Has Top Engine Ever Lost a Game?

[swami]

With the moderately fast hardware, it did lose a lot of games in 1990's, less so in 2000's, even lesser in 2010's.

Now, I'm hearing that for the past one year, the top engine, namely Stockfish hasn't lost a single game, and minimum it could achieve is a draw. Is that true? If so, that's impressive.

Ofcourse, games lost due to negative lines in openings don't count.
That (infinite draw) might happen in longer time controls, surely not in Rapid or Blitz - right?! That indicates, there's still a scope for improvement...

[Brunetti]

Now, I'm hearing that for the past one year, the top engine, namely Stockfish hasn't lost a single game, and minimum it could achieve is a draw. Is that true?

No.

Alex

[smatovic]

We are getting closer to the death by draw in computer chess:

viewtopic.php?p=967285#p967285

Modern tournaments like TCEC and CCCC (and FishTest) use unbalanced opening books, lowers draw rate.

--
Srdja

[Viz]

With decent enough hardware and enough time I do believe that no correspondence player or whoever will be able to win against sf from startposition.

[Jouni]

https://www.iccf.com/event?id=100104. You can lose only if You die . Tansel has 2x AMD EPYC 7V12 ddr4 2133 = 256 threads.

[jefk]

In ICCF correspondence chess you can also lose if you make an input mistake; happened to me in the past,
sometimes i was analyzing with Idea (with various engines) a position which would occur further down in the
tree, while playing sometimes two or three tournaments, so quite some games. Then when the Idea solution
came up, eg. a move like 22.f4!? whereas i was still in position 19 or so, i occasionally plugged it into
the Iccf interface, and then later noticed that 19.f4?? was a blunder.

So in some ways it seems to have become a data entry contest; a method to be promoted from CCM to IM nowaways
is to play lots of games, and inputting to ICCF your moves with high accuracy (double or triple check). Then wait
for an input mistake of your opponent, which happens in about 1 pct of all games, then when you got 1 point
extra in one of the (eg. the six) tourns you're playing, and all the other games drawn, you get an IM norm.
I've recently seen two guys getting an IM norm in such a way; maybe i'm also going to do it, starting end of the
year or so (although i'm now betraying my secrets, lol (but not all Iccf players are on talkchess i guess). Anyway
you won't become an CCM just like that, at least not fast, by plugging in SF moves because in principle anyone can do that.

Note that at lower levels, eg Iccf < 1900, some people (beginners/newbies?) with not much knowledge about (correspondence)
chess, openings, and computer chess in general, still can make a mistake which can cost them the game.
As eg. in this latest tourn where i participate:
https://www.iccf.com/event?id=106001

One game which i already won went as follows:
1.e4 e5 2.f4! exf4 3.Nf3 g5 4.Nc3!! Bg7 5.d4 g4 6.Bxf4! gxf3 7.Qxf3 d6 8.Bb5+ Bd7!?(*) 9.O-O! Nc6?!(**) 10.d5 Nd4 11.Qg3 Nxb5
12.Qxg7 Qf6 13.Bh6 Qxh6 14.Qxh8 Qe3+ 15.Kh1 O-O-O 16.Rf3!! Qd2 17.Nxb5 Bxb5 18.Rxf7 1-0 (Black resigns).

In this game I choose to play first for a KGA and then the -rare- Quaade (****)gambit, sacrificing a knight at move 6, to get a
sharp game (so *not* chosing simply Sf moves..), then lateron at (or in fact before) every move i analyzed various lines in
depth (yes with engines ofcourse) and then again went for either a good/strong or (initially) another very sharp continuation.
My unexperienced opponent played fast (one move per day), and wasn't aware he was lured into dangerous territories.
Somewhere, around move 8, Black went into a 'novelty' (simply a -weak- engine move i suppose), and i got chances.
At move 16, i looked for an Lc0(!) eval (not SF!) and found a move (not the SF first move) which could lead to a significant
advantage in the endgame (pawnstorm on White kings side). Then already at move 18 Black apparently noticed he was getting
into quite some endgame problems, and resigned. No real input mistakes were made by Black (but then black
also didn't use a modern SF on a fast modern comp, being maybe too optimistic about his chances).

In my game against Panugaling (1773) i now think i also can go for a win(***), and thus maybe win this tourn.
Next round obviously will most likely be more difficult or simply a gamble (or all draws).

(*) in Nepustil(2370)-Valuk(2319), ICCF, 2016, Black played 8...c6 and the game ended in a draw
(**) maybe an SF move (at not to big depth), but Obsidian thinks Bxb5 would have been better (-0.35)
(***) similar to the first game, this opponent also didn't make an input mistake i think, but apparently somewhere
during the opening stage apparently didn't use an engine (or a lousy one) and simply played a move (maybe too
fast) which he thought was sufficient; but actually turned out to be a mistake.
(****) Quaade, not Quade (as sometimes used), a Danish (not Dutch as sometimes thought) sea captain (1825-1906)
and chess player (just like Evans) btw; apparently some liked to play (lots of) chess in the 19th century on the sea
https://en.wikipedia.org/wiki/William_Davies_Evans

[Jouni]

I look from PGN to adjucated games ICCF. Final position was mostly 0,00 some 0,05.

