To put everything into place with Strelka

[Alexander Schmidt]

Hello Sergei,

it is unbelieveable what kind of witch hunt started because of Strelka. Maybe some of you know that I am kind of experienced in detecting clones. I posted following things in a german forum before it came to public that Sergei Markoff saw the sources of Strelka.

Strelka is not a clone of Rybka. It is easy to find out that there are essential differences. They have a different movegeneration and different evaluation and knowledge.

All we have is that Strelka is suspicious because it is somehow encrypted and because it behaves similar to Rybka in many illegal positions. Well, I saw a lot of encrypted free engines, one example is Homer which is for sure no clone.

And about the behaviour in illegal positions... Engines with similar techniques behave similar in illegal positions. Try out Crafty and Dragon. Both are Bitboard engines and botch behave in the same way in 95% of illegal positions. You cant say anything only with illegal positions, they are just a hint to take a closer look.

What you guys have done is, everyone searched for similaries and posted them, but noone looked for differences. Thats all.


Here are a few positions of me:

In such positions you can check the movegen, a lot of moves are a mate:

FEN: 6KB/3Q4/2Q5/2R1R3/3k4/2R1R3/2Q1Q3/8 b - - 1 1

Strelka:
1 00:00 1 1 +M283 h8g7
1 00:00 2 2 +M277 h8f6
1 00:00 12 12 +M0 e6d7
1 00:00 12 12 +319,00 e6d7


Rybka v1.0 Beta.w32:
Dc2-d1

So the movegen is different. What you can also see here is that Rybka plays the move immediately while Strelka looks deeper.


Now a look on the endgame knowledge:

FEN: 8/8/8/4k3/8/3K4/2B5/8 w - - 0 1

Rybka v1.0 Beta.w32:
3 00:00 242 247.808 -0,03 Lc2b1
4 00:00 1.822 1.865.728 +0,03 Lc2b1
5 00:00 3.158 3.233.792 -0,03 Lc2b1 Ke5f4
6 00:00 15.255 918.889 +0,03 Lc2b1 Ke5f4 Lb1a2

Rybka evaluates around 0 while Strelka:

FEN: 8/8/8/4k3/2K5/8/2B5/8 b - - 1 1

Strelka:
1 00:00 1 1 +2,75 d3c4
2 00:00 14 14 +2,81 d3c4 e5f4
3 00:00 67 67 +2,75 d3c4 e5f4 c2b1
4 00:00 321 321 +2,81 d3c4 e5f4 c2b1

thinks it has an advantage.

And btw. an illegal position which happened by accident in the previous test:

FEN: 8/2B5/8/4k3/2K5/8/8/8 b - - 0 1

Strelka:
1 00:00 1 1 +2,75 c7e5
1 00:00 12 12 +2,81 c7b8 e5e4
2 00:00 48 48 +2,75 c7b8 e5e4 b8h2
2 00:00 64 64 +2,81 c7e5 a5a4
3 00:00 217 217 +2,75 c7e5 a5b4 e5h2
3 00:00 288 288 +2,81 c7b8 e5e4 b8h2 e4e3


Rybka crashes here...

There is really not enough to say Strelka is a clone, and after Sergey Markoff saw the sources I am sure Strelka is OK.

There is no need for Ossipov to show anyone else the sources, first bring on more hints it could be a clone.

Ossipov has done much more than others and much more than he has to do.

Best,
Alex

[Sylwy]

Hello Alexander !
Happy to read a such well-mannered, honest & professional man like you.A true German spirit !
I'm so sick with the torrents of arrogance,hate & chauvinism of some people here-in the last few days.Wonder for me because I supposed here post only educated men !
www.uciengines.de ???-it lacks me a lot !!!
Thank you again Alexander !
Regards,
Sylwy

[Gerd Isenberg]

Hi Gerd,

maybe you overlooked my smiles ??
Its just my logical view of point and i wrote its just a choose of taste.
So i accept other ways of course.

