. Rb7 Kf5 428. Ke7 Ke4 429. Kd7 Rxb7+ 430. Kd6 Re7 431. Kc5 Re5+ 432. Kd6 Re7 433. Kc5 Rd7
434. Kc4 Rd5 435. Kc3 Rd2 436. Kb4 Kd5 437. Kc3 Ke6 438. Kc4 Rd4+ 439. Kb3 Rc4 440. Ka2 Kd6
441. Kb2 Rc3 442. Ka1 Kc5 443. Ka2 Rb3 444. Ka1 Rb2 1/2-1/2
```
Again, Lichess only analyzed the first 150 moves and here it marked practically every move as a blunder (the rest are probably blunders so genius that the quick analysis didn't even reveal the genius behind that stupidity), because of our different definition of a bad move. Here are the statistics. White: 1 inaccuracy, 1 mistake, 147 blunders, 1610 average centipawn loss, accuracy: 2%. Black: 1 inaccuracy, 0 mistakes, 148 blunders, 1613 average centipawn loss, accuracy: 2%.
What is **the rarest move**? Some [YouTube](youtube.md) video tried to investigate this with the help of Lichess database. Things that immediately come to mind like en passant checkmates and checkmates by promoting to a knight are rare but not insanely rare. A crazily rare kind of move, which only appeared ONCE in the whole database, was a doubly disambiguatated (i.e. with the necessary specification of both rank and file of the bishop) checkmate by a bishop (specifically Bf1g2#, occurring in a 2022 game) -- this is rare because to need a double disambiguation for a bishop move it is necessary to underpromote two pawns to a bishop and then place them correctly. Yet rarer moves, which NEVER appeared in the database, were a doubly disambiguated knight checkmate with capture and doubly disambiguated bishop checkmate with capture, latter of which was judged less likely and therefore probably the rarest move ever.
Anyway, you can try to derive your own stats, there are huge free game databases such as the Lichess [CC0](cc0.md) database of billions of games from their server.
@ -269,17 +425,19 @@ Some general tips and rules of thumb, mostly for beginners:
- Don't bring the queen out too early, the opponent can harass it and get ahead in development.
- Learn some universal setup openings or "systems" to play, e.g. London, King's Indian, the hippo etc.
- Develop your men before attacking, usually knights go out before bishops, bishops are well placed on the longest diagonals as "snipers".
- Learn basic tactics, especially **forks** (attacking two or more men at once so that one of them cannot escape capture) and **pins** (attack one man so that if he moves out of the way he will expose another one to be captured).
- Learn basic tactics, especially **forks** (attacking two or more men at once so that one of them cannot escape capture) and **pins** (attack one man so that if he moves out of the way he will expose another one to be captured), but also other things like double checks, sacrifices etc.
- Learn basic types of checkmates and mating patterns, for example mating with two rooks, with king and queen, back rank mates (especially dangerous when starting, always make an escape square), smothered mates, how not to stalemate etc.
- Watch out for hanging pieces! You usually want to have everything guarded.
- King safety is extremely important until endgame, castle very early but not extremely early. In the endgame (with queens out) king joins the battle as another active man.
- Learn when to exchange and when not -- usually when you're up material, trades are good for you because trading same value men increases the ratio of your material advantage. When attacking you probably don't want to trade because you need SOMETHING to attack with -- so when your opponent is attacking you and you're defending, you WANT to trade because he probably doesn't want that. Also when you have little space and are squeezed in your base, you probably want to trade and create more space. Etc.
- Pawn structure is very important (avoid doubled and isolated pawn, watch out for the weak back pawns etc.).
- Watch out for back rank checkmates, make an escape square for your king.
- Rooks want to be on open files, you also want to CONNECT them (have both guard each other). Also a rook in the opponents second row (2nd/7th rank) is pretty good.
- Stack rooks, i.e. place them on the same open file -- this is very powerful. You can also stack two rooks and a queen and create a so called legendary triple stack which is extremely powerful.
- If you find a good move, look for a better one. There seems to be this pattern in which if you spot a good move, it indicates that opponent's position is falling apart and usually there is a much more powerful, crashing move to play.
