Scratch Chess Engine - Game of Kings

HasiLover

I just ran games against Scratch Chess engines with 10+0 think time, I lost to BC in the endgame and drew amarillo 4, this means that Scratch Engines on my system are probably in the 2000 to 2300 glicko range. Considering you all have much more faster systems, I believe the strongest Scratch Chess Engines could have a Glicko of around 2500 maybe

Also, I am so impressed by my Game with Amarillo 4, both played 95% accuracy: https://lichess.org/1DjLU7HM#0

Last edited by HasiLover (March 4, 2026 20:56:01)

HasiLover

ArnoHu wrote:
HasiLover wrote:
ArnoHu wrote:
ArnoHu wrote:
ArnoHu wrote:
ArnoHu wrote:
GoK Classic vs. White Dove, 99% vs. 93%:
https://lichess.org/study/vXPhQaqz/EYrgmaoc

GoK Classic vs. Wolverine 2, 99% vs. 91%:
https://lichess.org/study/vXPhQaqz/xBiAN7Ui

GoK NNUE vs. Black Crow, 92% vs. 87%:
https://lichess.org/study/vXPhQaqz/tiSLUf4Z

GoK NNUE vs. Black Crow, 92% vs. 94%:
https://lichess.org/study/vXPhQaqz/JbGSWgKA

GoK NNUE vs. Black Crow, 92% vs. 85%:
https://lichess.org/study/vXPhQaqz/u8RC1BPV

GoK NNUE vs. Black Crow, 98% vs. 93%:
https://lichess.org/study/vXPhQaqz/201TRU1L

GoK NNUE vs. Black Crow, 96% vs. 91%:
https://lichess.org/study/vXPhQaqz/77wfkNEv

GoK Classic vs. White Dove, 93% vs. 91%:
https://lichess.org/study/vXPhQaqz/TIvkL6wf

GoK Classic vs. Wolverine 2, 98% vs. 95%:
https://lichess.org/study/vXPhQaqz/vxICGc9P

GoK NNUE vs. Black Crow, 96% vs. 91%:
https://lichess.org/study/vXPhQaqz/aXZv2qzc

GoK NNUE vs. Black Crow, 97% vs. 91%:
https://lichess.org/study/vXPhQaqz/M8iNZOAt

GoK Classic vs. Wolverine 2, 84% vs. 70%:
https://lichess.org/study/vXPhQaqz/bTtbd3lm

New GoK versions:

GoK NNUE vs. Black Crow, 98% vs. 93%:
https://lichess.org/study/vXPhQaqz/qVy5WlB6

GoK Classic vs. Wolverine 2, 93% vs. 90%:
https://lichess.org/study/vXPhQaqz/iXbvYA6W
About those new versions, GoK classic was looking at some awful blunders as its top moves, when I tested it. Are you sure the new version is stable?

The new version was just uploaded now. It fixed a pruning regression in the previous version.

Still you never know for sure, I only played a limited number of games.

I probably played the prior version in that case, maybe the new one fixed this bug already

ArnoHu

ArnoHu wrote:
ArnoHu wrote:
ArnoHu wrote:
ArnoHu wrote:
GoK Classic vs. White Dove, 99% vs. 93%:
https://lichess.org/study/vXPhQaqz/EYrgmaoc

GoK Classic vs. Wolverine 2, 99% vs. 91%:
https://lichess.org/study/vXPhQaqz/xBiAN7Ui

GoK NNUE vs. Black Crow, 92% vs. 87%:
https://lichess.org/study/vXPhQaqz/tiSLUf4Z

GoK NNUE vs. Black Crow, 92% vs. 94%:
https://lichess.org/study/vXPhQaqz/JbGSWgKA

GoK NNUE vs. Black Crow, 92% vs. 85%:
https://lichess.org/study/vXPhQaqz/u8RC1BPV

GoK NNUE vs. Black Crow, 98% vs. 93%:
https://lichess.org/study/vXPhQaqz/201TRU1L

GoK NNUE vs. Black Crow, 96% vs. 91%:
https://lichess.org/study/vXPhQaqz/77wfkNEv

GoK Classic vs. White Dove, 93% vs. 91%:
https://lichess.org/study/vXPhQaqz/TIvkL6wf

