HasiLover wrote:

S_P_A_R_T wrote:

ArnoHu wrote:

HasiLover_Test wrote:

ArnoHu wrote:

Scratch Chess Engine Ranking (on Scratch 3 Runtime)

Based on this lichess.org study: https://lichess.org/study/v3EKTlR2/ , calculated using the BayesELO tool. Data from games of recent two years, and I ran some more in addition.

Rank	Name		Elo	+	-	games	score	oppo.	draws
1	GoK		1746	159	125	20	100%	0	0%
2	Element		1659	303	269	3	67%	141	0%
3	Bonsai		1604	167	158	10	60%	111	0%
4	Archimedes	1530	264	266	4	50%	100	0%
5	Shallow Blue	1514	262	238	3	67%	21	0%
6	The Turk	1475	210	240	6	33%	124	0%
7	LowDoor		1466	241	237	4	50%	43	0%
8	Chip		1430	235	283	4	25%	106	0%
9	Pseudo		1423	296	438	2	0%	180	0%
10	White Dove	1421	259	411	3	0%	220	0%
11	HarleyK		1393	258	311	4	25%	132	0%
12	Scurious	1386	305	303	2	50%	-55	0%
13	Wolverine	1385	279	446	3	0%	203	0%
14	U0		1366	314	461	2	0%	156	0%
15	Frenchgamerlol	1345	284	301	3	33%	19	0%
16	Midecah		1276	232	338	4	0%	84	0%
17	Mystery		1230	268	386	3	0%	45	0%

Data is still limited, too few games played. Bottom ELO numbers should be even lower, some rankings certainly wrong - Archimedes, The Turk too high, White Dove too low (currently facing some Scratch 3 issues). Midecah is the engine which has never drawn let alone won a game and should be last.

BTW, Midecah was very close to a repetition draw against Scurious right now. I started even crossing fingers for Midecah, like a once in a lifetime chance, but no - instead of making the next identical check-evasion, it chose not to.

Thundershark is not in the list, because it was designed for TurboWarp only. U0 has been taken offline it seems.

You can find the engines in this studio: https://scratch.mit.edu/studios/15896412/

I must say this is currently rather Innacurate as Scurious can easily beat Chip, Archimides and Turk, also Frenchgamerlol is way stronger and Midecah chess definately isnt 1200. White Dove also is definately in the top 4.

Yes, as stated before, some calibration games missing, but the trajectory is correct. I did some of those yesterday, and it improved things immediately already. Three times I had White Dove playing, each time it messed up the board on P=2, so there was nothing I could do. I will try again today, as S_P_A_R_T has fixed that now.

And as said, those are not FIDE numbers. Also, for Midecah to fall to 1000 or similar, it must play more games.

WD's bug has still not been fixed sadly. I've been quite busy with my YT Channel (again). I've just been spending 100% of my time on that for the past 2 weeks to get it to 1000 subs (only 3 away right now!). But today, I'll hopefully start debugging the issue.

(The bug is also really weird in the way that you'd expect the bug to happen in TW, as it's just faster S3. But no… It just doesn't occur on TW. And I've even tested S3 without the q-search limiter and it still bugs out.)

Just subscribed, your pretty close to 1000 Subscribers.

HasiLover_Test wrote:

I have just released Scurious 2, after doing some Bug Fixes and Tweaks.

ArnoHu wrote:

HasiLover_Test wrote:

I have just released Scurious 2, after doing some Bug Fixes and Tweaks.

Scurious 2 just drew against Chip. Scurious was clearly ahead, could have promoted on same move, but then went for three-fold repetition instead.

As I had forgotten neither of the two engines has PGN export, I must replay anyway :-)

HasiLover wrote:

ArnoHu wrote:

HasiLover_Test wrote:

I have just released Scurious 2, after doing some Bug Fixes and Tweaks.

Scurious 2 just drew against Chip. Scurious was clearly ahead, could have promoted on same move, but then went for three-fold repetition instead.

As I had forgotten neither of the two engines has PGN export, I must replay anyway :-)

Yes, it still doesnt know about draw by repitition, I will be fixing it later.

ArnoHu wrote:

HasiLover wrote:

ArnoHu wrote:

HasiLover_Test wrote:

I have just released Scurious 2, after doing some Bug Fixes and Tweaks.

Scurious 2 just drew against Chip. Scurious was clearly ahead, could have promoted on same move, but then went for three-fold repetition instead.

As I had forgotten neither of the two engines has PGN export, I must replay anyway :-)

Yes, it still doesnt know about draw by repitition, I will be fixing it later.

