Single Comment

This is starting to get very mysterious. Whatever I do, it seems impossible
to create conditions where Guanacos convincingly beat Llamas, between equal
programs. The scores always converge around 51-53% in favor of the Llamas,
so far. There is an error of around 3-4% on this, as it was all done with
about 100 games, but still... It is totally ridiculous outcome, as the
Guanaco is obviously upward compatible.

What I did was this: First I established that it is a much better strategy
to centralize Guanacos than to leave them roam the board neutrally: A
version that centralizes wins symmetric positions where each side has two
Guanacos (replacing the Knights) very convincingly (66% score in 100
games).

Nevertheless, with this better Guanaco handling, the score of Guanacos
against Llamas (between identical programs) remained around 50%, like it
was with the poor Guanaco handling. This is not unusual: The fact that the
program having the Guanacos knows that they are better in the center is
largely offset by the fact that its opponent now knows this too, and tries
to keep them out of the center. I did this test with a value of 240 for
Guanaco and 180 for Llama (on a scale Bishop = 296).

So my first thought was that the high programmed value of the Guanaco was
hindering its proper use, by making the program too scared to risk it being
traded for other material. So I repeated the test with rogrammed values 180
and 190. But after 100 games the Llamas now lead there by 52%.

The only thing I can think off at the moment to explain this is that the
extra Guanaco moves (compared to the Llama) are nearly useless, but that
the fact that they are searched increases the branching ratio of the search
tree, reducing its depth. The side having the Guanacos suffers more from
this than its opponent, as at odd ply depth it searches one more level in
the tree with Guanaco moves than the program with the Llamas. But I cannot
imagine that this effect is very big. (There are not that many distant
Guanaco moves, compared to total moves in a typical positon.)

Anyway, it seems that the Guanaco hardly offers any advantage over the
Llama. Which is a bit surprising. I also did another test with to Guanacos
as Knight replacements on one side, between Guanaco-centtralizing engines,
and the Guanacos get clobbered by ~73% so far.
I will let these test runs overnight to get somewhat better statistics,
but everything so far points to nearly equal Guanaco and Llama value, both
significantly weaker than a Knight.