Scratch Chess Engine - Game of Kings

Destructor_chess

Wich think time for this depth? 5 sec?

waabooboo wrote:
New Wolverine version out, with some more improvements to the transposition table. This helps endgame search depth quite a bit.

https://imgur.com/a/fZhYKqb – I think Wolverine will not lose endgames like this any more

10?

waabooboo

Destructor_chess wrote:
Wich think time for this depth? 5 sec?
waabooboo wrote:
New Wolverine version out, with some more improvements to the transposition table. This helps endgame search depth quite a bit.

https://imgur.com/a/fZhYKqb – I think Wolverine will not lose endgames like this any more
10?

5 seconds, yes. Still around 200k nps, but it's a little slower than before because I generate moves before applying TT evals (just to ensure legality). For now, known wins from TT are +15000 and actual mates are 100000-ply.

waabooboo

ArnoHu wrote:
waabooboo wrote:
New Wolverine version out, with some more improvements to the transposition table. This helps endgame search depth quite a bit.

https://imgur.com/a/fZhYKqb – I think Wolverine will not lose endgames like this any more

Plays pretty well, esp. midgame:
https://lichess.org/study/7cqWfi6q/NjTsbPil

Nice, good to see! Midgame was back and forth there, looks like Wolverine had several chances to hold. But I can't blame him, that position is really complicated.

HasiLover

I am going to run a vote chess game between 5 scratch engines(GoK Classic, WD, Delta, SB and Amarillo) and SF 17.1, here is the broadcast link if you want to watch: https://lichess.org/broadcast/scratch-vs-stockfish/scratch-engines-stockfish-game/B951qxpR/aLg2ELab

Destructor_chess

waabooboo wrote:
Destructor_chess wrote:
Wich think time for this depth? 5 sec?
waabooboo wrote:
New Wolverine version out, with some more improvements to the transposition table. This helps endgame search depth quite a bit.

https://imgur.com/a/fZhYKqb – I think Wolverine will not lose endgames like this any more
10?

5 seconds, yes. Still around 200k nps, but it's a little slower than before because I generate moves before applying TT evals (just to ensure legality). For now, known wins from TT are +15000 and actual mates are 100000-ply.

Known wins are 15,000 and… 2 things: I guess you works with negamax and second thing, you saves a win score, is it a mate score, definitely? And how do you know in how much moves is it?

internet44

HasiLover wrote:
I am going to run a vote chess game between 5 scratch engines(GoK Classic, WD, Delta, SB and Amarillo) and SF 17.1, here is the broadcast link if you want to watch: https://lichess.org/broadcast/scratch-vs-stockfish/scratch-engines-stockfish-game/B951qxpR/aLg2ELab

nice to see a broadcast again, it has been way too long

even if we didn't really have a chance lol, just fun to hang out and chat live

HasiLover

The game has concluded, SF 17.1 won with 99% accuracy against the 94% of the Scratch Engines: https://lichess.org/broadcast/scratch-vs-stockfish/scratch-engines-stockfish-game/B951qxpR/aLg2ELab

waabooboo

Destructor_chess wrote:
waabooboo wrote:
Destructor_chess wrote:
Wich think time for this depth? 5 sec?
waabooboo wrote:
New Wolverine version out, with some more improvements to the transposition table. This helps endgame search depth quite a bit.

https://imgur.com/a/fZhYKqb – I think Wolverine will not lose endgames like this any more
10?

5 seconds, yes. Still around 200k nps, but it's a little slower than before because I generate moves before applying TT evals (just to ensure legality). For now, known wins from TT are +15000 and actual mates are 100000-ply.
Known wins are 15,000 and… 2 things: I guess you works with negamax and second thing, you saves a win score, is it a mate score, definitely? And how do you know in how much moves is it?

Yes, win score is definitely a mate. But it's a bit complicated to get that out of the hash table – that mate score could be after several unnecessary transpositions. So mates from the TT are always worth less, that way when Wolverine finds an actual mate, I have all the moves from start to finish.

