What kind of B.S. is CCRL trying to pull here!

CCRL has added Lc0 0.28.0 to their rating list. But tested Lc0 with the much inferior t40-1541 neural net.
This could have only been done to suppress Lc0's rating, or they are just incompetent.

Even the default net packaged with Lc0 0.28.0 is much better then t40-1541. Why did they not at least use the default net.....Hmmm.

If you look further they have used the t40-1541 net since version 25.1

I have used 609992.pb and it gave excellent results.

Admin wrote:

If you look further they have used the t40-1541 net since version 25.1

I have used 609992.pb and it gave excellent results.

I know and I have pointed this out to them in the past. The bad testing of Lc0. At some point you realize CCRL is doing this intentionally.
As there is no justification, reason, or anything else to use a very old, and weak neural net to test Lc0.

And now the #2 engine in the world if tested with the correct neural net. If you disregard the other Stockfish on their list Fat Frtiz 2.

No other engine gets such bad treatment on CCRL as Lc0.

I will put it on talkchess.

It's serious enough.

mwyoung wrote:

Admin wrote:

If you look further they have used the t40-1541 net since version 25.1

I have used 609992.pb and it gave excellent results.

I know and I have pointed this out to them in the past. The bad testing of Lc0. At some point you realize CCRL is doing this intentionally.
As there is no justification, reason, or anything else to use a very old, and weak neural net to test Lc0.

And now the #2 engine in the world if tested with the correct neural net. If you disregard the other Stockfish on their list Fat Frtiz 2.

No other engine gets such bad treatment on CCRL as Lc0.

Exactly as predicted with my testing. Lc0 is now the #2 behind the 2 Stockfish engines.