Replay. same opening, deterministic game by both, same result: https://lichess.org/study/v3EKTlR2/10eJhWR2

HasiLover wrote:

ArnoHu wrote:

HasiLover wrote:

ArnoHu wrote:

HasiLover_Test wrote:

I have just released Scurious 2, after doing some Bug Fixes and Tweaks.

Scurious 2 just drew against Chip. Scurious was clearly ahead, could have promoted on same move, but then went for three-fold repetition instead.

As I had forgotten neither of the two engines has PGN export, I must replay anyway :-)

Yes, it still doesnt know about draw by repitition, I will be fixing it later.

Replay. same opening, deterministic game by both, same result: https://lichess.org/study/v3EKTlR2/10eJhWR2

Its probably assuming that Black will play Ka8 and thats why it tried to play bc2 to wait, because it wanted the promotion to come with check.
I did a test on this Position:1k1K4/8/8/8/8/8/5p2/8 b - - 0 1 and Scurious didnt promote until it promoted with a check.

ArnoHu wrote:

HasiLover wrote:

ArnoHu wrote:

HasiLover wrote:

ArnoHu wrote:

HasiLover_Test wrote:

I have just released Scurious 2, after doing some Bug Fixes and Tweaks.

Scurious 2 just drew against Chip. Scurious was clearly ahead, could have promoted on same move, but then went for three-fold repetition instead.

As I had forgotten neither of the two engines has PGN export, I must replay anyway :-)

Yes, it still doesnt know about draw by repitition, I will be fixing it later.

Replay. same opening, deterministic game by both, same result: https://lichess.org/study/v3EKTlR2/10eJhWR2

Its probably assuming that Black will play Ka8 and thats why it tried to play bc2 to wait, because it wanted the promotion to come with check.
I did a test on this Position:1k1K4/8/8/8/8/8/5p2/8 b - - 0 1 and Scurious didnt promote until it promoted with a check.

OK, and the next game was against Archimedes, which went back and forth, until Scurious moved its pinned pawn trying capture the knight on d4, invalid move: https://lichess.org/Pmgee7a0#49

In one of the first versions, I had to force GoK by some minor bonus when promoting on earlier move.

ArnoHu wrote:

Is it just me, or did you also lose that lichess.org analysis board feature that would show you the next best move via a blue arrow once you activate local evaluation? I now have to hover over a variation in the PGN moves panel, and that only appears during live analytics (https://lichess.org/analysis/), not on a stored game.

ArnoHu wrote:

OK, whatever we might say about Archimedes Chess, but this old Swashbuckler of a Scratch Chess Engine won against White Dove (P2) after having faced mate in 3, won against Shallow Blue (Depth 2) after having faced mate in 6, and drew against Scurious (Ply 4) after having faced mate in 2, in just three consecutive games. True survival skills!

Those are chapters 42 - 44 in our Scratch 3 Runtime study: https://lichess.org/study/v3EKTlR2

It will take quite some wins to replace it from place 4. I think I will give Frenchgamerlol the next run, but who knows how this might end…

Rank	Name		Elo	+	-	games	score	oppo.	draws
1	GoK		1756	164	128	20	100%	1	0%
2	Element		1653	312	277	3	67%	126	0%
3	Bonsai		1602	170	160	10	60%	106	0%
4	Archimedes	1558	199	184	7	64%	50	14%
5	The Turk	1479	211	240	6	33%	128	0%
6	Shallow Blue	1475	232	232	4	50%	45	0%
7	LowDoor		1473	241	237	4	50%	50	0%
8	Chip		1435	205	231	5	30%	88	20%
9	Scurious	1427	213	212	4	50%	-13	50%
10	HarleyK		1397	258	311	4	25%	137	0%
11	Pseudo		1395	308	460	2	0%	165	0%
12	Wolverine	1393	279	446	3	0%	211	0%
13	U0		1374	315	462	2	0%	164	0%
14	White Dove	1371	238	383	4	0%	192	0%
15	Frenchgamerlol	1348	288	305	3	33%	25	0%
16	Midecah		1280	244	387	4	0%	105	0%
17	Mystery		1234	269	386	3	0%	50	0%

birdracerthree wrote:

ArnoHu wrote:

Is it just me, or did you also lose that lichess.org analysis board feature that would show you the next best move via a blue arrow once you activate local evaluation? I now have to hover over a variation in the PGN moves panel, and that only appears during live analytics (https://lichess.org/analysis/), not on a stored game.

This is a setting you can toggle with the settings button in the lichess analysis (to the right of the button that toggles Stockfish).