HasiLover

I am running a 5 round Swiss tourney, with 2 sec per move: https://lichess.org/broadcast/scratch-bullet-swiss-2025/round-1/CTgbgo8b/liAb0zru

waabooboo

HasiLover wrote:
I am running a 5 round Swiss tourney, with 2 sec per move: https://lichess.org/broadcast/scratch-bullet-swiss-2025/round-1/CTgbgo8b/liAb0zru

2 sec per move should be good for Wolverine, he does well under time pressure due to high nps compared to his peers. Looking forward to the results

HasiLover

Round 2 is continuing now: https://lichess.org/broadcast/scratch-bullet-swiss-2025/round-2/SGM4bluh#boards

HasiLover

Running Round 3 now

https://lichess.org/broadcast/scratch-bullet-swiss-2025/round-3/lzDElvpv/RF4nthla

HasiLover

Round 4 is starting: https://lichess.org/broadcast/scratch-bullet-swiss-2025/round-4/dC2kbPtm/Cxw5KlQL There are some intense matches incoming, like BC-GoK NNUE

HasiLover

Final Round starting now! https://lichess.org/broadcast/scratch-bullet-swiss-2025/final-round5/ZoUqCsDU/dYQuGHYB

ArnoHu

HasiLover wrote:
Final Round starting now! https://lichess.org/broadcast/scratch-bullet-swiss-2025/final-round5/ZoUqCsDU/dYQuGHYB

Great tournament with some surprising results!

ArnoHu

GoK Classic vs. Wolverine 2, 92% vs. 88%:
https://lichess.org/study/7cqWfi6q/gPgNzMtF

GoK NNUE vs. Black Crow, 96% vs. 92%:
https://lichess.org/study/7cqWfi6q/QdHJu3qP

Destructor_chess

ArnoHu wrote:
HasiLover wrote:
Final Round starting now! https://lichess.org/broadcast/scratch-bullet-swiss-2025/final-round5/ZoUqCsDU/dYQuGHYB

Great tournament with some surprising results!

The more surprising is gok classic vs tk lol (Ik you said it was a regression but….)

ArnoHu

Destructor_chess wrote:
ArnoHu wrote:
HasiLover wrote:
Final Round starting now! https://lichess.org/broadcast/scratch-bullet-swiss-2025/final-round5/ZoUqCsDU/dYQuGHYB

Great tournament with some surprising results!
The more surprising is gok classic vs tk lol (Ik you said it was a regression but….)

Yes, I refactored move generator variable names, and made a mistake, which caused a move generator bug. That doesnt change anything about the win. GoK Classic would have had a difficult time anyway, with two seconds on the clock and running on a slow system.

waabooboo

ArnoHu wrote:
GoK Classic vs. Wolverine 2, 92% vs. 88%:
https://lichess.org/study/7cqWfi6q/gPgNzMtF

GoK NNUE vs. Black Crow, 96% vs. 92%:
https://lichess.org/study/7cqWfi6q/QdHJu3qP

Thanks, there's an interesting case! Around move 35 when Wolverine is shuffling the rook back and forth, Wolverine gives evals of only around +0.5 for white – despite the fact that all three of black's pieces are completely paralyzed. It takes a while for white to win material by force, so Wolverine doesn't recognize the significance of it.

Dev version of Wolverine wins most games against me now – it will be interesting to see how he does against GoK in short time formats. Next release is coming soon…

Last edited by waabooboo (Yesterday 19:15:24)

ArnoHu

ArnoHu wrote:
GoK Classic vs. Wolverine 2, 92% vs. 88%:
https://lichess.org/study/7cqWfi6q/gPgNzMtF

GoK NNUE vs. Black Crow, 96% vs. 92%:
https://lichess.org/study/7cqWfi6q/QdHJu3qP

GoK Classic vs. White Dove, 98% vs. 91%:
https://lichess.org/study/7cqWfi6q/RLhSw609

GoK Classic vs. Delta, 95% vs. 89%:
https://lichess.org/study/7cqWfi6q/1jBXogfP

GoK Classic vs. Shallow Blue 3 Dev, 92% vs. 90%:
https://lichess.org/study/7cqWfi6q/Hi0kJdzd

Last edited by ArnoHu (Today 04:57:56)

Discuss Scratch