Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / testsuite 5356 engine mistakes
- - By gsgs (**) [de] Date 2017-04-03 09:17 Edited 2017-04-03 09:29
http://magictour.free.fr/FCP449.FEN

5356 FEN-positions from games at FCP with both players of Elo >2900 [link]
http://amateurschach.de/main/_conditions.htm
where the own eval and the opponent eval dropped by at least 0.30
after the move made

stockfish8 bench 32 1 xx FCP449.FEN depth > c:\ll

gives

d13:
d14:1655
d15:1676
d16:1650
d17:1625
d18:1542
d19:1463
d20:

1s,tolino,1 thread, asmfish Mar06 : 1720,1682
5s,tolino,4 threads, (1700kn/s)asmfish Mar06 : 1313
5s,tolino,4 threads, asmfish Dec07 =
5s,tolino,4 threads, asmfish Nov04 =
5s,tolino,4 threads, asmfish Oct04 =
5s,tolino,4 threads, asmfish Jan06 =
5s,tolino,4 threads, asmfish Frb06 =
30s,tolino,4threads,asmfish Apr06 =

the idea is to use such lists of mistakes specific to the playing conditions
to test engines faster than letting them play games

I have a small program to count the matches in FCP449.FEN and c:\ll
I also have a batch-file to run this for engines without the bench-feature
or the asmfish -"wait" feature

but maybe the GUIs handle this well, I don't know

post your results for other top-engines

the entries in FCP449.FEN per line are :

-------------------------------

the 6 fields of the FEN , separated by space

number of game in the FCP .PGN file)
number of ply in the game
Elo-White
Elo-Black
last eval of opponent
own eval given with the (bad) move
next eval of opponent
next own eval
(bad) move in short notation
(bad) move in long notation

------------------------------------
Parent - - By user923005 (****) [us] Date 2017-04-04 00:06
File http://magictour.free.fr/FCP449.FEN is neither EPD nor FEN.
If an engine can make any sense of it, it must be custom programmed.
There are also duplicate rows.
Parent - By user923005 (****) [us] Date 2017-04-04 00:23
Duplicate
2r1rbk1/2qn1p1p/p2p1np1/Pb4B1/1ppPP3/5NNP/1P1Q1PP1/RBR3K1 w
2r1rbk1/2qn1p1p/p2p1np1/Pb4B1/1ppPP3/5NNP/1P1Q1PP1/RBR3K1 w
Parent - - By user923005 (****) [us] Date 2017-04-04 00:25
duplicates:
r1q2rk1/1p2ppbp/6p1/pN1pP3/Pnb1nP2/R1N1B3/1PP1B1PP/3Q1R1K b
r1q2rk1/1p2ppbp/6p1/pN1pP3/Pnb1nP2/R1N1B3/1PP1B1PP/3Q1R1K b
Parent - - By gsgs (**) [de] Date 2017-04-04 05:10 Edited 2017-04-04 05:27
duplicates don't hurt, if some positions are more likely to occur in practice,
then they should also be tested more often. I will run the testsuite several times
anyway, so to see the deviation and compare with the deviation
of games-playing-results. Hmm, with one thread I may need some randomizing.

I had no problem so far with the format, stockfish,asmfish,houdini,komodo accept it.
But I can easily change it, if some engine or GUI has a problem with it ?!

Under Windows-cmd.exe I create a batch-file to feed the FENs , using a "wait.exe" - utility.
First 8 positions in http://magictour.free.fr/fcpt5.bat
[using wait.exe utility from Uwe Gabbert
https://server1.uwegabbert.info/download/wait/wait.exe ]
run  :  fcpt1.bat | stock8.exe > file
Under Android for asmfish I use the asmfish "wait" -command
(available in asmfish since Feb.2017)
more exact descriptions in the German CSS-forum

------------------------------------------

results update:
[other results are welcome]

stockfish 8 , depth 13-20 :

d13:1642
d14:1655
d15:1676
d16:1650
d17:1625
d18:1542
d19:1463
d20:1399

1s,tolino,1 thread, asmfish Mar06 = 1720,1682
1s,tolino,4 threads, asmfish Mar06 =

5s,tolino,4 threads, (1700kn/s in the startposition)

5s,tolino,4 threads, asmfish Oct04 =
5s,tolino,4 threads, asmfish Nov04 =
5s,tolino,4 threads, asmfish Dec07 = (error: unknown command wait)
5s,tolino,4 threads, asmfish Jan06 = (error: unknown command wait)
5s,tolino,4 threads, asmfish Feb06 = 1320
5s,tolino,4 threads, asmfish Mar06 = 1313
30s,tolino,4 threads,pedantfish Mar28 =
30s,tolino,4threads,asmfish Apr06 =
Parent - - By Peter Grayson (****) [gb] Date 2017-04-04 06:52
Unclear what the value would be for engine testing because as far as I can see the positions are already going adverse for one side. In many cases Stockfish 8 for example finds that out in a few seconds or less. I could understand their value if there was a key move that saves the day? In that case the key move should have been included as the benchmark but it seems to be a case of having to find the best of a bad bunch of moves?

PeterG
Parent - - By gsgs (**) [de] Date 2017-04-04 07:27 Edited 2017-04-04 07:41
well, it is supposed that there are better moves, than the move made, since both
evals before that move were at least 0.30 better.
In some cases -my wild guess is ~20% - it could be, that there are really no better moves
and it only takes a while that the engines realize how bad the position really is.

You would make these "blunder"-test positions from games that match the testing
conditions which you are doing, so different sets for STC and LTC.
Ideally the average tested engine would make ~half of the test-blunder-moves.

I have no experience with this, but I'd assume that it gives better results in shorter
time than playing games. You avoid all these non-important drawish games and moves ...
For the same purpose some testers are using sharp opening books now.

For now I am just counting the matching moves but you could improve this by assigning
evals to the possible moves.

[all evals are white-based, positions with evals >1.50 or <-1.50 were excluded]
Parent - - By Peter Grayson (****) [gb] Date 2017-04-04 10:13
It may be better to use more than one engine because the meaning of the engine output is not necessarily consistent with engines of say 10+ years ago when the score was very much in terms of +1 or -1 being a pawn advantage.

I ran the first 1600 positions at 5s per position with Stockfish 8 and there were several positions coming out at +2 or -2 or more. With several minutes for them and the score got worse. I've attached the zipped dump file of the 5s analysis. Open in Notepad or Wordpad and search for +2 or -2. The output was from Stockfish 8 32 bit in Windows XP home, CPU Q9550 at 2.8GHz and 1 Gb hash.

Just my take on it but I think it may be of more value if the position included a thematic to identify the engine's "understanding" of the position or a single key tactical move.

PeterG
Attachment: FC5sanalysis.zip - 5s dump file (1584k)
Parent - - By gsgs (**) [de] Date 2017-04-04 11:22 Edited 2017-04-04 12:29
thanks.
376 of the "blunders" were found in your output in those first 1619 FENs.
I had 452 at depth 20 in the first 1619 and 464 at depth 19.

I assume you are using some GUI or software for this ? While I feed this directly
from commandline to the engines.

It seems that the castling flags are not recognised correctly ,
so I should change their format, I always had 4 digits in {K,Q,k,q,-}^4

I will check this and update the file at the link above soon.
-------is updated now : http://magictour.free.fr/FCP449.FEN
-------edit-----------------------
ahh, yes , and your program doesn't recognize (anti-) solutions, maybe because it uses
still another move-notation, I added that too.

-----------edit------------------------------
I used the evals of the playing engines, these were >=2900 at FCP,
so about >3100 at CCRL. I could maybe make a similar list from
CCRL-games later.
Yes, stockfish immediately finds some of these blunders and often has
quite a different eval. But there should be enough diversity to produce
meaningful output.

My "thematic" is to use real mistakes from real games made in that
tournament, so to produce about the same Elos.
Whether you should use mistakes only - that is debatable.
I'll probably try the same thing later with (surprising) moves, after which both evals went up.
But I want to eliminate the majority of moves, where not much happens.

The engine's "understanding" of a position is hard to measure ..., hmm, maybe positions
where the evals differed a lot and then compare with the evals 5 moves later or such
(who was right in the first place ?) ?
That could just be calculated automatically from the logs without further engine running.
Parent - - By user923005 (****) [us] Date 2017-04-04 18:23
Looking at your data set, it looks to me that you are simply capturing fail high and fail low.
You have a move listed as avoid move, but it is often the best move.
There are also many positions that are simply dead lost like these:
8/2r2k2/6p1/8/6R1/1K6/1PP5/8 b - -

5r2/R7/2k5/8/2p3Pp/7P/p1K5/8 w - -


What is the actual goal of the data set?
Are they actually supposed to be avoid move problems?
Parent - By gsgs (**) [de] Date 2017-04-05 03:53
no fail-high/low, since only the final eval is available usually, when you download games.
And both playing engines saw a drop in eval in those positions, the evals are given after the fen.
I have no estimate yet in how many of the positions the made move is no mistake
and the drop in eval is just due to realizing that the real value of the position
is worse than thought. Could be up to 50% ? These could be filtered out later
to improve speed but should not falsify the rating much.
Stockfish at low depth wants to make ~31% of the moves (depth 15 )
(Houdini , Komodo ~33%, 1s. old hardware, <1% of the targeted depth of the played games)
and then it's going down to 26% at depth 20 and presumably further down at higher depths.
The goal is to use that number (at fixed reflection time) as a rating of the engine
hopefully better, more reliable, and faster to achieve than playing games.
Yes, avoid the given moves. (for now, later I may add evals of the best moves)

update of tested engines and depths is currently slow, but probably faster later
Parent - - By Peter Grayson (****) [gb] Date 2017-04-04 20:44
I used the Shredder GUI "Analyse EPD" option. The engine output window gives the option to record the full engine output to a dump file.

