Torch V1 - Chess324 Testing Results

https://ccrl.chessdom.com/ccrl/Chess324/index.html

Just to recap, this is a Top 15 revolving round-robin tournament using SPCC's Anti-Draw Chess324 opening book - SPCC 3-move Chess324_xxl_big_+090_+119. The book contains 70355 lines. Each pairing plays 100 openings reversed sides, so 200 games per pairing. Thanks to Stefan for his excellent work in creating these Chess324 books.

Revolving means that when a new engine version comes out, the old one is removed and the new one added. In this case also, because Torch came in as a new Top 15 engine, Seer at the bottom fell out.

This tournament was first published back in August and been being updated since then.


Torch is the #2 engine under these test conditions.

Thanks Ray...

Stockfish 20230813 vs Torch v1 : 127 - 73 = 36.5%

Elo difference = 52

Great start but a long way to go.

Yes it underperformed against Stockfish compared to its overall rating. In an online tournament maybe an alteration to the default contempt setting could help.

Ray wrote:

Yes it underperformed against Stockfish compared to its overall rating. In an online tournament maybe an alteration to the default contempt setting could help.

Thanks for the list but the data that I read in this discussion that Torch underperformed against Stockfish is misleading .

I saw from the table the following:
1)Torch underperformed by 52 elo against Stockfish
2)Torch is only 26 elo stonger than Dragon3.3

My wrong conjecture based on the data was that Torch is probably stronger against weak engines relative to Dragon but not against strong engines and that Torch scored less points against Stockfish but more points against the weak engines so Torch got a bigger elo.

I found that it is wrong by looking at the data
Torch did better than Dragon3.3 against Stockfish.
Torch lost 127-73 when Dragon3.3 lost 132-68.


I think that the problem is not with Torch.
The problem is with stockfish that in a systematical way underperform against the weaker engines and overperform against the stronger engines
maybe because it does not know to play for a win with black.

Yes that is true, not a Torch issue, it is Stockfish behaviour.

Uri Blass wrote:

Ray wrote:

Yes it underperformed against Stockfish compared to its overall rating. In an online tournament maybe an alteration to the default contempt setting could help.

Thanks for the list but the data that I read in this discussion that Torch underperformed against Stockfish is misleading .

I saw from the table the following:
1)Torch underperformed by 52 elo against Stockfish
2)Torch is only 26 elo stonger than Dragon3.3

My wrong conjecture based on the data was that Torch is probably stronger against weak engines relative to Dragon but not against strong engines and that Torch scored less points against Stockfish but more points against the weak engines so Torch got a bigger elo.

I found that it is wrong by looking at the data
Torch did better than Dragon3.3 against Stockfish.
Torch lost 127-73 when Dragon3.3 lost 132-68.


I think that the problem is not with Torch.
The problem is with stockfish that in a systematical way underperform against the weaker engines and overperform against the stronger engines
maybe because it does not know to play for a win with black.

Agreed, SFish plays too solidly with Black, as if not to lose.

Stockfish is also very strong at FRC, and given that Chess324 is a subset of DFRC, I guess some of that strengths carries over.