Why Stockfish Level 5 on Lichess is so Weak ?

[Chessqueen]

[pgn][Event "Casual Correspondence game"]
[Site "https://lichess.org/XVoxCn7a"]
[Date "2022.01.12"]
[Round "?"]
[White "ChessLearner1979"]
[Black "lichess AI level 5"]
[Result "1-0"]
[BlackElo "1700"]
[ECO "D02"]
[Opening "Queen's Pawn"]
[Variation "2.Nf3 g6"]
[WhiteElo "1700"]
[TimeControl "600"]
[Termination "normal"]
[PlyCount "137"]
[WhiteType "human"]
[BlackType "program"]

1. d4 d5 2. Nf3 {D02 Queen's Pawn Game: Zukertort Variation} g6 3. e3 Bg7
4. c3 Nf6 5. Bd3 O-O 6. O-O c5 7. Nbd2 c4 8. Bc2 Nc6 9. Re1 Nd7 10. e4 e6
11. exd5 exd5 12. Qe2 Qc7 13. Nf1 Rb8 14. g3 Qb6 15. Bf4 Qxb2 16. Bxb8
Ncxb8 17. Reb1 Qa3 18. Qd2 b6 19. Rb4 Qa5 20. a4 a6 21. Rab1 Nc6 22. R4b2
Bb7 23. Qf4 Nd8 24. Ne5 Ne6 25. Qe3 Bxe5 26. dxe5 Re8 27. Rxb6 Nxb6 28.
Qxb6 Qxc3 29. Qxb7 Qxc2 30. Qb2 Qxb2 31. Rxb2 Nc5 32. Ne3 c3 33. Rc2 d4 34.
Nd1 g5 35. f4 Nxa4 36. Nf2 a5 37. Nd3 g4 38. Kf2 Kg7 39. Ke2 f5 40. Kd1 Nb6
41. Ra2 Nc4 42. Ra4 Nb2+ 43. Nxb2 cxb2 44. Kc2 Rb8 45. Kb1 h6 46. Rxa5 Kf7
47. Ra7+ Ke6 48. Ra6+ Kd5 49. Rxh6 Ke4 50. e6 Re8 51. Kxb2 Re7 52. Kc2 Rc7+
53. Kd2 Re7 54. h3 Rc7 55. hxg4 d3 56. gxf5 Kxf5 57. Kxd3 Re7 58. Ke3 Rxe6+
59. Rxe6 Kxe6 60. Ke4 Kf7 61. g4 Kf6 62. g5+ Kg7 63. Kf5 Kg8 64. Kf6 Kh8
65. Kf7 Kh7 66. g6+ Kh8 67. g7+ Kh7 68. g8=Q+ Kh6 69. Qg6# {White wins by
checkmate.} 1-0[/pgn]

[Chessqueen]

Even on AI Level 6 of stockfish is NOT that solid, or is it that I have done too many practice against Komodo Dragon 2.6.1 and it has made me a stronger tactical player, I already have played more than 200 Blitz games against Komodo Dragon 2.6.1. since I got it and gradually I feel that I am becoming a better player. On is it that I got Lucky, or playing lots and lots of games has helped me tremendously against Komodo Dragon 2.6.1

[pgn][Event "Casual Correspondence game"]
[Date "2022.01.12"]
[Round "2"]
[White "ChessLearner1979"]
[Black "lichess AI level 6"]
[Result "1-0"]
[ECO "B01"]
[Opening "Scandinavian"]
[Variation "Schiller, 4.d4 Nf6"]
[WhiteElo "1700"]
[TimeControl "600"]
[Termination "normal"]
[PlyCount "171"]
[WhiteType "human"]
[BlackType "program"]

1. e4 d5 2. exd5 Qxd5 3. Nc3 Qd6 {B01 Scandinavian Defense: Gubinsky-Melts
Defense} 4. d4 Nf6 5. Nf3 c6 6. Ne5 Be6 7. a3 Nbd7 8. Nc4 Bxc4 9. Bxc4 e6
10. Be3 Qc7 11. Qf3 Nd5 12. Nxd5 cxd5 13. Bb3 Bd6 14. c3 Rd8 15. g3 Qc6 16.
Qg4 Qb6 17. Qxg7 Rf8 18. Ba4 Qa5 19. Bc2 Qb5 20. Bc1 a5 21. Qh6 Qc4 22. a4
Qa6 23. Qe3 Ke7 24. Qd3 Qxd3 25. Bxd3 h5 26. Be3 Rc8 27. O-O Rg8 28. Kh1 h4
29. gxh4 f5 30. Rg1 f4 31. Bd2 Nf6 32. f3 Nd7 33. Rae1 Rxg1+ 34. Rxg1 Nb6
35. b3 Kf7 36. h5 Rh8 37. Bg6+ Ke7 38. c4 Bb4 39. Bxf4 dxc4 40. Be5 Rc8 41.
h6 Nd7 42. bxc4 Nxe5 43. dxe5 Bd2 44. h7 Bf4 45. Re1 Bd2 46. Re4 b6 47. Rg4
Bb4 48. Rf4 Rh8 49. Rf7+ Ke8 50. Ra7+ Kf8 51. Rf7+ Ke8 52. Kg2 Be7 53. Rg7+
Kf8 54. Rf7+ Ke8 55. f4 Bb4 56. Ra7+ Kf8 57. Ra8+ Kg7 58. Rxh8 Kxh8 59. Kg3
Bc3 60. Kg4 Kg7 61. Kg5 Be1 62. h4 Bg3 63. h5 Bf2 64. f5 Be3+ 65. Kg4 exf5+
66. Kxf5 Kh8 67. e6 Bc5 68. Kf6 b5 69. e7 Bxe7+ 70. Kxe7 bxc4 71. Kd6 c3
72. Kc5 Kg7 73. Bc2 Kh8 74. Kb5 Kg7 75. Kxa5 Kh8 76. Kb6 Kg7 77. a5 Kh8 78.
a6 Kg7 79. a7 Kf6 80. a8=Q Kg5 81. Qe8 Kg4 82. h8=Q Kf3 83. Qf6+ Kg2 84.
Qeg6+ Kh2 85. Qf2+ Kh1 86. Qgg2# {White wins by checkmate.} 1-0[/pgn]

[Sopel]

It's weak because it's designed to be weak.