No need to do a rude personal attack on me!
i'm not inspired

Best,
Daniel

PS: I don't know Sergei, thats the truth. What should i write ? That we're friends since 10 years ? Here are so many programmers, you can't know all of them. It's just a fact and has no negative vibrations.

Hi Daniel,

Personal attack? May be you overlooked my smilies?? Since I used almost the same words as you did, does that imply your statement was a personal attack against Sergei as well?

I don't know much about you, but it seems that you're not a brother of ...

Explicitly stating that, implies "i don't trust you at all". May be i am a bit over sensitized - but imho this is social chauvinism.

Gerd

[Daniel Mehrmann]

Hi Gerd,

maybe you overlooked my smiles ??
Its just my logical view of point and i wrote its just a choose of taste.
So i accept other ways of course.

No need to do a rude personal attack on me!
i'm not inspired

Best,
Daniel

PS: I don't know Sergei, thats the truth. What should i write ? That we're friends since 10 years ? Here are so many programmers, you can't know all of them. It's just a fact and has no negative vibrations.

Hi Daniel,

Personal attack? May be you overlooked my smilies?? Since I used almost the same words as you did, does that imply your statement was a personal attack against Sergei as well?

I don't know much about you, but it seems that you're not a brother of ...

Explicitly stating that, implies "i don't trust you at all". May be i am a bit over sensitized - but imho this is social chauvinism.

Gerd

[Daniel Mehrmann]

Hi Daniel,

Personal attack? May be you overlooked my smilies?? Since I used almost the same words as you did, does that imply your statement was a personal attack against Sergei as well?

Don't be silly Gerd. I was talking about the name "Michael Ehrig".

With the name "Mr. Spock" i was explaining that its my logical way and nothing more.

With the name "Michael Ehrig" you're saying i'm cloning engines or at least i'm a cloner ! (should i show you the old CCC archives where you attacking me as well with the same spot ?)

So what ?

Explicitly stating that, implies "i don't trust you at all". May be i am a bit over sensitized - but imho this is social chauvinism.

Gerd

Look what i have written in the CSS-Forum in german, it's the same and yes maybe it sounds to hart.


Trying to send greetings
Daniel

[Chan Rasjid]

Hello,

As Rybka's code is private, the only way that was mentioned that a Rybka clone may be made is for Osipov to disassemble and then reassemble Rybka.
But Sergei has been long in chess programming and can easily know if the C source of Strelka is original. There is usually proper naming conventions in original sources and if Strelka's C codes is a translation from disassembled codes, it need to figure out how things work and use proper names that had had to have meaning related to usage. It is doubtful if it can be done and yet escape the notice of Sergei.


For the following position:-
[D]8/8/7R/5p1P/p7/r1n1k3/6PK/8 b - - 0 55

time - 60 moves per 40 secs.
Strelka - b
Glaurung - w
mate after 36 moves

Free Rybka - b
Glaurung - w
draw after 70 moves (remember Rybka do not know how to mate kqk)

SnailChess (Elo about TSCP + 50/100) - b
Glaurung - w
mate after 20 moves (the lower the ELO the better )

Main Hash Size( 24 MB )
No of Entry( 1572864 )
Entry Size( 16 byte)