"Surprising" moves could be interesting provided not too shallow a depth to find.

PeterG
Parent - - By gsgs (**) [de] Date 2017-04-05 03:59
it should work now correctly with the updated format
(castling flags, move notation)

maybe someone can run a smaller subset (10,100) at ~FCP-time control
to compare some different engines ?
Parent - By Peter Grayson (****) [gb] Date 2017-04-05 08:39
Looking at Frank's home page, there seems to be a hardware contradiction. He refers to the i7-4770K CPU as being 8 cores and the i7-5930k as being 6 cores that for the former would require hyperthreading to be active but later he refers to match conditions with hyperthreading being swtched off in the  BIOS.

The 512 Mb hash per engine philosophy may seem fair when read on the surface but in practice it is insufficient for some engines. Houdini and Stockfish for example are big hash users and giving 512Mb is not running them under optimised conditions even with just 1 thread and ponder off when Houdini 5.01 and Stockfish 8 for example can register 80%+ hash full during the game at the 40 moves in 10 minutes time controls on his hardware. 1 Gb hash for Stockfish and Houdini looks better in practice for game play under his conditions.

For position analysis keeping the hash value low but sufficient enough to prevent a full condition for the allocated analysis time generally gives good results although some hash values seem better than others. Too large a hash setting can add significant initialisation time particularly when LP hash is used that is quite noticeable in position batch testing. The mix of big hash and short analysis time is not good.

PetertG
Parent - - By gsgs (**) [de] Date 2017-04-05 05:47
this was initiated by this 111-tactical test :

http://forum.computerschach.de/cgi-bin/mwf/topic_show.pl?tid=9355
(in German, with link to 111 tactical testpositions from human games
and test-rating software)

and a software to calculate ELOs from the solution times
with a surprisingly good match with existing ELO-lists :

http://magictour.free.fr/ERETP2.GIF
Parent - By user923005 (****) [us] Date 2017-04-07 20:03
My analyzed eret records are attached.
Attachment: eret.epd - Walter Eigenmann: ERET (33k)
Parent - By user923005 (****) [us] Date 2017-04-07 19:46
Attached is a subset of the positions.
I did not analyze all of them, and some of them clearly are positions where the wrong move was played.
But a substantial portion I think the avoid move is probably as good or better than any other move.

In the attached file, there are positions like this one which are clearly won:
3r4/6k1/4p3/1n1rP2P/1p1N1P2/1PPR4/8/2K5 w - - acd 38; acs 900; am Nxe6+; bm Nxe6+; ce 1496; pm Nxe6+; pv Nxe6+;

and there are positions like this one which are clearly lost:
5r2/R7/2k5/8/2p3Pp/7P/p1K5/8 w - - acd 36; acs 287; am Rxa2; bm Rxa2; ce -32585; pm Rxa2;
pv Rxa2 Kb5 Ra1 Rf2+ Kc3 Rf3+ Kd4 Rd3+ Ke5 c3 g5 Rxh3 g6 Kb4 Rb1+ Ka3 Ra1+ Kb2 Ra8 Rg3 Rb8+ Ka2 Rc8 Kb3 Rb8+ Kc4 Rc8+ Kd3 Rd8+ Ke2 Rc8 h3 Kf6 Kf2 Rh8 Kg2 Re8 h2 Re2+ Kh3 Re1 Rg1 Re3+ Kh4 Re4+ Kg3 Re5 Kf2 Rf5+ Ke3 Re5+ Kd4 Kf5 Rxg6;
And there are positions like this one which are clearly drawn:
6k1/3n1b2/1p4p1/8/P2p1BP1/8/4Q2K/1q1B4 b - - acd 46; acs 900; am Nc5; bm Nc5; ce 0; pm Nc5;
pv Nc5 Be5 d3 Qd2 Qa2 Bb2 Kf8 Qc3 Ke7 Qf6+ Kd7 Qxb6 Bd5 Kg3 Ne4+ Kf4 g5+ Ke5 Bc6 Qa7+ Ke8 Qb8+ Kd7 Qa7+;

In all of the cases for the positions in the file, it is not clear to me that there are any better alternatives than the "avoid move"
Attachment: mistrakes.epd - Are there really any superior alternatives to the "avoid move" am field? (81k)
Parent - - By gsgs (**) [de] Date 2017-09-11 07:12 Edited 2017-09-11 10:04
I filtered 1231 of these : http://magictour.free.fr/fcp1231.fen ,
those where asmfish moves were different, and tested it on asmfishes,

http://magictour.free.fr/1231G1.GIF
(to be updated)

now on my new Ryzen Komodo 10.2 Linux popcnt seems to be
30 Elo better than asmfish Linux popcnt Nov.04 
on testpositions ?! Can it be ?

Also, stockfish,asmfish take the move with best eval from all threads,
but this looks buggy :

I can't easily find it in stockfish's search.cpp now,
but in CFish it's :

=========================

// Check if there are threads with a better score than main thread
Pos *bestThread = pos;

if ( !mainThread.easyMovePlayed
&& option_value(OPT_MULTI_PV) == 1
&& !Limits.depth
// && !Skill(option_value(OPT_SKILL_LEVEL)).enabled()
&& pos->rootMoves->move[0].pv[0] != 0)

{
for (size_t idx = 1; idx < Threads.num_threads; idx++) {
Pos *p = Threads.pos[idx];
if ( p->completedDepth > bestThread->completedDepth
&& p->rootMoves->move[0].score > bestThread->rootMoves->move[0].score)
bestThread = p;
}
}

==============================

for thread=1 to threads
if score(thread)>maxscore and depth(thread)>maxdepth then maxscore=score(thread)
next thread

for example if we have 4 threads with depths 19,17,20,19 and scores 40,63,37,48
then the move with score 48 at depth 19 from thread 4 is picked , while I assume
they wanted 37 to be picked.

However, in my testgames this "bug" gave better Elo and even better (3 Elo)
if they would pick the move with score 63 at depth 17 from thread 3. .

But in my testpositions it was vice versa and the best was to always pick
the move from thread 1 ("master"-thread) and ignore the others.
This gave ~3 Elo.

Now, with 13 threads on my Ryzen I get ~20 Elo for this in testpositions !
I haven't yet tested what this gives in game-playing, maybe someone
with a suitable GUI can test it , when I upload the FASM-executable
This also prints the values from all threads before bestmove is printed,
which I consider useful.

---I could still have an error somewhere---

================================================
actual (2017-09-11) code-snippet from
https://github.com/syzygy1/Cfish/

------------search.c , 2017/09/11, line 309----------------------

  // Check if there are threads with a better score than main thread
  Pos *bestThread = pos;
  if (   !mainThread.easyMovePlayed
      &&  option_value(OPT_MULTI_PV) == 1
      && !Limits.depth
//      && !Skill(option_value(OPT_SKILL_LEVEL)).enabled()
      &&  pos->rootMoves->move[0].pv[0] != 0)
  {
    for (int idx = 1; idx < Threads.num_threads; idx++) {
      Pos *p = Threads.pos[idx];
      Depth depthDiff = p->completedDepth - bestThread->completedDepth;
      Value scoreDiff = p->rootMoves->move[0].score - bestThread->rootMoves->move[0].score;
      // Select the thread with the best score, always if it is a mate
      if (    scoreDiff > 0
          && (depthDiff >= 0 || p->rootMoves->move[0].score >= VALUE_MATE_IN_MAX_PLY))
        bestThread = p;
    }
  }

  mainThread.previousScore = bestThread->rootMoves->move[0].score;

[... then print bestmove and ponder and exit]

================================================
Parent - - By user923005 (****) [us] Date 2017-09-12 19:26
For all of these, I doubt if there are any better alternatives.  For instance, for the first one, Qc7+ really only draws.  But the same is true for all the other best alternatives:
8/7k/6p1/5rPp/7P/2Q5/5p2/5K2 w - - acd 92; acs 120; am Qc7+; bm Qc7+; ce 502; pm Qc7+;
pv Qc7+ Kg8 Qb7 Rf4 Qc6 Kg7 Qc8 Rf5 Qe8 Rf7 Qe5+ Kh7 Qg3 Rf8 Qg2 Kg7 Qb7+ Rf7 Qb3 Rf5 Qh3 Rf4 Qd7+ Rf7 Qa4 Kh7 Qc4 Kg7 Qa2 Kf8 Qa4 Kg7 Qb4 Kh7 Qa5 Kg7 Qb5 Rf8 Qb1 Kh7 Qd1 Rf7 Qc1 Kg7 Qb1 Kh7 Qb4 Rf5 Qe4 Rf7 Qc2 Kg7 Qb2+ Kg8 Qe2 Kg7 Qc2 Kh7 Qc6 Kg7 Qe8 Rf8 Qe5+ Kh7 Qh2 Rf7 Qh1 Rf5 Qh3 Kg8 Qa3 Kg7 Qg3 Kg8 Qh3 Kh8 Qc3+ Kh7 Qc6 Rf7 Qb5 Rf8 Qc4 Kg7 Qe6 Kh7 Qe7+ Kg8 Qd6 Kg7 Qd4+ Kg8 Qb6 Kg7 Qc5 Kg8 Qa5 Rf4;
r3k1r1/4pp2/p2p1b2/qp1P1N1R/2n1P1n1/6B1/P3BPP1/1R1Q1K2 w q - acd 36; acs 900; am Bxg4; bm Bxg4; ce 415; pm Bxg4; pv Bxg4;