HasiLover wrote:

ArnoHu wrote:

OK, whatever we might say about Archimedes Chess, but this old Swashbuckler of a Scratch Chess Engine won against White Dove (P2) after having faced mate in 3, won against Shallow Blue (Depth 2) after having faced mate in 6, and drew against Scurious (Ply 4) after having faced mate in 2, in just three consecutive games. True survival skills!

Those are chapters 42 - 44 in our Scratch 3 Runtime study: https://lichess.org/study/v3EKTlR2

It will take quite some wins to replace it from place 4. I think I will give Frenchgamerlol the next run, but who knows how this might end…

Rank	Name		Elo	+	-	games	score	oppo.	draws
1	GoK		1756	164	128	20	100%	1	0%
2	Element		1653	312	277	3	67%	126	0%
3	Bonsai		1602	170	160	10	60%	106	0%
4	Archimedes	1558	199	184	7	64%	50	14%
5	The Turk	1479	211	240	6	33%	128	0%
6	Shallow Blue	1475	232	232	4	50%	45	0%
7	LowDoor		1473	241	237	4	50%	50	0%
8	Chip		1435	205	231	5	30%	88	20%
9	Scurious	1427	213	212	4	50%	-13	50%
10	HarleyK		1397	258	311	4	25%	137	0%
11	Pseudo		1395	308	460	2	0%	165	0%
12	Wolverine	1393	279	446	3	0%	211	0%
13	U0		1374	315	462	2	0%	164	0%
14	White Dove	1371	238	383	4	0%	192	0%
15	Frenchgamerlol	1348	288	305	3	33%	25	0%
16	Midecah		1280	244	387	4	0%	105	0%
17	Mystery		1234	269	386	3	0%	50	0%

It seems Archimides is just Nr. 4 on Scratch I guess

ArnoHu wrote:

HasiLover wrote:

ArnoHu wrote:

OK, whatever we might say about Archimedes Chess, but this old Swashbuckler of a Scratch Chess Engine won against White Dove (P2) after having faced mate in 3, won against Shallow Blue (Depth 2) after having faced mate in 6, and drew against Scurious (Ply 4) after having faced mate in 2, in just three consecutive games. True survival skills!

Those are chapters 42 - 44 in our Scratch 3 Runtime study: https://lichess.org/study/v3EKTlR2

It will take quite some wins to replace it from place 4. I think I will give Frenchgamerlol the next run, but who knows how this might end…

Rank	Name		Elo	+	-	games	score	oppo.	draws
1	GoK		1756	164	128	20	100%	1	0%
2	Element		1653	312	277	3	67%	126	0%
3	Bonsai		1602	170	160	10	60%	106	0%
4	Archimedes	1558	199	184	7	64%	50	14%
5	The Turk	1479	211	240	6	33%	128	0%
6	Shallow Blue	1475	232	232	4	50%	45	0%
7	LowDoor		1473	241	237	4	50%	50	0%
8	Chip		1435	205	231	5	30%	88	20%
9	Scurious	1427	213	212	4	50%	-13	50%
10	HarleyK		1397	258	311	4	25%	137	0%
11	Pseudo		1395	308	460	2	0%	165	0%
12	Wolverine	1393	279	446	3	0%	211	0%
13	U0		1374	315	462	2	0%	164	0%
14	White Dove	1371	238	383	4	0%	192	0%
15	Frenchgamerlol	1348	288	305	3	33%	25	0%
16	Midecah		1280	244	387	4	0%	105	0%
17	Mystery		1234	269	386	3	0%	50	0%

It seems Archimides is just Nr. 4 on Scratch I guess

Haha, yes. And I am not saying anything any more, but here is Frenchgamerlol vs. Archimedes: https://lichess.org/study/v3EKTlR2/nEUUO34Y

ArnoHu wrote:

HasiLover wrote:

ArnoHu wrote:

OK, whatever we might say about Archimedes Chess, but this old Swashbuckler of a Scratch Chess Engine won against White Dove (P2) after having faced mate in 3, won against Shallow Blue (Depth 2) after having faced mate in 6, and drew against Scurious (Ply 4) after having faced mate in 2, in just three consecutive games. True survival skills!