1 a3b3 ( 368) depth( 6) pvL(10) maxPly(15) nps(175209) stage(1) Qnode( 9%)
Probe/HitGE (184314, 26.2% ) nEval/Hit (70580, 1.0% ) nGen( 18)
attk (0.0, eg 0.0) t(alloc 1.30, 1.05) ftime(6) Poll(0)
1 h6c6
2 a4a3 ( 356) depth( 6) pvL(11) maxPly(18) nps(191850) stage(1) Qnode( 8%)
Probe/HitGE (245754, 24.0% ) nEval/Hit (63782, 2.5% ) nGen( 24)
attk (0.0, eg 0.0) t(alloc 1.26, 1.28) ftime(6) Poll(0)
2 h5h6
followPV
3 e3d4 ( 399) depth( 7) pvL(11) maxPly(18) nps(181860) stage(1) Qnode(13%)
Probe/HitGE (180217, 27.6% ) nEval/Hit (48224, 3.7% ) nGen( 24)
attk (0.0, eg 0.0) t(alloc 1.22, 0.99) ftime(7) Poll(1)
3 c6d6
4 d4e5 ( 406) depth( 8) pvL(12) maxPly(19) nps(168791) stage(1) Qnode(15%)
Probe/HitGE (184311, 22.5% ) nEval/Hit (54353, 2.2% ) nGen( 6)
attk (0.0, eg 0.0) t(alloc 1.19, 1.09) ftime(8) Poll(1)
4 d6a6
5 a3a2 ( 406) depth( 7) pvL( 6) maxPly(15) nps(163007) stage(1) Qnode(22%)
Probe/HitGE (192504, 25.8% ) nEval/Hit (66585, 3.0% ) nGen( 21)
attk (0.0, eg 0.0) t(alloc 1.15, 1.18) ftime(7) Poll(1)
5 h6h7
followPV
6 b3b8 ( 546) depth( 9) pvL( 2) maxPly(17) nps(204345) stage(1) Qnode(12%)
Probe/HitGE (184312, 26.0% ) nEval/Hit (34052, 4.3% ) nGen( 21)
attk (0.0, eg 0.0) t(alloc 1.11, 0.90) ftime(9) Poll(0)
6 a6a5
7 e5d4 ( 598) depth( 7) pvL( 8) maxPly(15) nps(190619) stage(1) Qnode(15%)
Probe/HitGE (167929, 19.3% ) nEval/Hit (40096, 2.3% ) nGen( 9)
attk (0.0, eg 0.0) t(alloc 1.08, 0.88) ftime(7) Poll(0)
7 g2g3
8 b8h8 ( 662) depth( 6) pvL( 7) maxPly(15) nps(199804) stage(1) Qnode(14%)
Probe/HitGE (172027, 22.4% ) nEval/Hit (32522, 3.8% ) nGen( 27)
attk (0.0, eg 0.0) t(alloc 1.05, 0.86) ftime(6) Poll(1)
8 h2g1
9 d4e4 ( 669) depth( 6) pvL( 7) maxPly(17) nps(235453) stage(1) Qnode( 9%)
Probe/HitGE (217082, 19.3% ) nEval/Hit (29994, 2.7% ) nGen( 21)
attk (0.0, eg 0.0) t(alloc 1.02, 0.92) ftime(6) Poll(0)
9 g3g4
10 f5g4 ( 869) depth( 7) pvL( 8) maxPly(14) nps(191921) stage(1) Qnode(12%)
Probe/HitGE (155646, 12.8% ) nEval/Hit (27561, 0.4% ) nGen( 22)
attk (0.0, eg 0.0) t(alloc 0.99, 0.81) ftime(7) Poll(0)
10 g1g2
11 h8h7 ( 875) depth( 7) pvL( 7) maxPly(15) nps(227746) stage(1) Qnode(10%)