8/8/R3bk2/P1r2p1p/5KpP/4NP2/6P1/8 w - - acd 32; acs 123; am Ra8; bm Ra8; ce 149; pm Ra8; pv Ra8;

8/6p1/1bR2pk1/4p2p/4N2P/5PP1/6K1/1r6 b - - acd 39; acs 120; am Be3; bm Be3; ce 109; pm Be3;
pv Be3 Rc3 Rb2+ Kf1 Bb6 Rc6 Rb1+ Kg2 Bd4 Rc7 Rb2+ Kf1 Rb3 Kg2 Rb1 Kh3 Rg1 Rc2 Rh1+ Kg2 Re1 Rc6 Re3 Rc7 Re2+ Kf1 Rh2 Rc6 Rh1+ Kg2 Ra1 Kh3 Ra3 Kg2 Ra2+ Kf1 Ra1+ Kg2 Be3 Rc2 Kf5 Rc7 Ra2+ Kh1 Ra7 Rc6 Kg6 Kg2 Ra2+ Kf1;
8/1P1npkbp/2Np2p1/4p3/4P3/2K3P1/r6P/2B1R3 w - - acd 32; acs 120; am Bg5; bm Bg5; ce 89; pm Bg5;
pv Bg5 Ra6 Rf1+ Ke8 Nxe7 Bh6 Bf6 Rb6 Nc8 Rc6+ Kb2 Bf8 Rf3 Rc4 Rb3 Rxe4 Nb6 Nb8 Kc2 d5 Bg5 Bd6 Rb5 Rb4 Rxb4 Bxb4 Nxd5 Bd6 Nf6+ Ke7 Kd3 Nc6 Nxh7+ Kf7 Ke4 Na5 Be3 Nxb7 Ng5+ Kf6 Kd5 Be7 Ne4+ Kf7 Kxe5 Na5 Kf4;
1r4k1/r4p1q/4p1p1/1p1bN2n/3P1P2/b4B1P/2RB1Q1K/6R1 b - - acd 35; acs 900; am Bxf3; bm Bxf3; ce 85; pm Bxf3; pv Bxf3;

4rk2/PR3p1p/4p1p1/2K5/1N6/5P2/6PP/r7 w - - acd 34; acs 120; am Kb6; bm Kb6; ce 74; pm Kb6;
pv Kb6 Rb1 Kb5 Ra1 Na6 Rb1+ Kc6 Rc1+ Kd6 Ra1 Nc7 Rd8+ Ke5 Ra5+ Kf4 Rd2 g4 Rxh2 Rb8+ Ke7 a8=Q Rxa8 Nxa8 h5 Rb7+ Kf8 Nb6 hxg4 Nd7+ Kg7 fxg4 Rf2+ Kg5 Re2 Ra7 Re4 Ra5 Rb4 Rc5 Rb1 Rc7 Re1 Nc5 Kf8 Nd3 Rf1 Nb4 Rf3 Ra7 Rc3 Kf6 Rf3+ Ke5 Re3+ Kf4 Rc3;
8/5ppk/1R3n1p/2r2P2/1p4P1/5N1P/7K/8 b - - acd 31; acs 120; am Rc3; bm Rc3; ce 45; pm Rc3;
pv Rc3 Ne5 b3 Nxf7 Nd7 Rb5 Rc2+ Kg3 b2 g5 Kg8 g6 Kf8 Kf4 Nc5 Ne5 Ke7 Rb6 Re2 f6+ gxf6 g7 Rg2 Kf3 Rxg7 Nc4 Nd3 Nxb2 Ne5+ Kf2 Rg5 Nd1 Rf5+ Kg2 Nd7 Ra6 Re5 Kf2 Nc5 Ra7+ Ke6 Ne3 Ne4+ Kg2 h5 Ra6+ Kf7 Nc4 Rc5 Nd6+ Nxd6 Rxd6 Rc3;
r4rk1/5p1p/p1pb2q1/3n2p1/Pp1PR3/1BP3Pb/1P1N1P1P/R1BQ2K1 w - - acd 32; acs 120; am cxb4; bm cxb4; ce 18; pm cxb4;
pv cxb4 Nxb4 Re3 Nd5 Bxd5 cxd5 Nf3 f6 Ne1 f5 Qd3 Qh5 Qe2 Qxe2 Rxe2 f4 b3 Rae8 Rxe8 Rxe8 Bd2 Re2 Bc3 fxg3 hxg3 Kg7 Nd3 Bf5 Nc5 Rc2 Bb4 Bxc5 Bxc5 Rb2 b4 Kf6;
8/5bk1/r3p1p1/P1pp3p/2P2P1P/R5P1/1r6/4RNK1 b - - acd 37; acs 120; am Kf8; bm Kf8; ce 16; pm Kf8; pv Kf8;