- Bishops are generally seen a bit more valuable than knights, especially in pairs -- if you can trade your knight for opponent's bishop, it's often good. If your opponent has two bishops and you only have one, you want to trade yours for his so he doesn't have the pair. A knight pair is also pretty powerful though, especially when the knights are guarding each other.
- Consider the bishop difference: one only covers white squares, the other only black ones. Take this into account when exchanging bishops, sacrificing them, placing your pieces on white vs dark squares etc.
- "Knight on a rim is dim" (knights are best placed near the center).
- Blocking the opponent's man so that he can't move (i.e. making it inactive) is almost as good as taking it. And vice versa: you want to activate all your men if possible, put them on good squares and make them do something.
- Blocking the opponent's man so that he can't move (i.e. making it inactive) is almost as good as taking it. And vice versa: you want to activate all your men if possible, put them on good squares and make them do something. Take space from opponent and "squeeze" him, having little space and few moves is generally bad.
- Nubs are weak against long range bishops, they can't see them. Place a bishop to corner on the long diagonal and just snipe the opponent's material. See also fianchetto.
- Don't play "hope chess", always suppose your opponent will play the best move he can. Don't give a check just because you can, always try to invalidate the move you want to play and only play it if you can't find an easy counter to it.
- If you can achieve something with multiple men, usually it's best to do it with the weakest one.
@ -297,6 +455,7 @@ WORK IN PROGRESS, pls send me more tips :)
- OTB (over the board) only:
- Turn your knights to face backwards or in another weird way (always face the opponent's king etc.). Also place the pieces unevenly on the squares to piss off opponents with OCD and autism.
- Play anti-computer chess against human opponent.
- Play `1. Qe9#`.
- Behave weird, make weird faces, walk extremely far away from the board and walk in circles (or just get up and stand up directly behind your opponent in a completely upright position staring into the distance without moving at all like a robot lol), constantly sneeze (try to sneeze every time the opponent touches a piece), make very long unbroken eye contact with the opponent while smiling as if you know what he's thinking, call the referee constantly, go to the toilet after every move, pretend to fall asleep from boredom etc. Overeat on beans before the game so you fart a lot and always try to fart as loud as possible. Wear nice clothes but right before the game go sweat to the gym so that you smell like a pig and distract the opponent with toxic fume. If you're a [wimmin](woman.md) behave sexually, keep grabbing your boobs, lick your lips and opponent's captured pieces and silently moan sometimes as if you're having an orgasm, pretend to masturbate under the table; if your opponent is male he is almost definitely smarter than you, you gotta use your woman weapons, but it will probably work easily on the chess virgins.
- In a tournament change play based on opponent's [race](race.md) or sex, for example play only one opening against white people and another opening against black people, see if anyone notices the pattern :D
- Outside tournament take advantage of the fact that you can do whatever the fuck you want: have one hand constantly on the clock and play with the other hand (considered rude and often forbidden), touch and knock over your opponent's pieces, take back your moves, ... and of course when you're losing, "accidentally" knock over the whole board and be like "oops, let's consider it a draw then" :D
@ -307,7 +466,7 @@ WORK IN PROGRESS, pls send me more tips :)
- Be annoying and offensive in chat, if opponent blunders write `gg`, spam `ez` when you win. If he wins say it was a shit game and accuse him of [cheating](cheating.md).
- Constantly ask for takebacks, offer draws, report legit opponents for cheating and offensive behavior.
- ...
- Play the bongcloud, fool's mate, 1. h3 or similar offensive opening, especially against a stronger player. Offer a draw after 1st move. Just play knight F3 and back constantly. Castle manually even if you don't have to. Play the exact mirror of opponent's moves -- if he tries to break it then just always try to get back to mirrored position or do some similar shit.
- Play the bongcloud, fool's mate, 1. h3, 1. g4 or similar offensive opening, especially against a stronger player. Offer a draw after (or even before) the 1st move. Just play knight f3 and back constantly. Castle manually even if you don't have to. Play the exact mirror of opponent's moves -- if he tries to break it then just always try to get back to mirrored position or do some similar shit.
- When losing constantly offer draws, prolong the game AS MUCH AS POSSIBLE, before the very last move just let the clock run out.
- Repeatedly try to make swastikas on the board, especially against colored opponents.
- Underpromote pawns to knights or bishops.