Probe/HitGE (161922, 22.6% ) nEval/Hit (26792, 3.2% ) nGen( 20)
attk (0.0, eg 0.0) t(alloc 0.97, 0.71) ftime(7) Poll(0)
11 a5a3
12 h7b7 ( 1026) depth( 9) pvL( 2) maxPly(16) nps(226060) stage(1) Qnode(11%)
Probe/HitGE (172027, 27.4% ) nEval/Hit (33503, 2.5% ) nGen( 29)
attk (0.0, eg 0.0) t(alloc 0.94, 0.76) ftime(9) Poll(0)
12 a3a8
13 b7b1 ( 1030) depth( 6) pvL( 6) maxPly(14) nps(213042) stage(1) Qnode(13%)
Probe/HitGE (125906, 32.3% ) nEval/Hit (25383, 4.2% ) nGen( 29)
attk (0.0, eg 0.0) t(alloc 0.92, 0.59) ftime(6) Poll(0)
13 g2f2
14 a2a1q ( 1063) depth( 6) pvL(13) maxPly(14) nps(229420) stage(0) Qnode(11%)
Probe/HitGE (167929, 23.4% ) nEval/Hit (29421, 1.7% ) nGen( 27)
attk (0.0, eg 0.0) t(alloc 0.90, 0.73) ftime(6) Poll(0)
14 a8a1
followPV
15 b1a1 ( 1077) depth( 8) pvL( 9) maxPly(15) nps(293744) stage(1) Qnode( 3%)
Probe/HitGE (211782, 26.5% ) nEval/Hit (21258, 3.2% ) nGen( 27)
attk (0.0, eg 0.0) t(alloc 0.87, 0.72) ftime(8) Poll(0)
15 f2g3
followPV
16 e4f5 ( 1117) depth( 8) pvL( 9) maxPly(17) nps(299247) stage(1) Qnode( 2%)
Probe/HitGE (158593, 50.8% ) nEval/Hit (31350, 6.0% ) nGen( 27)
attk (0.0, eg 0.0) t(alloc 0.85, 0.53) ftime(8) Poll(0)
16 g3g2
17 a1a2 ( 1497) depth(18) pvL( 2) maxPly(27) nps(220822) stage(1) Qnode( 3%)
Probe/HitGE (101796, 51.0% ) nEval/Hit (13533, 8.5% ) nGen( 30)
attk (0.0, eg 0.0) t(alloc 0.83, 0.46) ftime(18) Poll(0)
17 g2g1
18 f5e4 ( 7993) depth( 7) pvL( 8) maxPly( 9) nps(141252) stage(1) Qnode( 1%)
Probe/HitGE (24010, 45.1% ) nEval/Hit ( 473, 4.1% ) nGen( 29)
attk (0.0, eg 0.0) t(alloc 0.81, 0.17) ftime(10) Poll(0)
18 g1f1
followPV
19 e4f4 ( 7995) depth( 5) pvL( 6) maxPly( 4) nps(141252) stage(1) Qnode( 0%)
Probe/HitGE ( 485, 76.5% ) nEval/Hit ( 0, -1.$% ) nGen( 29)
attk (0.0, eg 0.0) t(alloc 0.81, 0.00) ftime(3) Poll(0)
19 f1e1
followPV
20 f4g3 ( 7997) depth( 3) pvL( 4) maxPly( 2) nps(141252) stage(1) Qnode( 0%)
Probe/HitGE ( 57, 64.9% ) nEval/Hit ( 0, -1.$% ) nGen( 29)
attk (0.0, eg 0.0) t(alloc 0.81, 0.01) ftime(2) Poll(0)
20 e1f1
followPV
Result 0-1{SnailChess mate}
0-1