3r1rk1/3q1pb1/2bpp1p1/1np4P/1p1nPN2/3PN1P1/1PP2QBK/R1B2R2 b - - acd 37; acs 900; am g5; bm g5; ce 8; pm g5;
pv g5 h6 Be5 Nh3 f5 exf5 exf5 Bxc6 Nxc6 Nxg5 f4 Qg2 Kh8 gxf4 Rxf4 Rxf4 Bxf4+ Kh1 Rg8 Qh3 Qxh3+ Nxh3 Bxh6 Nd5 Bxc1 Rxc1 Rg4 Rg1 Rxg1+ Kxg1 Ncd4 Ne3 Kg7 Nf4 Kf6 Kf2 Ke5 Ng6+ Ke6 Nf4+ Kd7 Nfd5 Nc7 Nxc7 Kxc7 Kf1 Kc6 c3;
2r1b1k1/R4ppp/4p3/1pp1P3/4p3/1PP1P3/4BKPP/8 b - - acd 41; acs 120; am g5; bm g5; ce 8; pm g5;
pv g5 c4 bxc4 bxc4 Kg7 Bh5 Kf8 Bd1 Rb8 Rc7 Rb1 Bh5 Rb2+ Kf1 Rb3 Kf2 Ra3 Be2 Ra1 Bh5 Ra4 Be2 Ra8 Kf1 Ra2 Bh5 Ra5 Kf2 h6 Be2 Ra1 Bh5 Ra2+ Kg1 Ra8 Be2 Kg7 Bh5 Ra3 Kf2 Kf8 Be2 Ra2 Kf1 Kg7;
2b3k1/2n4q/3p1b2/p1pP2p1/Pp2P1p1/1P4P1/2QR3P/2B3NK w - - acd 42; acs 900; am Rd1; bm Rd1; ce 7; pm Rd1;
pv Rd1 Qg6;
7r/5kp1/8/5rP1/P1ppR3/5PK1/8/2R5 w - - acd 43; acs 120; am Kg4; bm Kg4; ce 7; pm Kg4;
pv Kg4 Rd5 Rxc4 Rhd8 Rc2 R8d6 Ra2 d3 Rd2 Rg6 f4 Re6 Kf3 Rxe4 Kxe4 Ra5 Ra2 d2 Rxd2 Rxa4+ Kf5 g6+ Ke5 Ra5+ Rd5 Ra7 Ke4 Re7+ Kf3 Ra7 Rd4 Ra3+ Ke4 Ra5 Rb4 Ra6 Rb7+ Ke8 Ke5 Kf8 Rd7 Ra4 Rd4 Ra5+ Ke6 Ra1 Rd6 Kg7 Rd7+ Kf8 Rd3 Kg7 Rd5;
r4r1k/ppbq3p/2p5/P2ppP2/1PP5/3P2nP/2Q3PN/R1BR2K1 w - - acd 33; acs 120; am f6; bm f6; ce 0; pm f6;
pv f6 Rxf6 cxd5 cxd5 Ng4 Rf5 Bh6 Bd6 Re1 Bxb4 Rxe5 Rc8 Qa4 Qxa4 Rxa4 Bc5+ Kh2 Rxe5 Nxe5 Nf1+ Kh1 Ng3+ Kh2;
2rq1rk1/pp2ppbp/2n3p1/3pNb2/3PnB2/2N1P1P1/PP3PBP/R2Q1RK1 w - - acd 35; acs 120; am Nxc6; bm Nxc6; ce 0; pm Nxc6;
pv Nxc6 Rxc6 Qb3 Nxc3 bxc3 e6 Qxb7 Qb6 Qxb6 axb6 Rfc1 Rfc8 a4 Bf6 h4 Rxc3 Rxc3 Rxc3 a5 bxa5 Rxa5 h5 Ra7 Rc1+ Kh2 Rc2 Ra8+ Kh7 Ra7 Kg8;
rr4k1/6b1/2p1p2B/P2b3q/2p3N1/R4P1p/1P4PQ/4R2K w - - acd 35; acs 120; am gxh3; bm gxh3; ce 0; pm gxh3;
pv gxh3 Qh4 Rg1 Rxb2 Qf4 Qxh3+ Nh2 Rxh2+ Qxh2 Bxf3+ Rxf3 Qxf3+ Qg2 Qh5+ Qh2 Qd5+ Qg2 Rxa5 Qxd5 Rxd5 Rxg7+ Kh8 Kg2 Rh5 Rg4 Rxh6 Rxc4;
k2r4/1pb5/p1nq4/3Bp1r1/2PpR1pN/PQ3pP1/1P3P1P/4R1K1 b - - acd 36; acs 120; am Rf8; bm Rf8; ce 0; pm Rf8;
pv Rf8 c5 Qxc5 Be6 d3 Qxd3 Bb6 Rf1 Rd8 Qc3 Rd4 Qxc5 Bxc5 Rfe1 Rd2 Rf1 Rd4;
2r5/r5k1/1q1pQ1p1/1P2p2p/PR2Pp2/7P/5bPB/3R3K b - - acd 37; acs 120; am Rc2; bm Rc2; ce 0; pm Rc2;
pv Rc2 Rbb1 Bd4 Bg1 Rxa4 Qe7+ Kh6 h4 Rc7 Qf6 Rg7 Rbc1 Qb7 Bxd4 Rxd4 Rxd4 exd4 Qxf4+ Kh7 Rc6 d3 Rxd6 Rd7 Kh2 Rxd6 Qxd6 Qxe4 Qd7+ Kh6 Qd8 Qf4+ Kg1 Qe3+ Kh2 Kg7 Qc7+ Kf6 Qd8+ Kg7;
3b1r1k/2q4p/b1Bp2p1/2nPprP1/1pN3RP/pP2B3/P1P5/2K1R2Q w - - acd 37; acs 120; am Qg1; bm Qg1; ce 0; pm Qg1;
pv Qg1 Qg7 Nd2 Ba5 Kd1 Rd8 Bxc5 dxc5 Qxc5 e4 Rgxe4 Qa1+ Nb1 Qxb1+ Kd2 Qb2 Re8+ Rxe8 Rxe8+ Kg7 Re7+ Kg8 Re8+ Kg7;
5k2/R7/1p1P2p1/7p/2PK4/8/4r3/8 b - - acd 39; acs 120; am g5; bm g5; ce 0; pm g5;
pv g5 Rh7 Rd2+ Ke5 g4 Ke6 Re2+ Kd5 g3 Rxh5 Rc2 Rg5 g2 d7 Ke7 Rg7+ Kd8 Kd6 Rd2+ Kc6 Rd4 c5 Rd2 Kxb6 Rf2 c6 Rb2+ Kc5 Rc2+ Kb5 Rb2+ Ka5 Kc7 Rg8 Ra2+ Kb4 Rd2 Kc5 Rc2+ Kb5 Rb2+ Ka5;
6R1/2p5/3b1p2/p2k1P2/r1p5/4B1PP/4KP2/8 w - - acd 40; acs 120; am h4; bm h4; ce 0; pm h4;
pv h4 Ra2+ Kf3 c3 h5 c2 Bd2 Ra1 h6 Rh1 Kg2 Rd1 Be3 Bc5 Bf4 Bd6 Be3;
8/p3r2k/P2p2b1/1PpP3p/6p1/3p2P1/3Q4/5K2 b - - acd 41; acs 209; am Re2; bm Re2; ce 0; pm Re2;
pv Re2 Qg5 Rh2 Qe7+ Kh6 Qf8+ Kh7 Qe7+;
1r5k/3R1p1P/5K2/1p3P1R/pp6/1P4r1/P7/8 b - - acd 42; acs 120; am Rg7; bm Rg7; ce 0; pm Rg7;
pv Rg7 Rd2 Rb6+ Ke5 f6+ Kf4 Rb8 Rh6 Rf8 Rhh2 Rxh7 Rxh7+ Kxh7 Rd5 Rb8 Ke3 Kh6 Kd4 Rc8 Rxb5 a3 Rxb4 Rc2 Ra4 Rxa2 Kc3 Kg5 b4 Kxf5 Kb3 Rb2+ Kxa3;
8/8/2n2kp1/p4p1p/2K2N1P/1P6/5PP1/8 b - - acd 42; acs 120; am g5; bm g5; ce 0; pm g5;
pv g5 Nh3 gxh4 Kc5 Nb4 Kb5 Ke5 Kxa5 Nd5 b4 Ke4 b5 Kd3 Ka6 Nc7+ Kb6 Nxb5 Kxb5 Ke2 f3 Kf1 Nf4 Kf2 Kb6 Kg3 Ng6 Kxg2 Nxh4+;
1k1r4/pPp3q1/P2b3p/6p1/1PQpR3/1K1Bp2P/8/8 b - - acd 46; acs 209; am Bg3; bm Bg3; ce 0; pm Bg3;
pv Bg3 Qd5 Qf8 Qb5 Bd6 Rxd4 Qf3 Re4 h5 Qxg5 Be7 Qb5 Bh4 Kc2 Qg2+ Kb3 Qf3;
5rk1/4R2p/1p6/1p3pp1/1P2n3/R2NP1P1/P2r1PKP/8 w - - acd 46; acs 900; am Raa7; bm Raa7; ce 0; pm Raa7;
pv Raa7 Rd8 g4 R2xd3 Rg7+ Kh8 Rxh7+ Kg8 Rag7+ Kf8 Rb7 Kg8 Rhg7+ Kh8 Rh7+ Kg8;
8/2k5/2n1P2p/3K2p1/2P3P1/p5P1/N7/8 w - - acd 49; acs 120; am c5; bm c5; ce 0; pm c5;
pv c5 Ne7+ Kc4 Nc6 Nc1 Kd8 Kb3 Ke7 Kxa3 Kxe6 Ka4 Ne5 Ka5 Nxg4 c6 Ne5 c7 Kd7 Kb6 Nc4+ Kb7 Na5+ Kb8 Nc6+ Kb7 Na5+;
3Qb3/5p1k/4r1p1/1p6/p2p4/P1r1nB1P/6P1/4R1K1 b - - acd 50; acs 900; am Rxa3; bm Rxa3; ce 0; pm Rxa3;
pv Rxa3 Bg4 Re4 Bf3 Re5 Qxd4 Re6 Be4 Nf5 Qd2 Nd6 Bc2 Nc4 Qb4 Rae3 Rxe3 Nxe3 Bd3 Nc4 g4 Re3 Bxc4 Re4 Kf2 Rxc4 Qe7 Rc2+ Kg3 Rc8 h4 Ra8 Qe4 Rc8 Qe7;
8/8/5k2/p1K3p1/P5r1/1P1PR3/8/8 b - - acd 51; acs 209; am Rf4; bm Rf4; ce 0; pm Rf4;
pv Rf4 d4 g4 Rg3 Kf5 d5 Rf3 Rg1 g3 Kc6 Rc3+ Kd7 Kf4 d6 Kf3 Kd8 g2 d7 Kf2 Rd1 g1=Q Rxg1 Kxg1 Ke7 Re3+ Kf7 Rd3 Ke7 Re3+;
8/1p6/p3rpkp/P2R4/1P3K1P/4P3/8/8 b - - acd 54; acs 120; am h5; bm h5; ce 0; pm h5;
pv h5 Rd1 Rc6 Kf3 Re6 Rg1+ Kf5 Rg7 b6 axb6 Rxb6 e4+ Ke6 Rh7 Rb5 Kf4 a5 bxa5 Rxa5 Rb7 Rc5 Rb6+ Kf7 Rb2 Rc7 Kf5 Rc5+ Kf4;
6k1/1p3p2/p4P2/2PbN3/7p/P3K2P/8/8 b - - acd 64; acs 209; am Be6; bm Be6; ce 0; pm Be6;
pv Be6 Kf4 Bxh3 Kg5 Be6 Kxh4 Kf8 Kg5 Ke8 Kh6 Kf8 Nf3 Bd5 Nd4 Ke8 Kg7 Kd7 Nf5 a5 a4 Bb3 Nd6 Bxa4 Nxf7 Bb3 Ne5+ Ke8 Nf3 a4 Nd4 Bc4 c6 bxc6 Nxc6;
4k3/4p3/3q2r1/3P3p/P1p2P2/2Nb4/1P3Q2/2K1R3 b - - acd 32; acs 120; am Kf7; bm Kf7; ce -7; pm Kf7;
pv Kf7 f5;
2rqkb1r/pp1b1p2/2n1p3/3p2pP/3P4/2N1PN2/PPQ1BPP1/R1R3K1 b k - acd 30; acs 120; am Rxh5; bm Rxh5; ce -15; pm Rxh5;
pv Rxh5 g4 Rh6 Kg2 e5 dxe5 Bxg4 Rh1 Rxh1 Rxh1 Bg7 Qh7 Kf8 Nd4 Bd7 Rd1 Nxe5 Nxd5 Ng6 Bh5 Rc5 Nc3 Qb6 Bxg6 fxg6 Kg1 Qxb2 Ne4 Rc1 Rxc1 Qxc1+ Kg2;
8/4k3/3Rp3/5p2/5B2/2qp1B1P/6K1/8 w - - acd 28; acs 120; am Rb6; bm Rb6; ce -35; pm Rb6;
pv Rb6 e5 Bg5+ Kd7 Rb7+ Kc8 Rf7 Qc2+ Kg3 d2 Rf8+ Kc7 Bd8+ Kd7 Bg5 e4 Rd8+ Kc7 Rxd2 Qb3 Rf2 Kd6 h4 exf3 Rxf3 Qb1 Kf2 Qb2+ Kf1 Qa1+ Kf2 Ke6 h5 Qd4+ Be3 Qh4+ Ke2 Kf6 h6 Kg6 h7 Qh1 Rf2;
2Q5/3brpkp/6p1/1p2q3/1P1bp3/1B6/4RPPP/1R5K w - - acd 34; acs 48; am Qd8; bm Qd8; ce -43; pm Qd8;
pv Qd8 Qf6 Bd5 Bf5 Rf1 e3 fxe3 Bb6 Qa8 Ra7 Qc6 Qxc6 Bxc6 Bd3 Ree1 Rc7 Be8 Re7 Bc6 Bxf1 Rxf1 Bxe3 g3 Bd2 Rd1 Bxb4 Rd7 Re1+ Kg2 Re2+ Kh1 Be7 Bxb5 Re3 Kg2 Kf6 Rd3 Re2+ Kh3 Rf2 Rd1 h5 Rf1 Rxf1 Bxf1 Ke5 Bb5 Bc5 Bc4;
r3r3/5qkp/1np2pp1/p2np3/P7/B2Q2P1/P1P2PBP/3RR1K1 b - - acd 36; acs 209; am Nxa4; bm Nxa4; ce -53; pm Nxa4; pv Nxa4;