[Peter Berger]

Note that at lower levels, eg Iccf < 1900, some people (beginners/newbies?) with not much knowledge about (correspondence)
chess, openings, and computer chess in general, still can make a mistake which can cost them the game.
As eg. in this latest tourn where i participate:
https://www.iccf.com/event?id=106001

One game which i already won went as follows:
1.e4 e5 2.f4! exf4 3.Nf3 g5 4.Nc3!! Bg7 5.d4 g4 6.Bxf4! gxf3 7.Qxf3 d6 8.Bb5+ Bd7!?(*) 9.O-O! Nc6?!(**) 10.d5 Nd4 11.Qg3 Nxb5
12.Qxg7 Qf6 13.Bh6 Qxh6 14.Qxh8 Qe3+ 15.Kh1 O-O-O 16.Rf3!! Qd2 17.Nxb5 Bxb5 18.Rxf7 1-0 (Black resigns).

In this game I choose to play first for a KGA and then the -rare- Quaade (****)gambit, sacrificing a knight at move 6, to get a
sharp game (so *not* chosing simply Sf moves..), then lateron at (or in fact before) every move i analyzed various lines in
depth (yes with engines ofcourse) and then again went for either a good/strong or (initially) another very sharp continuation.
My unexperienced opponent played fast (one move per day), and wasn't aware he was lured into dangerous territories.
Somewhere, around move 8, Black went into a 'novelty' (simply a -weak- engine move i suppose), and i got chances.
At move 16, i looked for an Lc0(!) eval (not SF!) and found a move (not the SF first move) which could lead to a significant
advantage in the endgame (pawnstorm on White kings side). Then already at move 18 Black apparently noticed he was getting
into quite some endgame problems, and resigned. No real input mistakes were made by Black (but then black
also didn't use a modern SF on a fast modern comp, being maybe too optimistic about his chances).

If we assume that black used little or no computer assistance all his moves look pretty respectable to me.
HIs major mistake was that he "apparently noticed he was getting into quite some endgame problems, and resigned".

In the final position he has 18. ..Nh6 (not too obvious for a human). I suppose play continues 19. Rf8 Nf7 20. Rxd8+ Nxd8 21. Qxh7 Qxc2 . I don't see how you want to win this against the likes of Stockfish.

Peter

[Viz]

Opponent resigned in a drawable position, "At move 16, i looked for an Lc0(!) eval (not SF!) and found a move (not the SF first move) which could lead to a significant advantage in the endgame (pawnstorm on White kings side). " - Rf3 is sf 1st line (at least at some depths), also all game apart from the opening is sf either 1st or 2nd line with score almost equal to 1st one, while opponent played some weak moves that sf never approved here and there.
So idk what to brag about, sincerely. You played against a person who probably doesn't even use SF and maybe uses some old ass fritz or smth and managed to "win" by playing stockfish moves, "win" being not a real win since opponent prematurely resigned.
What an achievement! And a completely irrelevant one to this topic.

[jefk]

PB wrote

If we assume that black used little or no computer assistance all his moves look pretty respectable to me.

only a very strong human player could play like that, usually humans go astray in lines like this (look at eg. GM Shaw's
book about the King's gambit, where he recommends 4.Nc3! ( instead of Bc4), the Quaade
https://qualitychess.co.uk/products/ope ... john_shaw/
In this book he only talks about 8.Bc4 instead of 8.Bb5+! btw. and suggests it will be very difficult for humans at all levels,
but maybe not so good at top correspondence level. In my database i see four games with 8.0-0-0 and one
with 8.Bb5 (as earlier mentioned which ended in a draw). The four games with 8.0-0-0 all were ICCF games,
from 2009-2020 and one game was won with White [Milde(2325) -Lindberg(2128), 2020).
So i think 8.Bb5 and 0-0-0 besides at human level (although it would be new) can also be played at correspondence level
(and at the time hadn't looked even at Shaw's book, buty maybe his recommendation 4.Nc3 was in the back of my mind).
So I think the opponent used a computer but only in a superficial way, to assist him in avoiding to make
obvious blunders, some sort of 'advanced chess' as in the old times, not deep computer analysis.

In the final position he has 18. ..Nh6 (not too obvious for a human). I suppose play continues 19. Rf8 Nf7 20. Rxd8+ Nxd8 21. Qxh7 Qxc2

yep and i suppose he also noticed that with his simple (or shallow) engine .

I don't see how you want to win this (against the likes of SF)

good point, i also was a bit surprised he resigned, this -also- looks to have been mostly a human
judgment , Lc0 gives 22.b3 (not b4?! as SF) in line with a (human) plan to advance the double
free pawns on the g and h line. Lc0 gives WDL 282/523/195, clearly (a bit) better for White.
Presumably my opponent was not interested in playing long defensive end game and apparently
also didn't know how strong the modern SF's (or similar) can be at sufficient calculation time(s) eg. a
few min's per move on a fast comp (so mostly some sort of newbie problem i guess).
But i also -like you- know that SF often can defend tenaciously especially at the end with egtbs.
(and before that probably bring some pieces and later the king to the right flank of the board;
it would an interesting study maybe to analyze this in depth as a practical endgame, no time for this now
(maybe later, at the moment i don't have a fast GPU (for Lc0 and with my experience with the latest Nnue
SF's it remains doubtful it the small Lco (wdl) advantage could be increased gradually in this endgame.