Best Regards
Rasjid

[Zlaire]

I completely agree with Ed Schröder here. And I think it's important that any new (strong) engine goes through the process of eliminating it from any clone suspicions.

Better to thoroughly verify every new engine than just accept it and waste hours and hours of testing time on something that indeed turns out to be a clone.

There is nothing wrong with that. It would be more wrong to blindly accept it in my opinion.

[Rob]

Hello Sergei,

it is unbelieveable what kind of witch hunt started because of Strelka. Maybe some of you know that I am kind of experienced in detecting clones. I posted following things in a german forum before it came to public that Sergei Markoff saw the sources of Strelka.

Strelka is not a clone of Rybka. It is easy to find out that there are essential differences. They have a different movegeneration and different evaluation and knowledge.

All we have is that Strelka is suspicious because it is somehow encrypted and because it behaves similar to Rybka in many illegal positions. Well, I saw a lot of encrypted free engines, one example is Homer which is for sure no clone.

And about the behaviour in illegal positions... Engines with similar techniques behave similar in illegal positions. Try out Crafty and Dragon. Both are Bitboard engines and botch behave in the same way in 95% of illegal positions. You cant say anything only with illegal positions, they are just a hint to take a closer look.

What you guys have done is, everyone searched for similaries and posted them, but noone looked for differences. Thats all.


Here are a few positions of me:

In such positions you can check the movegen, a lot of moves are a mate:

[D] 6KB/3Q4/2Q5/2R1R3/3k4/2R1R3/2Q1Q3/8 b - - 1 1

Strelka:
1 00:00 1 1 +M283 h8g7
1 00:00 2 2 +M277 h8f6
1 00:00 12 12 +M0 e6d7
1 00:00 12 12 +319,00 e6d7


Rybka v1.0 Beta.w32:
Dc2-d1

So the movegen is different. What you can also see here is that Rybka plays the move immediately while Strelka looks deeper.

Strelka simply says "1-0 {White mates}"

Rybka says "bestmove a1a1"

Both identify the position correctly as an unplayable endposition.
The difference is about the interface code wich is obviously
different between them. UCI doesn't have the Winboard output "1-0 {White mates}"

Now a look on the endgame knowledge:

[D] 8/8/8/4k3/8/3K4/2B5/8 w - - 0 1

Rybka v1.0 Beta.w32:
3 00:00 242 247.808 -0,03 Lc2b1
4 00:00 1.822 1.865.728 +0,03 Lc2b1
5 00:00 3.158 3.233.792 -0,03 Lc2b1 Ke5f4
6 00:00 15.255 918.889 +0,03 Lc2b1 Ke5f4 Lb1a2

Rybka evaluates around 0 while Strelka:

[D] 8/8/8/4k3/2K5/8/2B5/8 b - - 1 1

Strelka:
1 00:00 1 1 +2,75 d3c4
2 00:00 14 14 +2,81 d3c4 e5f4
3 00:00 67 67 +2,75 d3c4 e5f4 c2b1
4 00:00 321 321 +2,81 d3c4 e5f4 c2b1

thinks it has an advantage.

Strelka lacks some drawn endgame knowledge and gives more to white.

And btw. an illegal position which happened by accident in the previous test:

[D] 8/2B5/8/4k3/2K5/8/8/8 b - - 0 1

Strelka:
1 00:00 1 1 +2,75 c7e5
1 00:00 12 12 +2,81 c7b8 e5e4
2 00:00 48 48 +2,75 c7b8 e5e4 b8h2
2 00:00 64 64 +2,81 c7e5 a5a4
3 00:00 217 217 +2,75 c7e5 a5b4 e5h2
3 00:00 288 288 +2,81 c7b8 e5e4 b8h2 e4e3


Rybka crashes here...

Not illegal at all. It's just the black king in check.
Both move the king out of the way.
Again, Strelka lacks some drawn endgame knowledge and gives more to white.

There is really not enough to say Strelka is a clone, and after Sergey Markoff saw the sources I am sure Strelka is OK.

There is no need for Ossipov to show anyone else the sources, first bring on more hints it could be a clone.

Ossipov has done much more than others and much more than he has to do.

Best,
Alex

I have no doubt that Strelka is a clone. The only problem is to prove it without betraying Rybka secrets.

[Andre van Ark]

Hi all,

What astonishes me: Why does the programmer do the effort to make an engine, but takes he no effort to tell something about it. For example some release notes, FAQ and so on.

André

[Sylwy]

Hi !

1.Strelka is for free and comes from another region with other standards & exigencies.
2.I heared-these days-a lot of stupidities here: the no.1 in top-far away-is the name of Strelka used like reason to be considered a clone.
3.Why the new Loop isn't it thought that to be a Rybka clone despite the history with the very first List( versus Crafty), despite the use of bitboards,despite some similar analyses,despite....... ?
I like to see the same severe manners to be applied for every new strong release but not in a selective way
So sorry !
Regards,
Sylwy