5rk1/ppR2bpp/8/8/5p2/PPB1rP1P/1KQR2P1/5q2 b - - acd 37; acs 209; am h6; bm h6; ce -60; pm h6;
pv h6 a4 Qa6 Bd4 Re1 Qf5 Re2 Rxe2 Qxe2+ Rc2 Qe8 Qxf4 Bxb3 Qg3 Qd7 Bxg7 Bxc2 Bxh6+ Kh7 Bxf8 Bxa4 Qf4 Kg8 Bc5 Qb5+ Bb4 Qd5 Qe3 a5 Be1 Bc6 h4 Kf7 Bc3 Qb5+ Kc2 a4 Qf4+ Ke8 Qe5+ Qxe5 Bxe5 Kf7 h5 Bb5 Bb2 Bf1 g4;
1kq1r3/bpn2p2/p1prp1p1/B4p2/B1PP1P1P/PR3QP1/1P6/1K1R4 b - - acd 32; acs 120; am Qd7; bm Qd7; ce -63; pm Qd7;
pv Qd7 Rbd3 Qc8 R1d2 b5 Bb3 Qb7 h5 gxh5 Qxh5 Rd7 g4 fxg4 c5 Red8 Qe5 Ka8 f5 Bb8 Qf6 Re8 fxe6 fxe6 Bxc7 Qxc7 Qg6 Rde7 Qxg4 Kb7 Rf3 Rf7 Rxf7 Qxf7 Re2 Rg8 Qxe6 Rg1+ Kc2 Qf1 Qf7+ Qxf7 Bxf7 Rf1 Re7+ Kc8;
2r3b1/8/prN1pp1k/N5pn/p1P5/5PP1/1P1R1K2/7R b - - acd 39; acs 210; am Rcxc6; bm Rcxc6; ce -64; pm Rcxc6;
pv Rcxc6 Nxc6 Rxc6 g4 Rxc4 Rxh5+ Kg6 Rh8 Kg7 Rh1 Kg6 Kg3 e5 Rd8 Bf7 Rdh8 Kg7 R1h7+ Kg6 Rh2 Kg7 Ra8 Rc6 Ra7 Kg6 Rh8 Rb6 Rc8 Be6 Rc5 Ba2 Rcc7 f5 Rab7 f4+ Kf2 Rxb7 Rxb7 Bc4 Re7 Kf6 Re8 Bd3 Rd8 e4 fxe4 Bxe4 Rd6+ Kf7 Rxa6 Bc2 Ra5 Kg6 Ra8 Kg7;
1r5r/1k2p3/1B1p3R/pP1Pbqp1/P1P1R3/5P2/6K1/4Q3 w - - acd 32; acs 120; am Rxe5; bm Rxe5; ce -67; pm Rxe5;
pv Rxe5 dxe5 Qxa5 Qc2+ Kg3 Ra8 Rxh8 Rxa5 Bxa5 Qxa4 Bd2 Qd1 Be3 e4 Rf8 exf3 Rxf3 Qe1+ Kh3 Qh1+ Kg3 Qh4+ Kg2 Qxc4 Bxg5 Qa2+ Kg3 Qxd5 Be3 Qe4 Kf2 Qh4+ Ke2 Qh5 Kf2 e5 Rf8 Kc7 Kf1 Qh1+ Kf2 Qh4+ Kf3 Qh5+ Ke4 Qg6+ Kf3 Kb7 Ke2 Qc2+ Kf1 Qb3 Kf2;
5rk1/p4pbp/qr1pn1p1/3Q4/6P1/2NP2B1/PP5P/1R3R1K w - - acd 35; acs 209; am Rbe1; bm Rbe1; ce -67; pm Rbe1; pv Rbe1;

2b2rk1/2p2r1p/p2p2p1/P2Pp1q1/1NP1P3/3Q3P/4RBP1/2b2R1K w - - acd 32; acs 209; am c5; bm c5; ce -69; pm c5;
pv c5 Qh6;
8/8/p2k2p1/4pp2/P3P1Pp/3KNPnP/8/8 b - - acd 44; acs 209; am Nh1; bm Nh1; ce -69; pm Nh1;
pv Nh1 exf5 Nf2+ Kd2 gxf5 Nxf5+ Kd5 Ke3 Nxh3 a5 Nf4 Nxh4 Ne6 Nf5 Nc5 Ne7+ Ke6 Nc6 Nb7 Nb4 Nxa5 Nxa6 Nb3 Nb4 Nd4 Ke4 Ne2 Na2 Kf6 Ke3 Nf4 Nc3 Ne6 Kd3 Ke7 Ne4 Nf4+ Kc4 Ke6 Kc3 Ne2+ Kd3 Nf4+ Kd2 Ke7 Ke3 Ne6 Nd2 Ng5 Nc4 Kf6 Nd6;
3b2k1/7p/2bpq1p1/1p1Np3/2r1P3/P1N2RP1/1P2Q2P/6K1 b - - acd 33; acs 209; am Qc8; bm Qc8; ce -75; pm Qc8; pv Qc8;

r1b2r1k/1p5p/1q2p2p/p2p3Q/P2n1N2/6P1/1P3PB1/2R1R1K1 b - - acd 36; acs 209; am Nf5; bm Nf5; ce -79; pm Nf5;
pv Nf5 g4;
r2r2k1/1p1qbpp1/1Pbp1n2/p3pPB1/P3P3/2NQ1B2/2P4P/3R1R1K b - - acd 29; acs 120; am Nh7; bm Nh7; ce -81; pm Nh7;
pv Nh7 Bc1;
r4rk1/2qn1pp1/p1b1p2p/1p6/3R1PnB/1BN3Q1/PPP4P/2KR4 b - - acd 33; acs 210; am h5; bm h5; ce -85; pm h5;
pv h5 h3 Nh6 Rxd7 Bxd7 Bf6 Nf5 Qg5 Qa7 Bxg7 Nxg7 Rg1 Qe3+ Kb1 Qxg1+ Qxg1 Bc6 f5 exf5 Qb6 Bg2 a3 Rae8 Qxa6 b4 Nd5 Bxd5 Bxd5 Rd8 Bb3 Rd1+ Ka2 bxa3 bxa3 Rfd8 Qf6 R8d7 Kb2 Kf8 a4 R1d6 Qc3 f4 a5 Ne6 Qh8+ Ke7 Qxh5 Ke8 Qf5 Kd8;
8/7p/5kp1/2n1pP2/1p2P1P1/3BK2P/8/8 w - - acd 49; acs 209; am h4; bm h4; ce -87; pm h4;
pv h4 g5;
6r1/7p/B7/1P3p1p/5P1P/1k1p3K/1N3R2/6r1 w - - acd 37; acs 209; am Kh2; bm Kh2; ce -91; pm Kh2; pv Kh2;

rr4k1/2q1ppb1/b2p2p1/2pP2Nn/2P1P3/2N3P1/P1Q2PBP/R3R1K1 b - - acd 33; acs 209; am Qb6; bm Qb6; ce -96; pm Qb6;
pv Qb6 e5 Qb2 Qxb2 Rxb2 exd6 exd6 Nce4 Bd4 Nf3 Bxc4 Nxd4 cxd4 Nxd6 Rbxa2 Rxa2 Bxa2 Ra1 d3 Nc4 Rc8 Ne3 d2 Rxa2 Rc1+ Bf1 d1=Q Nxd1 Rxd1 d6 Rxd6 f3 Rd1 Kf2 Kg7 Rc2 Nf6 Be2 Rd7 f4 Ne4+ Ke3 f5 Rc6 Rd2;
r3nrk1/pp1n3p/1bpp2p1/4pP2/2P1P1Pq/6NP/PPQB2K1/3RRB2 b - - acd 29; acs 209; am Ng7; bm Ng7; ce -97; pm Ng7; pv Ng7;

2r2r2/1p2pqb1/p2p3k/5pn1/2PPbNp1/4P1P1/1B3PB1/R1R1Q1K1 w - - acd 30; acs 120; am Bxe4; bm Bxe4; ce -108; pm Bxe4; pv Bxe4;