[Chessqueen]

It's weak because it's designed to be weak.

Thanks, for the answer Sopel, I was NOT expecting the full strength Stockfish, but at least to be more competitive at that Level, Anyway thanks. Probably you should try it at Level 6 as well, but of course you are much much better player.

[jtwright]

IIRC one of the basic ways SF plays at lower levels is by intentionally choosing the not-best move from time-to-time. This is particularly apparent when playing against level 1, where you could just hang a queen for several turns and it feels like SF actively avoids taking it. I've made a game before out of trying to maximize number of "blunders" you could get in the post-game analysis in a game against SF level 1.

If this concept does extend to the higher levels, then I would imagine that against other AB engines, the stated Elo reported by Lichess is not particularly accurate. Playing a suboptimal move in certain tactical lines will be obvious to even 1700-2000 Elo engines, for instance, where they might be missed by human players with similar Elo (depending on the line). In those cases, the other engine will immediately capitalize and convert on the advantage.

I feel like I've had a similar experience with the Boris-Trapsky BOT on lichess. Its gimmick is that it tries to intentionally play moves sometimes that it knows are sub-optimal but thinks are very "tricky/gambity". This can make for some interesting games for humans (which is the intent in this case), but other engines are generally unimpressed and just punish the mistake effortlessly.

I'm no expert on SF levels though, this is my impression from only a brief glance and a few games with it on lichess. I know Larry Kaufman has put a lot of thought on levels for Komodo and probably has a lot of interesting knowledge on accurately simulating "weaker" play and what it takes.

[purechess]

Lichess Stockfish level has NOTHING to do with "Skill Level" settings on Stockfish.

Their levels are a combination of max depth, move time and "Skill level".
You can look up this in their source code.

[Chessqueen]

Lichess Stockfish level has NOTHING to do with "Skill Level" settings on Stockfish.

Their levels are a combination of max depth, move time and "Skill level".
You can look up this in their source code.

You are correct, but Level 6 plays better than level 5, and level 5 plays better than level , I just got lucky to play on my first game and won, since I went back and played 6 more games at level6 and lost them all

Note: Most of the players playing at Level 6 are at least rated 2000 by Lichess and they are not having an easy game against Stockfish Level 6, but at bullet time control or faster time control even 2400 to 2500 human rated by Lichess are having a hard time at level 2. At Bullet, I can not get a win in 100 games, can NOT think that fast not even at Level 1.
https://lichess.org/tv/computer

Finally got another win, but NOT on Level 6, this time I lowered myself to Level 5 playing at 10 minutes. I will try playing at 3 minutes level 5 next.

[pgn]][Event "Casual Correspondence game"]
[Date "2022.01.13"]
[Round "?"]
[White "ChessLearner1979"]
[Black "lichess AI level 5"]
[Result "1-0"]
[ECO "D10"]
[Opening "Slav"]
[Variation "Exchange"]
[WhiteElo "1700"]
[TimeControl "600"]
[Termination "normal"]
[PlyCount "59"]
[WhiteType "human"]
[BlackType "program"]

1. d4 d5 2. c4 c6 3. cxd5 {D10 Slav Defense: Exchange Variation} Qxd5 4.
Nc3 Qa5 5. e4 Nf6 6. Qc2 e6 7. Nf3 Nbd7 8. Be2 c5 9. O-O Qd8 10. b4 cxd4
11. Nb5 Ke7 12. Nfxd4 Rg8 13. e5 Nxe5 14. Qc5+ Kd7 15. Qxe5 a6 16. Bc4 Qb6
17. Nf5 Nd5 18. Rd1 Kd8 19. Rxd5+ Ke8 20. Nc7+ Qxc7 21. Qxc7 exd5 22. Bg5
Be6 23. Bxa6 bxa6 24. Re1 f6 25. Bxf6 Bc5 26. bxc5 Ra7 27. Rxe6+ Kf8 28.
Re8+ Kxe8 29. Nd6+ Kf8 30. Qd8# {White wins by checkmate.} 1-0[/pgn]

[Chessqueen]

I have a new opening book with traps, and I used this trap against Stockfish at level 5, and once I got the advantage I did NOT feel like playing out this easy won game, so I resigned in a won position. I will continue reading this old book of opening traps.

[pgn][Event "Casual Correspondence game"]
[Date "2022.01.13"]
[Round "?"]
[White "ChessLearner1979"]
[Black "lichess AI level 5"]
[Result "0-1"]
[ECO "B13"]
[Opening "Caro-Kann"]
[Variation "Panov-Botvinnik, 5.Nc3"]
[WhiteElo "1700"]
[TimeControl "300"]
[Termination "normal"]
[PlyCount "24"]
[WhiteType "human"]
[BlackType "program"]

1. e4 c6 2. d4 d5 3. exd5 cxd5 4. c4 Nf6 5. Nc3 {B13 Caro-Kann Defense:
Panov Attack} Bf5 6. Qb3 Qc7 7. Nb5 Qc6 8. cxd5 Nxd5 9. Qxd5 Qxd5 10. Nc7+
Kd8 11. Nxd5 Bd7 12. Ne3 Nc6 {. White resigns} 0-1[/pgn]