Those are chapters 42 - 44 in our Scratch 3 Runtime study: https://lichess.org/study/v3EKTlR2

It will take quite some wins to replace it from place 4. I think I will give Frenchgamerlol the next run, but who knows how this might end…

// snip

It seems Archimides is just Nr. 4 on Scratch I guess

Haha, yes. And I am not saying anything any more, but here is Frenchgamerlol vs. Archimedes: https://lichess.org/study/v3EKTlR2/nEUUO34Y

Rank	Name		Elo	+	-	games	score	oppo.	draws
1	GoK		1764	164	126	21	100%	4	0%
2	Element		1651	313	280	3	67%	123	0%
3	Bonsai		1601	171	161	10	60%	105	0%
4	Archimedes	1512	171	165	9	56%	63	22%
5	The Turk	1476	212	243	6	33%	127	0%
6	LowDoor		1476	241	237	4	50%	54	0%
7	Shallow Blue	1455	234	234	4	50%	25	0%
8	Chip		1436	207	234	5	30%	90	20%
9	Frenchgamerlol	1421	236	249	4	38%	46	25%
10	HarleyK		1413	255	307	4	25%	150	0%
11	Scurious	1405	213	212	4	50%	-35	50%
12	Wolverine	1398	280	448	3	0%	218	0%
13	Pseudo		1381	313	461	2	0%	159	0%
14	U0		1378	316	462	2	0%	170	0%
15	White Dove	1355	241	389	4	0%	182	0%
16	Mystery		1271	269	400	3	0%	83	0%
17	Midecah		1258	246	393	4	0%	89	0%

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

HasiLover_Test wrote:

GoK(white) wins against Scurious 2 in 41 Moves :https://lichess.org/VC2cvCUT#81 The new GoK Version played at an Impressive 97% accuracy.

Thanks, good to see GoK castled at the right time. GoK Medium also played a good game against WD already: https://lichess.org/geJ2aqvs#129

Game #2 against WD was similar, and as before, the one blunder does not happen on Difficult, search-depth related: https://lichess.org/rGFuqJL6#143

I left the aggressive LMR activated for Scratch 3, it has some advantages there, think time among them.

You can see the changeset here: https://github.com/ArnoHue/scratch/commit/b799c201c91b76ca907286de9562c8e3862fb8bc

Game #3 ( https://lichess.org/vjRjyTAu#102 ) was also good, although GoK left the shortest checkmate path twice due to a refactoring bug I had made on a constant usage, this is the bugfix:

https://github.com/ArnoHue/scratch/commit/70b9730e26b08fac9e681dda1735e9ece607b10e

github is such a big help!

Also, castling timing was not perfect, tweaked just one evaluation factor, and fixed.

Game #4 was played on Medium, nearly perfect, 96% vs 90%, two accuracies, one of which does not happen on Difficult: https://lichess.org/hou6oFy4#101

S_P_A_R_T wrote:

ArnoHu wrote:

HasiLover wrote:

ArnoHu wrote:

OK, whatever we might say about Archimedes Chess, but this old Swashbuckler of a Scratch Chess Engine won against White Dove (P2) after having faced mate in 3, won against Shallow Blue (Depth 2) after having faced mate in 6, and drew against Scurious (Ply 4) after having faced mate in 2, in just three consecutive games. True survival skills!

Those are chapters 42 - 44 in our Scratch 3 Runtime study: https://lichess.org/study/v3EKTlR2

It will take quite some wins to replace it from place 4. I think I will give Frenchgamerlol the next run, but who knows how this might end…

Rank	Name		Elo	+	-	games	score	oppo.	draws
1	GoK		1756	164	128	20	100%	1	0%
2	Element		1653	312	277	3	67%	126	0%
3	Bonsai		1602	170	160	10	60%	106	0%
4	Archimedes	1558	199	184	7	64%	50	14%
5	The Turk	1479	211	240	6	33%	128	0%
6	Shallow Blue	1475	232	232	4	50%	45	0%
7	LowDoor		1473	241	237	4	50%	50	0%
8	Chip		1435	205	231	5	30%	88	20%
9	Scurious	1427	213	212	4	50%	-13	50%
10	HarleyK		1397	258	311	4	25%	137	0%
11	Pseudo		1395	308	460	2	0%	165	0%
12	Wolverine	1393	279	446	3	0%	211	0%
13	U0		1374	315	462	2	0%	164	0%
14	White Dove	1371	238	383	4	0%	192	0%
15	Frenchgamerlol	1348	288	305	3	33%	25	0%
16	Midecah		1280	244	387	4	0%	105	0%
17	Mystery		1234	269	386	3	0%	50	0%

It seems Archimides is just Nr. 4 on Scratch I guess

Haha, yes. And I am not saying anything any more, but here is Frenchgamerlol vs. Archimedes: https://lichess.org/study/v3EKTlR2/nEUUO34Y

I think the reason we they're isn't an analysis bar is because it's disabled in the study. (Normally it would show up in other studies.)