8/7k/2Pp4/q2Pp2p/4Pp1p/5PrP/1r4PK/2N2QR1 w - - acd 44; acs 210; am Qd1; bm Qd1; ce -108; pm Qd1;
pv Qd1 Qd2 Qxd2 Rxd2 Nb3 Rc2 Na5 Kg8 Nb7 Rg6 Nd8 Kg7 Nb7 Kf7 Nd8+ Ke7 Ne6 Rg8 Kh1 Rg3 Nc7 Rb2 Ne6 Rg8 Nc7 Ra2 Ne6 Rc2 Kh2 Rg6 Kh1 Kf7 Kh2 Rg8 Kh1 Rg3 Kh2 Ke7 Kh1 Kf6 Nd8 Rg8 Ne6 Rb8 Nc7 Ke7 Ra1 Kf7;
6r1/1k5p/p1p1r1q1/Pp1nP1p1/1P1P1p2/5Q1P/3BR1P1/2R3K1 w - - acd 34; acs 209; am Qe4; bm Qe4; ce -112; pm Qe4;
pv Qe4 g4;
7k/6b1/p3q2p/6p1/6n1/5N1B/P3PP2/1R2R1K1 b - - acd 33; acs 209; am Qe4; bm Qe4; ce -118; pm Qe4;
pv Qe4 Rb8+ Kh7 Reb1 a5 R1b6 Bd4 Nxd4 Qxd4 e3 Qd1+ Bf1 Qe1 R8b7+ Kg8 Rb2 Qc1 R7b6 Ne5 Kg2 Nd7 Rb7 g4 Be2 Qc6+ Kg1 Qc1+ Kh2 g3+ fxg3 Nc5 R7b6 Nd7 R6b5 Qxe3 Rxa5 Kh7 Bg4;
2krq2r/pp3pp1/2b5/2bR1nPp/2P5/P3PN1P/5PB1/2BQ1RK1 b - - acd 31; acs 120; am Bxd5; bm Bxd5; ce -122; pm Bxd5;
pv Bxd5 cxd5 Qe4 Bb2 Rxd5 Nd4 Nxe3 Bxe4 Rxg5+ Kh1 Nxd1 Rxd1 Bxd4 Bxd4 Re8 Rc1+ Kb8 Be3 Rb5 Bf4+ Ka8 Rc4 g5 a4 Rb2 Be3 Kb8 Bd5 f6 Bf7 Rd8 Bd4 Rxd4 Rxd4 Rxf2 Bxh5 Kc7 Bg6 Rf3 Kg2 Ra3 Bf5 Kc6 Be4+ Kb6 Rb4+ Kc5 Rxb7;
4r3/2p1k3/1pPp3q/1P1B1p2/5P1P/6R1/PQ5K/4r3 w - - acd 39; acs 210; am Kh3; bm Kh3; ce -122; pm Kh3;
pv Kh3 Kd8;
3r1k2/1pr2p2/3N2p1/1PP1bb1p/4p3/4B2P/1n3PP1/1RR2BK1 b - - acd 34; acs 209; am Nd3; bm Nd3; ce -123; pm Nd3;
pv Nd3 b6 Rcd7 Nxb7 Rxb7 Rc4 Bd7 c6 Bxc6 Rxc6 Rd5 g3 Ke7 Rb3 Kd7 Rc4 f5 f3 Nb2 Rcb4 Nd1 fxe4 fxe4 Bc1 e3 Ba6 Rb8 Re4 Bd6 Be2 Rc8 Ba3 Nc3 Rxe3 Nxe2+ Rxe2 Bxa3 Rxa3 Rb8 Ra7+ Kd6 b7;
2rq2k1/pp1b1ppp/8/b1p2PPn/3P4/2NBQ2P/PP2N3/4R1K1 w - - acd 33; acs 210; am Rd1; bm Rd1; ce -124; pm Rd1; pv Rd1;

2kr3r/bb3ppp/p3p3/P3n3/1p2P3/3q2P1/1P1N1PBP/R1BQR1K1 w - - acd 34; acs 209; am Qh5; bm Qh5; ce -124; pm Qh5;
pv Qh5 Qb5;
2r3kb/1b3p1p/1pq2PpP/2ppP3/8/2B1PN2/6P1/R4QK1 b - - acd 32; acs 120; am b5; bm b5; ce -125; pm b5;
pv b5 Bb2 d4 exd4 c4 Bc3 Qb6 Ng5 Bd5 Ra5 Qb7 Qf4 Re8 Bb4 Qb6 Ra1 Qc6 Re1 Qc7 Ne4 Bxe4 Rxe4 Qd7 Re1 Ra8 Qe4 Ra4 Bc5 Qe8 Qf4 Qe6 Rb1 Qf5 Qxf5 gxf5 Rxb5;
7R/8/2k1pp2/2n2n2/1p5N/8/6PP/2K5 b - - acd 29; acs 209; am Nd6; bm Nd6; ce -130; pm Nd6;
pv Nd6 Nf3 e5 h4 Nd3+ Kd1 e4 Nd2 Ne5 h5 f5 Ke2 Ng4 g3 Kd7 Nb3 Nf6 Nd4 b3 Kd2 e3+ Kd3 b2 Kc2 Ke7 h6 Nf7 Rb8 Nxh6 Rxb2 Kf7 Kd3 Nh5;
b2r2k1/q1r1b1p1/1n2pp1p/p7/Pp1PP1N1/1P1B1N2/3Q2PP/3R1R1K b - - acd 33; acs 209; am Nd7; bm Nd7; ce -130; pm Nd7;
pv Nd7 Nxh6+ gxh6 Qxh6 Bf8 Qh3 Re8 e5 f5 Ng5 Nb6 Qh5 Rg7 Qxe8 Rxg5 Qxe6+ Qf7 Qxf7+ Kxf7 Rxf5+ Rxf5 Bxf5 Nd5 Re1 Nc3 Kg1 Bd5 Bc2 Be7 Kf2 Bg5 g3 Be6 Kf3 Bd5+ Kg4 Bd2 Rf1+ Ke7 Rf6 Be6+ Kf3 Nd5 Rg6 Be3 Bd3 Bxd4 Bc4 Kf7 Rxe6 Kxe6 Ke4 Bxe5 Bxd5+ Kf6 Bc4 Bc7 Kd5 Bb6 h4 Bc7 g4 Bg3 g5+ Kg7;
8/2k5/3p2p1/p1pPn2p/P1P1P2P/q5P1/2Q1BK2/8 w - - acd 36; acs 209; am Qd1; bm Qd1; ce -130; pm Qd1; pv Qd1;

r5k1/1q1bbprp/1p1ppp2/pP3P2/P1P1P2R/2NB2P1/4Q1KP/5R2 b - - acd 31; acs 120; am Qc7; bm Qc7; ce -139; pm Qc7;
pv Qc7 Rh5;
r2q3k/Nb4r1/4n1p1/2Bnp3/3p1p2/5P2/PP1QBRN1/2R3K1 w - - acd 29; acs 120; am Nb5; bm Nb5; ce -141; pm Nb5;
pv Nb5 Rh7;
3B2k1/1p6/p1p4p/1b1p2PP/3PrP2/1P3K2/3R4/8 b - - acd 44; acs 210; am Bf1; bm Bf1; ce -142; pm Bf1;
pv Bf1 g6 Bh3 Rd1 Bf5 Bc7 Re8 Ra1 Bc2 b4 Rc8 Be5 Be4+ Kg3 Bf5 Kf2 Bg4 Rg1 Bf5 Kg3 Re8 Kh4 Rc8 Re1 Re8 Ra1 Rf8 Ra2 Rc8 Re2 Re8 Rf2 Rf8 Kg3 Re8 Kf3 Rf8 Re2 Re8 Re1 Re6 Rd1 Re8 Rg1 Re6 Rc1 Re7;
8/q4p1k/3p1npp/2bPp3/1p2P1B1/1P1Q1PP1/r2B3P/3R3K w - - acd 38; acs 209; am Bh3; bm Bh3; ce -143; pm Bh3;
pv Bh3 Qa6 Qxa6 Rxa6 Bf1 Ra2 g4 g5 Bb5 Rb2 Ba4 Kg6 Be1 h5 h3 h4 Rd2 Rb1 Re2 Ng8 Kg2 Ne7 Bf2 Bxf2 Rxf2 Rc1 Ra2 Kg7 Bb5 Ng6 Kf2 Nf4 Bf1 Rb1 Ra6 Rb2+ Ke3 Rxb3+ Kf2 Rb2+ Ke3 b3 Rxd6 Rc2 Rb6 b2 d6 Kf6;
7k/6p1/3b1p1p/4pP1b/P2p4/1P6/3B4/K5R1 w - - acd 37; acs 209; am a5; bm a5; ce -150; pm a5;
pv a5 e4;
8/N1Pbkp2/1r6/p7/3KP3/2R3p1/6P1/8 b - - acd 32; acs 210; am Ke6; bm Ke6; ce -153; pm Ke6;
pv Ke6 c8=R;
k6r/1p6/p5p1/P1pPnp1r/2P3p1/2P3Pn/3BB1K1/4RR2 w - - acd 35; acs 209; am Bc1; bm Bc1; ce -156; pm Bc1; pv Bc1;

rn3rk1/pb1q1pb1/3p1n1p/1ppPP1p1/8/2N2NP1/PP3PBP/R1BQ1RK1 b - - acd 29; acs 120; am dxe5; bm dxe5; ce -158; pm dxe5;
pv dxe5 Nxe5;
6b1/3k3p/8/1p3P2/p2Pr1PR/2P4P/1P1K4/8 w - - acd 33; acs 120; am Rh6; bm Rh6; ce -162; pm Rh6;
pv Rh6 Bc4 Rxh7+ Kc6 Rh6+ Kc7 Rh7+ Kb6 f6 Re2+ Kc1 Rf2 Rh5 Rxf6 Rf5 Re6 Re5 Rxe5 dxe5 Kc6 h4 Kd5 Kd2 Kxe5 Ke3 Bd5 h5 Bf7 Kf3 Bb3 Kg3 Bd5 h6 Be4 g5 Bd3 Kg4 Bh7 Kf3 Bg6 Kg4;
b3r3/p3q1k1/Pp2rpp1/1Pp1n2p/2PpP3/5P1P/4BKP1/2QRRN2 w - - acd 35; acs 209; am Qd2; bm Qd2; ce -165; pm Qd2; pv Qd2;