GoK Classic vs. Wolverine 2, 98% vs. 95%:
https://lichess.org/study/vXPhQaqz/vxICGc9P

GoK NNUE vs. Black Crow, 96% vs. 91%:
https://lichess.org/study/vXPhQaqz/aXZv2qzc

GoK NNUE vs. Black Crow, 97% vs. 91%:
https://lichess.org/study/vXPhQaqz/M8iNZOAt

GoK Classic vs. Wolverine 2, 84% vs. 70%:
https://lichess.org/study/vXPhQaqz/bTtbd3lm

New GoK versions:

GoK NNUE vs. Black Crow, 98% vs. 93%:
https://lichess.org/study/vXPhQaqz/qVy5WlB6

GoK Classic vs. Wolverine 2, 93% vs. 90%:
https://lichess.org/study/vXPhQaqz/iXbvYA6W

GoK Classic vs. White Dove, 96% vs. 93%:
https://lichess.org/study/vXPhQaqz/prprXVt7

GoK Classic vs. Black Crow, 99% vs. 99%:
https://lichess.org/study/vXPhQaqz/TC7vuLmN

GoK Classic vs. Wolverine 2, 99% vs. 90%:
https://lichess.org/study/Yq49IAcO/ImXrlMnZ

I think we are good…

Last edited by ArnoHu (March 5, 2026 22:03:48)

birdracerthree

Hasi was ahead of his time

ArnoHu

ArnoHu wrote:
ArnoHu wrote:
ArnoHu wrote:
ArnoHu wrote:
ArnoHu wrote:
GoK Classic vs. White Dove, 99% vs. 93%:
https://lichess.org/study/vXPhQaqz/EYrgmaoc

GoK Classic vs. Wolverine 2, 99% vs. 91%:
https://lichess.org/study/vXPhQaqz/xBiAN7Ui

GoK NNUE vs. Black Crow, 92% vs. 87%:
https://lichess.org/study/vXPhQaqz/tiSLUf4Z

GoK NNUE vs. Black Crow, 92% vs. 94%:
https://lichess.org/study/vXPhQaqz/JbGSWgKA

GoK NNUE vs. Black Crow, 92% vs. 85%:
https://lichess.org/study/vXPhQaqz/u8RC1BPV

GoK NNUE vs. Black Crow, 98% vs. 93%:
https://lichess.org/study/vXPhQaqz/201TRU1L

GoK NNUE vs. Black Crow, 96% vs. 91%:
https://lichess.org/study/vXPhQaqz/77wfkNEv

GoK Classic vs. White Dove, 93% vs. 91%:
https://lichess.org/study/vXPhQaqz/TIvkL6wf

GoK Classic vs. Wolverine 2, 98% vs. 95%:
https://lichess.org/study/vXPhQaqz/vxICGc9P

GoK NNUE vs. Black Crow, 96% vs. 91%:
https://lichess.org/study/vXPhQaqz/aXZv2qzc

GoK NNUE vs. Black Crow, 97% vs. 91%:
https://lichess.org/study/vXPhQaqz/M8iNZOAt

GoK Classic vs. Wolverine 2, 84% vs. 70%:
https://lichess.org/study/vXPhQaqz/bTtbd3lm

New GoK versions:

GoK NNUE vs. Black Crow, 98% vs. 93%:
https://lichess.org/study/vXPhQaqz/qVy5WlB6

GoK Classic vs. Wolverine 2, 93% vs. 90%:
https://lichess.org/study/vXPhQaqz/iXbvYA6W

GoK Classic vs. White Dove, 96% vs. 93%:
https://lichess.org/study/vXPhQaqz/prprXVt7

GoK Classic vs. Black Crow, 99% vs. 99%:
https://lichess.org/study/vXPhQaqz/TC7vuLmN

GoK Classic vs. Wolverine 2, 99% vs. 90%:
https://lichess.org/study/Yq49IAcO/ImXrlMnZ

I think we are good…

Fixed a regression:

GoK NNUE vs. Black Crow, 97% vs. 91%:
https://lichess.org/study/Yq49IAcO/6xr9fVqv

GoK Classic vs. Wolverine 2, 98% vs. 95%:
https://lichess.org/study/Yq49IAcO/J1gPV08i

HasiLover

Turns out the Amarillo tuner found a bug before I did

The CorHist had a bug that caused massive false positives/negatives, thats why the tuner learned to not use CorHist. I just fixed the bug, ran more tuning games and I think Amarillo 4 is ready for release

ArnoHu

GoK Classic vs. White Dove, 98% vs. 92%:
https://lichess.org/study/Yq49IAcO/bLycGeYz

GoK Classic vs. Wolverine 2, 97% vs. 90%:
https://lichess.org/study/Yq49IAcO/4w2wXHk3

GoK NNUE vs. Black Crow, 96% vs. 93%:
https://lichess.org/study/Yq49IAcO/EMAZ15qZ

ArnoHu

HasiLover wrote:
Turns out the Amarillo tuner found a bug before I did The CorHist had a bug that caused massive false positives/negatives, thats why the tuner learned to not use CorHist. I just fixed the bug, ran more tuning games and I think Amarillo 4 is ready for release

Congrats! Will you post here once it is available?

HasiLover

ArnoHu wrote:
HasiLover wrote:
Turns out the Amarillo tuner found a bug before I did The CorHist had a bug that caused massive false positives/negatives, thats why the tuner learned to not use CorHist. I just fixed the bug, ran more tuning games and I think Amarillo 4 is ready for release

Congrats! Will you post here once it is available?

I am gonna release it in a few minutes, after some testing I can confirm that Amarillo 4 can compete with Shallow Blue 3 and may even be a bit stronger

Edit: Its out!

Last edited by HasiLover (March 6, 2026 21:57:13)

ArnoHu

HasiLover wrote:
ArnoHu wrote:
HasiLover wrote:
Turns out the Amarillo tuner found a bug before I did The CorHist had a bug that caused massive false positives/negatives, thats why the tuner learned to not use CorHist. I just fixed the bug, ran more tuning games and I think Amarillo 4 is ready for release

Congrats! Will you post here once it is available?
I am gonna release it in a few minutes, after some testing I can confirm that Amarillo 4 can compete with Shallow Blue 3 and may even be a bit stronger
Edit: Its out!

How can I play a 5sec/move game?

HasiLover

ArnoHu wrote:
HasiLover wrote:
ArnoHu wrote:
HasiLover wrote:
Turns out the Amarillo tuner found a bug before I did The CorHist had a bug that caused massive false positives/negatives, thats why the tuner learned to not use CorHist. I just fixed the bug, ran more tuning games and I think Amarillo 4 is ready for release

Congrats! Will you post here once it is available?
I am gonna release it in a few minutes, after some testing I can confirm that Amarillo 4 can compete with Shallow Blue 3 and may even be a bit stronger
Edit: Its out!

How can I play a 5sec/move game?

Click on the button that looks like a chess clock(The one under “play”)

ArnoHu

HasiLover wrote:
ArnoHu wrote:
HasiLover wrote:
ArnoHu wrote:
HasiLover wrote:
Turns out the Amarillo tuner found a bug before I did The CorHist had a bug that caused massive false positives/negatives, thats why the tuner learned to not use CorHist. I just fixed the bug, ran more tuning games and I think Amarillo 4 is ready for release

Congrats! Will you post here once it is available?
I am gonna release it in a few minutes, after some testing I can confirm that Amarillo 4 can compete with Shallow Blue 3 and may even be a bit stronger
Edit: Its out!

How can I play a 5sec/move game?
Click on the button that looks like a chess clock(The one under “play”)

OK I saw that. 5sec seems to be default anyway, right?

Here is a first game against Classic, Classic won at 96% vs. 92%, nice accuracy:
https://lichess.org/study/Yq49IAcO/rC0O7G3r

But also a bug, Amarillo didn't allow black's completely valid move 34, and I had to re-import the board.

HasiLover

ArnoHu wrote:
HasiLover wrote:
ArnoHu wrote:
HasiLover wrote:
ArnoHu wrote:
HasiLover wrote:
Turns out the Amarillo tuner found a bug before I did The CorHist had a bug that caused massive false positives/negatives, thats why the tuner learned to not use CorHist. I just fixed the bug, ran more tuning games and I think Amarillo 4 is ready for release

Congrats! Will you post here once it is available?
I am gonna release it in a few minutes, after some testing I can confirm that Amarillo 4 can compete with Shallow Blue 3 and may even be a bit stronger
Edit: Its out!

How can I play a 5sec/move game?
Click on the button that looks like a chess clock(The one under “play”)

OK I saw that. 5sec seems to be default anyway, right?

Here is a first game against Classic, Classic won at 96% vs. 92%, nice accuracy:
https://lichess.org/study/Yq49IAcO/rC0O7G3r

But also a bug, Amarillo didn't allow black's completely valid move 34, and I had to re-import the board.

Thanks for running a game! I have never noticed a UI movegen bug in Amarillo before, thanks for the notice! I am quite happy with the 92% accuracy performance by Amarillo 4, since that is a big improvement from the prior version.

ArnoHu

ArnoHu wrote:
GoK Classic vs. White Dove, 98% vs. 92%:
https://lichess.org/study/Yq49IAcO/bLycGeYz

GoK Classic vs. Wolverine 2, 97% vs. 90%:
https://lichess.org/study/Yq49IAcO/4w2wXHk3

GoK NNUE vs. Black Crow, 96% vs. 93%:
https://lichess.org/study/Yq49IAcO/EMAZ15qZ

GoK Classic vs. Amarillo 4, 96% vs. 92%:
https://lichess.org/study/Yq49IAcO/rC0O7G3r

GoK Classic vs. Shallow Blue 3, 93% vs. 86%:
https://lichess.org/study/Yq49IAcO/ICZSGhbI

GoK Classic vs. Delta, 96% vs. 91%:
https://lichess.org/study/Yq49IAcO/yNiI7ejm

GoK Classic vs. TurboKnight 5, 98% vs. 90%:
https://lichess.org/study/Yq49IAcO/ACgQsEvh

GoK NNUE vs. Black Crow, 97% vs. 91%:
https://lichess.org/study/Yq49IAcO/ZnLFpGUX

The overall Scratch chess engine quality at 5sec TurboWarp games is really amazing! Also Shallow Blue played a perfect game for 49 moves, before it started blundering during endgame.

Last edited by ArnoHu (March 7, 2026 00:32:12)

iceysnowman

I was really excited to see the Amarillo update (I've been checking in every day to see when it would finally come out hahaha)! Great to see it and looks wonderful! I randomly ran a test game of it vs Turboknight on turbowarp 5 second game. Didn't go so well but I think it's because my system is terrible (school chromebook T-T) (also I probably set something up wrong hahaha): https://lichess.org/study/b0xzzwZ9/XwLEZFmi.
Either way great to see the update out and to see you back again Hasi!

ArnoHu

This game had it all - and at the end BC was saved because Classic thinks KB-KP is a great position to be in

GoK Classic vs. Black Crow, 90% vs. 91%:
https://lichess.org/study/Yq49IAcO/Q6HcInRT

ArnoHu

iceysnowman wrote:
I was really excited to see the Amarillo update (I've been checking in every day to see when it would finally come out hahaha)! Great to see it and looks wonderful! I randomly ran a test game of it vs Turboknight on turbowarp 5 second game. Didn't go so well but I think it's because my system is terrible (school chromebook T-T) (also I probably set something up wrong hahaha): https://lichess.org/study/b0xzzwZ9/XwLEZFmi.
Either way great to see the update out and to see you back again Hasi!

Can't open the study, most likely you have to edit visibility.

Destructor_chess

iceysnowman wrote:
I was really excited to see the Amarillo update (I've been checking in every day to see when it would finally come out hahaha)! Great to see it and looks wonderful! I randomly ran a test game of it vs Turboknight on turbowarp 5 second game. Didn't go so well but I think it's because my system is terrible (school chromebook T-T) (also I probably set something up wrong hahaha): https://lichess.org/study/b0xzzwZ9/XwLEZFmi.
Either way great to see the update out and to see you back again Hasi!

I don ́t agree, why against turboknight… it didn’t ask anything

btw, the link isn’t working on my side

HasiLover

I should 100% run a new chess engine tournament, maybe with 10sec/move?

Destructor_chess

https://lichess.org/Ol1ykpx0 TK white 5 sec vs Amarillo 5 sec

btw Hasi, arno will a 5 sec tournament https://lichess.org/broadcast/spring-2026-scratch-chess-engine-tournament/scet-spring-2026-round-1/YHpCYQDz

Last edited by Destructor_chess (March 7, 2026 08:48:02)

Discuss Scratch