ArnoHu wrote:

S_P_A_R_T wrote:

ArnoHu wrote:

HasiLover wrote:

ArnoHu wrote:

OK, whatever we might say about Archimedes Chess, but this old Swashbuckler of a Scratch Chess Engine won against White Dove (P2) after having faced mate in 3, won against Shallow Blue (Depth 2) after having faced mate in 6, and drew against Scurious (Ply 4) after having faced mate in 2, in just three consecutive games. True survival skills!

Those are chapters 42 - 44 in our Scratch 3 Runtime study: https://lichess.org/study/v3EKTlR2

It will take quite some wins to replace it from place 4. I think I will give Frenchgamerlol the next run, but who knows how this might end…

Rank	Name		Elo	+	-	games	score	oppo.	draws
1	GoK		1756	164	128	20	100%	1	0%
2	Element		1653	312	277	3	67%	126	0%
3	Bonsai		1602	170	160	10	60%	106	0%
4	Archimedes	1558	199	184	7	64%	50	14%
5	The Turk	1479	211	240	6	33%	128	0%
6	Shallow Blue	1475	232	232	4	50%	45	0%
7	LowDoor		1473	241	237	4	50%	50	0%
8	Chip		1435	205	231	5	30%	88	20%
9	Scurious	1427	213	212	4	50%	-13	50%
10	HarleyK		1397	258	311	4	25%	137	0%
11	Pseudo		1395	308	460	2	0%	165	0%
12	Wolverine	1393	279	446	3	0%	211	0%
13	U0		1374	315	462	2	0%	164	0%
14	White Dove	1371	238	383	4	0%	192	0%
15	Frenchgamerlol	1348	288	305	3	33%	25	0%
16	Midecah		1280	244	387	4	0%	105	0%
17	Mystery		1234	269	386	3	0%	50	0%

It seems Archimides is just Nr. 4 on Scratch I guess

Haha, yes. And I am not saying anything any more, but here is Frenchgamerlol vs. Archimedes: https://lichess.org/study/v3EKTlR2/nEUUO34Y

I think the reason we they're isn't an analysis bar is because it's disabled in the study. (Normally it would show up in other studies.)

Thanks, yes, I understand that might be the case for a study - but I also experience that after I import PGN data into an analysis board, without any settings at all.

S_P_A_R_T wrote:

White Dove's strange S3 bug has finally been found!

Due to the terrible S3 speed, and a minor code oversight that never caused any problems during TW, WD was actually so slow, that it didn't even finish evaluating the FIRST move of the lowest depth!

This was caused by the enormous amount of extensions, which sometimes caused WD to get to ply-8 (on Scratch 3!). This removed all performance during more tactical lines.

The new version should hopefully fix this, improve the opening book, and tweak some evaluation parameters.

Though this will take some time, I'm pretty confident this will have a massive impact on WDs S3 performance, so I want to make sure I do it right

(Also @ArnoHu , can you turn on exporting & eval in the study, as this makes it much easier to test & debug, thanks!)

ArnoHu wrote:

S_P_A_R_T wrote:

White Dove's strange S3 bug has finally been found!

Due to the terrible S3 speed, and a minor code oversight that never caused any problems during TW, WD was actually so slow, that it didn't even finish evaluating the FIRST move of the lowest depth!

This was caused by the enormous amount of extensions, which sometimes caused WD to get to ply-8 (on Scratch 3!). This removed all performance during more tactical lines.

The new version should hopefully fix this, improve the opening book, and tweak some evaluation parameters.

Though this will take some time, I'm pretty confident this will have a massive impact on WDs S3 performance, so I want to make sure I do it right

(Also @ArnoHu , can you turn on exporting & eval in the study, as this makes it much easier to test & debug, thanks!)

Great to hear! Does this also cover the disappearing pieces? I have activated export and analysis for members, which as far as I understand means you only have to be logged in. I had run too many analysis yesterday and reached my daily quota.

I let White Dove and GoK play one more match on S3, and it was great to watch (91% vs. 97% accuracy)! Nice level for two engines running at 15 and 10 seconds think time: https://lichess.org/study/v3EKTlR2/w693R8Ri . White Dove will certainly move up the ranks quickly to a position where it belongs.

Funny details about lichess analysis, when importing PGN, lichess showed 3 resp. 1 blunders. But when loading the same PGN into the study, those blunders were downgraded to inaccuracies, white the overall accuracy nearly staying the same. Different SF search depth I guess?