3r2k1/pR3ppp/4p2n/q1p1P3/P1P3P1/7P/3r1PB1/R3Q1K1 b - - acd 35; acs 209; am Kf8; bm Kf8; ce -168; pm Kf8;
pv Kf8 Bf1 Qc3 Rd1 a5 Rxd2 Qxd2 Qxd2 Rxd2 Ra7 f5 exf6 gxf6 Rxa5 Ra2 Rxc5 Rxa4 Rc8+ Kf7 Rc7+ Kf8 c5 Ra1 Rc6 Rc1 Rxe6 Ng8 Rc6 Kf7 Rc7+ Ne7 Kg2 h6 Bd3 Rc3 Ba6 Ke6 f4 Kf7 h4 Ke8 g5 hxg5 hxg5 fxg5 fxg5;
qrb1kb1r/5p2/2p1p3/p1npP2p/3Q2pP/1PN3B1/P1P2PP1/1K1R1B1R w k - acd 34; acs 209; am Bf4; bm Bf4; ce -170; pm Bf4;
pv Bf4 a4;
6k1/8/3n4/pb1p1p1p/3PpPp1/1P4P1/1R1BP1KP/r4B2 w - - acd 39; acs 209; am Rc2; bm Rc2; ce -178; pm Rc2; pv Rc2;

1r6/4qpk1/8/3p3p/3P1bPP/R1P5/1r1N1P2/2QR1BK1 w - - acd 36; acs 210; am g5; bm g5; ce -181; pm g5;
pv g5 Qxa3;
2b2bk1/5rpp/4p1q1/2Pp4/3Np3/4P1PB/3Q1P1P/2R3K1 b - - acd 34; acs 120; am Qh6; bm Qh6; ce -191; pm Qh6;
pv Qh6 Kg2 Qf6 Rb1 g6 Rb6 Kg7 Bxe6 Bxc5 Bxf7 Bxb6 Bxd5 Bf5 Nxf5+ Qxf5 Kg1 Kh6 Bb7 Bc5 Qc2 Be7 Bxe4 Qe6 Kg2 Kg7 Qb2+ Qf6 Qb5 h5 h4 Kh6 Qc4 Bd8 Qd5 Be7 Qb5 Bd8 Bf3 Be7 Qc4 Bd6 Be4 Be7 Bd5 Bd8;
r5k1/p3bp2/2q2n1p/1r2p1p1/2N1P1P1/2pQ2BP/PP3P2/1K1RR3 w - - acd 36; acs 209; am b3; bm b3; ce -192; pm b3;
pv b3 Rc8;
k6r/pp3pb1/1npQb1q1/4n2p/4N1p1/1P4P1/P3PPBP/R2RN1K1 b - - acd 32; acs 209; am h4; bm h4; ce -193; pm h4;
pv h4 Qc7 h3 Rd8+ Rxd8 Qxd8+ Bc8 Nd6 Qf6 Qxf6 Bxf6 Be4 Nf3+ Bxf3 Bxa1 Nxc8 Nxc8 Bxg4 Ne7 Bxh3 Kb8 Kg2 Kc7 Bg4 Kd6 Nd3 Bc3 Bh5 Nd5 Bxf7 Nb4 Nf4 Ke7 Be6 Nxa2 Kf3 b5 h4 Nb4;
6k1/6b1/1R6/3Pp3/1p2PnNp/7r/P2B4/1K6 b - - acd 24; acs 140; am Kf7; bm Kf7; ce -195; pm Kf7;
pv Kf7 Bxf4 exf4 e5 Bf8 e6+ Kg7 d6 Rh1+ Kc2 Bxd6 Rxd6 Re1 Kd2 Re4 Kd3 Re1 Rb6 h3 Rxb4 Rxe6 Rxf4 Re1 Nh2 Rb1 Rf2 Kg6 Ke3 Kg5 Kf3 Kh4 Nf1 Rc1 Ne3 Kg5 Nd5;
r2r2k1/2qnbpp1/b4n1p/p2Pp1B1/1pp1P3/2P2NN1/PP4PP/1K1RQB1R w - - acd 27; acs 120; am Bxf6; bm Bxf6; ce -195; pm Bxf6;
pv Bxf6 Bxf6;
4r1k1/6p1/5np1/2p2bN1/ppP5/1Pbq1N1P/P2B1QP1/5R1K w - - acd 39; acs 209; am Re1; bm Re1; ce -196; pm Re1;
pv Re1 Rxe1+ Bxe1 Ne4 Nxe4 Bxe4 Ng5 Bf6 Bd2 axb3 axb3 Bxg5 Bxg5 Qxh3+ Kg1 Qxb3 Be3 Qd1+ Qf1 Qd6 Qe2 Bf5 Bc1 Qd4+ Qe3 Qd1+ Kh2 Qc2 Kg1 Be4 Qd2 Qxd2 Bxd2 Bd3 Kf2 Bxc4 Be3 b3 Bc1 Bd5 g3 Kf7 Ke3 g5 Kd3 g4 Kc3 g6 Ba3 c4 Kd4 Ke6 Bb2 Kd6 Bc1 Kc6;
4r1k1/1q2pp1p/1p1p2p1/p1rP1P2/3NP1n1/1P4P1/PR1Q2KP/5R2 b - - acd 34; acs 209; am Nf6; bm Nf6; ce -198; pm Nf6;
pv Nf6 Qh6 g5 Qxg5+ Kh8 Re2 Rg8 Qh6 Rgc8 Rfe1 a4 Nc6 b5 bxa4 bxa4 Qd2 Ng4 Qg5 Nf6 Qf4 Rc3 Qd2 R8xc6 dxc6 Qxc6 Kh3 Nd7 Rb1 a3 Qg5 Qc4 Qh5 Rc2 Rxc2 Qxc2;
8/1q3ppk/1P4b1/3pPp1p/3P1N1P/p1r1rPP1/R5K1/1R1Q4 b - - acd 34; acs 210; am Rc4; bm Rc4; ce -198; pm Rc4;
pv Rc4 Rf2;
4r1bk/1p2r1qp/pb6/2Npn1p1/1P1B1p2/2PB1P1P/P2Q2P1/2R1R2K b - - acd 35; acs 209; am Bf7; bm Bf7; ce -202; pm Bf7; pv Bf7;

r5k1/p3r2p/P1bp2pB/3pPq2/1R6/2P3QP/6P1/1R4K1 b - - acd 36; acs 209; am Qxe5; bm Qxe5; ce -203; pm Qxe5;
pv Qxe5 Bf4;
6k1/3r4/2n3q1/p1Ppbp1p/Pp2p1pP/1P4P1/2R1NPK1/2QR4 w - - acd 40; acs 209; am Nf4; bm Nf4; ce -212; pm Nf4; pv Nf4;

8/6k1/3bq3/5pQ1/1p5P/6PK/5P2/3B4 b - - acd 33; acs 209; am Qg6; bm Qg6; ce -225; pm Qg6;
pv Qg6 Qd2 Be7 Bc2 Bc5 f3 Bd6 f4 Kf8 Kh2 Ke7 Bb3 Qf6 Qe2+ Kf8 Qc4 Kg7 Kh3 Qg6 Qd4+ Qf6 Qd5 Qg6 Bc4 b3 Bd3 b2 Bxf5 Qf7 Qd3 Bc7 Qb5 Kf8 Kg4 Qf6 Qc5+ Bd6 Qc8+ Kg7 Kf3 Be7 Bd3 Qf7 Qb7;
8/8/1n3k1p/2pP1p1P/2P5/pK4P1/1b1NN3/8 w - - acd 35; acs 209; am Ka2; bm Ka2; ce -240; pm Ka2;
pv Ka2 Kg5 d6 Kg4 Nf1 Kf3 Nf4 Be5 Ng6 Bxd6 Nh4+ Ke4 Nd2+ Ke5 Kxa3 Ke6 Nhf3 Bxg3 Kb2 Nd7 Nb3 f4 Kc3 Kf5 Kd3 Kg4 Nbd2 Kxh5 Ke4 Kg4 Kd5 h5 Kd6 Nb6 Kxc5 Bf2+ Kb5 h4 Nxh4 Kxh4 Ne4 Be3 c5 Nd5 Kc4 Ne7;
7k/2n1br2/2q4p/3nP3/pp1P2Q1/3B2P1/1P3B1P/5RK1 w - - acd 33; acs 209; am Qe4; bm Qe4; ce -242; pm Qe4; pv Qe4;

1k1rr3/1p3p2/p1n3q1/P4Npp/1bNp2P1/1P3PP1/2PR4/1KQ2R2 w - - acd 35; acs 209; am Rd3; bm Rd3; ce -244; pm Rd3;
pv Rd3 hxg4 fxg4 Bc3 Rfd1 f6 Rxc3 dxc3 Rxd8+ Rxd8 Qe1 Ne5 Nfd6 Nxc4 Nxc4 Rd4 Qxc3 Qe4 Ka2 Qxg4 Qb4 Qd1 Kb2 Rd8 Qe7 Qd7 Qxf6 Qd4+ Qxd4 Rxd4 Ne3 Re4 Nf5 Kc7 Kc1 Re5 Nd4 Rxa5 Kd2 Kd6 g4 Ke5 Ke3 Kf6 c3 Ra1 Kf3 Rg1 Nf5 Ke5 Nd4;
8/2pp4/p4k2/1p4pp/2PPP3/P1P2K1P/8/8 w - - acd 37; acs 209; am c5; bm c5; ce -252; pm c5; pv c5;

k7/2r5/1p2RQ2/8/p1q4P/3p4/3P2PK/8 b - - acd 36; acs 209; am Rb7; bm Rb7; ce -275; pm Rb7;
pv Rb7 Rc6;
3br1k1/1bq5/p2p1ppB/3P4/PpnNP1N1/5QPP/1p3P2/4R1K1 w - - acd 30; acs 209; am Rd1; bm Rd1; ce -280; pm Rd1;
pv Rd1 f5;
8/3b2k1/4p1p1/3pP1Pp/1p1P3P/P1r2RK1/P5N1/8 b - - acd 49; acs 209; am Rxa3; bm Rxa3; ce -300; pm Rxa3;
pv Rxa3 Rxa3;
7r/1rqb1pk1/p2p1n1p/Pp1P2p1/2p1P1P1/2P2PNR/Q3B1P1/1K5R w - - acd 30; acs 209; am Kc2; bm Kc2; ce -324; pm Kc2;
pv Kc2 b4;
2b5/r1r4p/pRp2p1k/P3pPpP/2BpP1P1/8/2P2K2/1R6 b - - acd 34; acs 209; am Bb7; bm Bb7; ce -356; pm Bb7; pv Bb7;

5b1k/2r1qppb/p1n1p3/P2pP2N/3P2QN/4B2P/2r2PRK/R7 b - - acd 33; acs 209; am Rc4; bm Rc4; ce -359; pm Rc4;
pv Rc4 Bh6;
5q2/1bb3pk/p4r1p/2p1n3/PpN1Pr2/3PQ2P/1P1N1PB1/R4RK1 w - - acd 32; acs 209; am Rac1; bm Rac1; ce -372; pm Rac1;
pv Rac1 Nf3+;
2nb2Nk/q4p2/4p2p/pQ1pPnpN/1p1P4/1P2B1PP/P4PK1/8 b - - acd 35; acs 209; am Nxe3+; bm Nxe3+; ce -386; pm Nxe3+;
pv Nxe3+ Kh2 Kxg8 Qe8+ Kh7 Qxd8 Nf1+ Kg2 Qxd4 Nf6+ Kg6 Qg8+ Kf5 Ng4 Ne3+ Nxe3+ Kxe5 Qh8+ Ke4 Qxc8 Qc3 Qb7 Qe5 Qxf7 Kd3 Qh5 Qe4+ Kh2 Kd4 Qxh6 Kc3 Qh5 d4 Nc4 Qe1 Qf3+ d3 Qf6+ e5 Ne3 Kb2 Qxe5+ Kxa2 Qxa5+ Kxb3 Qd5+ Kb2 Kg2 d2 Qd3 Qc1 Nc4+ Qxc4 Qxc4 d1=Q Qxb4+ Ka2 Qc4+ Kb2 h4 g4 h5 Qf3+ Kg1 Kb1;
3B4/7p/p2bp3/6P1/P2KPk2/8/8/8 w - - acd 53; acs 48; am Bf6; bm Bf6; ce -388; pm Bf6;
pv Bf6 Bc7 Kd3 Kg4 Ke2 Bf4 g6 hxg6 Kf2 Bc7 Kg2 Kf4 Bh4 Kxe4 Bf2 g5 Bc5 Be5 Bf8 Bf4 Kf2 Bc7 Ba3 Kf4 Bf8 Ke5 Bb4 Bd6 Be1 Kf5 Kf3 Bc7 Bb4 Be5 Kg2 Bf4 Be1 Bc7 Bb4 g4 Kf2 Kf4 Bf8 Ke5 Bh6 Bd6 Kg3 Kf5+ Kf2 Bc5+ Kg3 Be7 Bg7 Bd6+ Kg2 Kf4 Bf6;
8/p1k5/1p1p2r1/1Pp1p1pp/P1P1Pr2/1n1Q1PqP/4B1P1/2RR2K1 w - - acd 34; acs 209; am Qd5; bm Qd5; ce -395; pm Qd5;
pv Qd5 Nxc1;
8/2b3r1/1N6/8/B3k3/7K/8/8 w - - acd 44; acs 210; am Bc6+; bm Bc6+; ce -408; pm Bc6+;
pv Bc6+ Ke5;
1r3bk1/5ppp/1P1p4/1Q2pqP1/8/1R2B3/1P3P1P/2K5 b - - acd 34; acs 209; am Qc8+; bm Qc8+; ce -429; pm Qc8+;
pv Qc8+ Rc3;
2r5/p5pk/3P3p/1p2P3/3Q2q1/3R4/6P1/6K1 b - - acd 31; acs 209; am Qe6; bm Qe6; ce -973; pm Qe6;
pv Qe6 Qd5 Rc1+ Kh2 Qf5 Rg3 Qh5+ Rh3 Qg4 Qd3+ Kg8 d7 Qf4+ Qg3 Qd4 Qb3+ Kh8 Qe3 Qxd7 Qxc1 b4 Qc5 Qf7 Qxb4 Qf5 Qd4 Qf7 Qd8+ Kh7 Qd3+ Kh8 Rf3 Qe7 Qe4 Qe8 Rb3 Kg8 Qd5+ Kh7 e6 Kh8 Qe4 Qh5+ Rh3 Qd1 e7;
5r2/R7/2k5/8/2p3Pp/7P/p1K5/8 w - - acd 183; acs 287; am Rxa2; bm Rxa2; ce -32585; dm -91; pm Rxa2;
pv Rxa2 Kb5 Ra1 Rf2+ Kc3 Rf3+ Kd4 Rd3+ Ke5 c3 g5 Rxh3 g6 Kb4 Rb1+ Ka3 Ra1+ Kb2 Ra8 Rg3 Rb8+ Ka2 Rc8 Kb3 Rb8+ Kc4 Rc8+ Kd3 Rd8+ Ke2 Rc8 h3 Kf6 Kf2 Rh8 Kg2 Re8 h2 Re2+ Kh3 Re1 Rg1 Re3+ Kh4 Re4+ Kg3 Re5 Kf2 Rf5+ Ke3 Re5+ Kd4 Kf5 Rxg6;

There are also a large number of positions that are lost no matter what you do.  So there is no value in avoiding some particular move because the other moves lose anyway.
Parent - By gsgs (**) [de] Date 2017-09-13 04:06 Edited 2017-09-13 04:53
this is for 5s per move with 1700 kn/s in the startposition.
It doesn't matter, what the eval at high-depth is, the engines can't know it.
If they make a move and the eval goes down, then it was a bad move,
it makes sense to avoid it. That;s how these engines work, how they
are programmed.
It doesn't matter if someone later with superior depths shows that the eval was wrong.

We can make a testsuite for games at longer time control, that one would contain
different positions. But then you get fewer iterations in the same time.

And yes, it makes sense to filter out those where the drop in eval just
results from the recognition that the position was worse than thought one move ago,
no matter what move was made.
But this won't affect the overall result much, it just adds randomness
(like adding random testpositions), so you need more iterations
to get into the same confidence interval.

The above charts show, that I have enough iterations to distinguish
differences of ~2 Elo in testposition-analysis. (of this kind)
Better than e.g. nextchessmove.com with 10000 games.

The iterations of 1231 positions give an estimated deviation of 10-15 Elo
or a 95% confidence interval of ~20-30 Elo *2 .

So for 150 iterations taking ~11 days on one tablet it is ~+- 2 Elo  [rough estimate]
This could probably be improved by measuring the time where the bad move was
first avaided ...
Parent - - By gsgs (**) [de] Date 2017-11-06 02:28
I wrote on 2017/09/11 :

> I filtered 1231 of these : http://magictour.free.fr/fcp1231.fen ,
> those where asmfish moves were different, and tested it on asmfishes,
>
> http://magictour.free.fr/1231G1.GIF
> (to be updated)
> now on my new Ryzen Komodo 10.2 Linux popcnt seems to be
> 30 Elo better than asmfish Linux popcnt Nov.04 
> on testpositions ?! Can it be ?

the reason seems to be that by {filtering positions where several versions of asmfish disagree on the best move}
I also filtered the positions where asmfish is more likely to make the "bad mave" in that position.

And then Komodo is better with those positions.

When I now run Komodo on all 5356 positions, it's weaker than asmfish (1243.1 vs. 1179.1 in average,
10 runs of 5356 positions at 0.8s per position , 13 cores , 14700 kn/s in the startposition)

Now I try to filter positions where different versions of Komodo disagree , expecting those then
being "solved" better by asmfish ...
Parent - By Labyrinth (*****) [us] Date 2017-11-06 16:36

>I also filtered the positions where asmfish is more likely to make the "bad mave" in that position.


What's a 'bad mave'? Bad move? The 'o' is so far from the 'a' though even on German keyboard layouts.
Up Topic The Rybka Lounge / Computer Chess / testsuite 5356 engine mistakes

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill