Comments/Ratings for a Single Item

Joe, these pieces can also make a non-capturing move to capture square. The Alpaca is interesting because it can sometimes invade the enemy position by jumping, and drive away enemy pieces while they have a higher value. However, it is a common theme that a light piece is exchanged for Alpaca + pawn. Pieces valued 2 are interesting, as long as they are short-range, it seems. I have now also implemented the Vicuña, which captures only on the second square, and jumps continuously. A peculiarity is that the Guanaco can continue jumping although it has a capture possibility on the first square. This is wholly logical. I decided to call this category Lama pieces, after genus 'Lama'. I don't know how good they are, but they can be tested in my programs. They are described here:
http://hem.passagen.se/melki9/llamapieces.htm
/Mats

OK, I started some tests of the Lama pieces. As they are regular Chess
pieces that fit in the Betza classification scheme, Fairy-Max can handle
them without modification. Both Lama and Alpaca are handicapped versions of
the Woody Rook (Betza WD). I tested this Woody Rook in the context of Great
Shatranj, and IIRC it tested as slightly less than a Knight. From my
current understanding of short-range leaper values, WD should have been one
of the weakest of all 8-target leapers, due to the small number of forward
moves. And forward moves seem to contribute about twice as much to the
overall value as backward or sideway moves. But I guess the global property
of having mating potenial is worth some extra bonus.

The Lama and Alpaca should be practicaly equal in value; moves of 1 step
seem to contribute to the value of a piece about as much as moves of 2
steps in the same direction. If I would have to do a prediction, I woud use
the formula Value = (30+5/8*N)*N, which would put 8-target SR leapers at
280 (centi-Pawn). I would count non-capture moves in the second factor N
only for 1/3, though, so that we would get 35*16/3 = 187 cP.

For rider-like pieces like the Guanaco I have no theory yet.

I started tests by replacing the Knights of one side by Lamas, to see how
badly they are beaten. I will then probably switch to replacing two Knights
by 3 Lamas.

On the other core of my C2D I am playing a substitution of 2 Bishops by 2
Guanacos. I will keep you posted.

(I cannot promise I will be able to complete all necessary tests shortly,
As I might need the computer to test my Xiangqi engine HaQiKi D, which I
entered in the World Championship in Pamplona, taking place early May. And
I still have a lot to improve on that before I will have any chance on
winning there...)

Thanks for this. One factor which is very important in the Guanaco and the
Alpaca is that they become quite strong in endgames with pawns. They can
rather easily attack a pawn chain while they can move to an adjacent square
of a different colour, and hop over pawns. To attack an enemy pawn chain
could be impossible for a bishop of the wrong colour, and be somewhat
difficult for a knight, while it has to maneuvre many moves. I found that
Zillions has difficulties handling the Guanaco in the beginning while it
demands planning. You must maeuvre it in order to use it, so it's not the
ideal piece for computers. Therefore one cannot expect that the replacement
test will work very well. The Guanaco is very strong in the endgame, and
two Guanacos plus king seem able to checkmate an enemy king (needs to be
verified). If this is true then two Alpacas would also be able, together
with the king, to give mate. The Llama has fewer capture squares than the
Alpaca. Only on the inner 6x6 square it has four capture squares. So I
suppose it's worth somewhat less. On the other hand it can threaten at a
two square distance. Notice also that the Guanaco *slides* on the squares
on which it stands. In a sense, it is an orthogonal bishop.
/Mats

Hmm, I am starting to doubt that the Guanaco is anywhere near a Bishop in
value. In the match of two Guanacos vs two Knights the Knights lead by 80%.
Strange thing is that two Llamas vs one Knight seems to go towards a very
clear victory for the Llamas (64%, which suggests a difference of the order
of a Pawn). All this after a mere 100-150 games, so with a sizable error
bars. But these results would put a Guanaco quite close to a Llama, which
is a bit strange, as it is clearly upward compatible with it.

Perhaps I should start with a test of two Guanacos against two Llamas, (say
as Knight replacements), to get a direct measure of how much the difference
between the two really is. It might be that the distant non-captures are
pretty useless. Thee are only few squares from which you can actually make
them on 8x8, and perhaps the Llama is already fast enough.

One thing I can confirm straight away, though: Both two Llamas and two
Alpacas make a generally won end-game. (Maximum distance to mate 43 or
51 moves, respectively.)

It is obvious that the Guanaco is very strong in endgames with pawns, so
the results are strange. In a replacement test it's necessary to tweak the
value of the knight to be the same as the value of the Guanaco. Otherwise
the knights become dominant while they can chase away the Guanacos. The
Alpaca and the Llama have this advantage against the knight, too.
/Mats

Yes, it seems I have to go 'back to basics', and first make sure that the
program handles the pieces well. I was running the test where the Guanaco
did so poorly with programmed-in values that had the Guanaco worth slightly
less than a Knight (240 vs 259 in the peculiar micro-Max internal units
that have B=296). So the problem you mention could not have occurred.

I did not program the Guanaco as a piece that has to be centralized,
however. Fairy-Max knows two kind of pieces, those that are lightly) drawn
to the center (Kings, Knights and Bishops), and those that move neutrally
over the board (which works better for Rooks and Queens). I programmed the
Guanaco as belonging to the latter group, because I figurered that its rider
moves woud make it behave like a Rook, while the Llama was obviously short
range and not very valuable, so the center was the obvious place for that.
But perhaps the Guanaco has to be centralized to perform well. This can
matter a lot: for the Lion (FWADN) I initially figured that it was too
valuable to centralize it (it would only be chased), but giving the program
a drive to centralize upped its value by nearly half a Pawn.

So the first thing I will do now (last night's tests were spoiled by
auto-updates...) is to test a symmetric position with Guanaco-for-Knight
substitution, between two differently configured versions of Fairy-Max, one
that centralizes the Guanaco, the other that doesn't. (And perhaps repeat
it with Guanaco-for-Bishop substtutions, or GG for BN.) This should tell me
if a Guanaco gets you more when it is centralized.

After that I will do a a Guanaco vs Llama test, between idenical engines
that handle the Guanaco in the way that proved best. In this test I will
configure the Guanaco only very slightly above the Llama in value. We know
the Guanaco must be better. This should result in an initial empirical
Guanaco-Llama difference. The test should then be repeated with this value
programmed in, for confirmation.

One question: would the end-game you show not be equally won when black
had a Llama rather than a Guanaco? That it is won might be due more to
thefact that black has an extra Pawn, while white has two very awkward
Pawns (one backward, one isolated) than to the Guanaco being so powerful.

Oops! I have made a programming error in the Guanaco. The black Guanaco can only make one two-square jump toward the white position. This makes it much weaker than the white Guanaco, of course. The reason why I missed this is because I only tested the game with the white side, and the white Guanaco moves correctly. (I had simply forgotten to insert the 2-jump directions into the symmetry statement.) The programs which contain the Guanaco are the following, which have been corrected.
http://hem.passagen.se/melki9/guanacochess.htm
http://hem.passagen.se/melki9/pilgrimchess.htm
http://hem.passagen.se/melki9/pioneerchess.htm
/Mats

True, that diagram example was badly chosen. Forget about it. It is very
likely that the Guanaco should be centralized because it has such a short
capture range. Those four capture squares should cover important fields,
which are in the centre. I might, too, need to tweak the Guanaco so that it
maneuvers more in the middlegame
/Mats

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.

Gentlemen, I have a question on shortrange piece placement on larger board
sizes. I wasn't going to pursue it, but after seeing HG's last comment,
I've re-considered and will happily try to muddy the water a bit. 

One of the things I've noticed is that 'centralized' as a concept might
be somewhat misleading, in that on an 8x8, getting the shortrange pieces
into the area of the board where their moves are maximized means getting
them into the central 4x4 square. However, on the Capa board, the area is 6
wide and 4 deep, and on larger boards, it expands more. On a 12x16, the
area of maximum movement for a knight is an 8x12 area that encompasses far
more than the center of the board. Further, since the forces are opposed in
a linear array, anywhere near the midline is actually advantageous, even on
the wings, because the piece is still close enough to the enemy forces to
threaten. And if a bunch of slow-moving pieces clog the center of the
board, that restricts the mobility of the rook and bishop types. A knight
on b5, for example, is not necessarily in a bad position, and it may be a
quite good position. Now, a piece like one in the lama family, being so
shortrange, is about as well positioned anywhere in the area of the
midline, not just the 'center' of the board, isn't it?

[HG, you had a similar problem with a lion, I believe; when you removed 2
backward moves, the less-mobile piece did better than the full piece, if I
remember correctly. Try chopping out the odd potential move or two, and see
what happens, if this makes sense to you.]

The Guanaco is a much faster piece than the Llama. Moreover, the Llama has,
in practice, fewer capture squares than the Guanaco. However, it's easier
for the Llama to create threats, while the capture square is two squares
away. If an enemy piece occupies a central square, the Llama can chase it
away from a distance of two squares. It doesn't matter if there is a
friendly pawn in between. Thus, the Llama is the perfect piece for
controlling the centre, and that's why I am somewhat sceptical about it.
(The Alpaca is different.) Anyway, since computer programs are threat- and
capture-oriented, they can find good use for the Llama. 
The Guanaco, however, is a piece that demands planning. It can easily
penetrate into the enemy position. But one must think like Capablanca and
decide where one should position it. Probably it is better handled by the
human brain. So the solution to the Guanaco mystery would be that the
computer is too stupid to handle it. The Guanaco's capabilities are, in
the middlegame, beyond the horizon. A way of verifying this is to set up
practical endgames, (e.g. remove the queen, a rook, a bishop, a knight and
a few pawns) and set Guanacos against Llamas. The Guanacos would win
against the Llamas in a practical endgame. In this case the Guanaca
positioning would be within the horizon of the computer.
/Mats

First, I have to apologize for creating confusion: I mixed up pieces, and
everywhere in my previous posts where I said Llama, I actually meant Alpaca
(Betza WmD). So I have been comparing Alpaca with Guanaco (Betza WmDD). And
it is obvious the Guanaca must be the stronger of the two, as it is fully
upward compatible with the Alpaca. I did not do any tests wit the Llama
(DmW) yet.

[I wll probbaly edit my previous posts to correct this error, soon.]

About the Lion: you are right, I had nearly forgotten about this peculiar
artifact. Now that you reminded me I remember: The Lion seemed to get
stronger by deleting some of its backward moves! But this was only when I
programmed it as a neutrally moving piece, and could be well understood in
this context: if an assymmetric piece moves randomly over the board (i.e.
each move is chosen with equal probbility) it tends to gravitate in the
direction with the most moves. (Actually towards the center of gravity of
its footprint.) So the Lion with fewer backward moves had a net forward
drive, which turned out much more important for its impact on the game than
a few extra means of retreating. (As it already had so many of those.)

But this effect disappeared entirely when I programmed the Lion as a piece
that should be centralized. Then both the nomal Lion (FWADN) and the one
missing a few backward move got much stronger, but the latter was clearly
weaker than the former, as it should. So articacts like this can only have
an impact if the original strategy fr using the piece s far from optimum,
and not very stable (i.e. easily affected by minor issues.)

About the center: Centralization is something different from mobility.
(Although the two happen to correlate in practice.) The reasons why it is
good to centralize pieces are:
1) Pieces like B can attack the opponent's lines in two places, that he
might not be able to simultaneously defend with non-cenralized pieces.
2) Short-range pieces have their worst-case travel time to any place on
the board minimized.
3) Pieces for which (1) or (2) do not apply can attack and defend squares
in the center to chase away or create safe squares for pieces for which it
does.
None of these is related to the number of moves of the pieces per se: a
Wazir only loses moves (and just a sngle one) at the edge of the board, but
it is far better positioned on e4 than on b2. Pawns in the center are
important because of (3). I guess the Guanaco should be centralized also
because of (3), (with only a single forward direction, (1) does not apply,
and because of its rider moves distance (2) plays no role either), which I
initially overlooked.

Btw, it seems we simply were unlucky: where after 100 games the Guanaco
vs. alpaca was only at 48%, after 250 games it had recovered to a 56% lead
for the Guanacos. I will let it run to 400 games (2% statistical error).
Nevertheless, a 6% advantage for two Guanacos over two Alpacas is not very
impressive; it is not even half a Pawn. (I will repeat the test with the
f-pawn deleted for the Guanacos, to get a precise interpolation.) So it
seems the value difference between a single Guanaco and Alpaca is at most a
quarter Pawn.

Hey, HG, can I ask a favor if/when you edit your earlier remark? I'd
appreciate it if you add the edit either after or before the comment you
are correcting, and leave both. Don't just change the original, so we have
context for the following remarks when people go through this topic in the
future. [I'd like to see that as a general policy.]

Now, as for centralization and mobility, I see what you are saying - there
are differences between the two, although in chess, they correlate
strongly. Are there situations where they don't? Let's build up to this
slowly. I'm interested in establishing some [very] general principles.

First, while a wazir on e4 is much better positioned than one on b2, look
at the wazir on b4. Is it better-positioned than the one on e2? Both
require 2 turns to get to e4, which we will agree, for now, is the 'best'
square. However, the piece on b4 is already 2 steps closer to the enemy,
and thus much better positioned, no? 

What's the mean free path of a bishop? Stipulate that it changes during
the game, starting at zero in the setup and approaching the length of the
side/shortest dimension of the board at game's end. During 'mid-game',
the path length would be roughly 3-4 squares, no? On a small board, ie:
8x8, that still means that the bishop, for greatest effect, must be in the
center. Suppose the board is bigger? On our 12x16, there's a 6x10 area
where, on an empty board, a bishop on any square can travel at least 3
squares in any of its 4 allowed directions. 

Next, suppose no piece in the game moves more than 3 squares. Now what
have you got? A very large game, lots of pieces, all with short range moves
- how does this affect the equation? How about having more than 1 goal
target? Suppose there are 2 kings per side, not all that near each other.
Or more kings. Make it multi-move. 

On an infinite chessboard, there is no center. - Unless you define it by
goal positions, which change as kings move. I'm sure I'm looking at this
far too broadly for most people's taste, but who gets anywhere new by
following the crowd?

I let Zillions play four games with Alpacas against Guanacos and the
Guanacos won by 3½-½. Just as expected if the Guanaco is worth around 3. 
But it's to few games. However, it could be a program problem. Maybe
Zillions is intelligent enough to handle this piece well. In fact, Zillions
overestimates it somewhat. So I had to tweak it to a lower value, to
somewhat lower than a bishop. Zillions overestimates movability without
capture-capability.
/Mats

I agree that it is important where the opponent is. If he would huddle with
all its pieces in one corner, (and there is no need to defend the back rank
in the other corner against promotions), the usefulness of being in the
center deminishes. There are some default assumptions here, and one of
those is that the opponent's pieces will be stretched out along the entire
back rank on his side of the board. In the end-game that assumption might
fail. If both sides only have Pawns on the King side (say on f-, g-, and
h-file) plus Knights and Bishops, there is absolutely no point being in the
center.

I would think g4 indeed a better square for a Wazir than e2, because it is
closer to g7, and not further from b7. This is only from the point of view
of attacking, though; A Wazir can also be useful as a defender, and that
might be a reason to keep it on g2 (with Kg1, and Pawns on f2, g3, h2). But
that pertains to King Safety, which is yet another, independent evaluation
term.

Mats: 'But it's to few games. However, it could be a program problem.
Maybe Zillions is intelligent enough to handle this piece well.'

Well, 3.5-0.5 is indeed far to few games to conclude anything. It could
easily occur in a match between two exactly equal opponents if one of them
is slightly lucky (e.g. because his opponent blunders away a single win).

I am at 366 games now, and the Guanacas lead by 57% over the Alpacas. When
I reach 400 games there, I will delete the f-pawn of the Guanaca side, and
see how they do then. (The Alpacas should win then by a similar amount.)
Two Knights vs two Guanacos is at about 250 games now, and the Knights lead
by 77%.

Intelligence is usually not in the vocabulary of Chess programs. It is all
brute search power, going through millions of positions with a very
simplistic evaluation. (Counting wood or moves.) Especially for a
generalist program like Zillions, that has no specific guidelines
programmed in for the Guanaco for sure.

I don't believe that handling Knights (or Alpacas, for that matter)
requires less intelligence than handling Guanacas. I don't believe that
Zillions would be any better at handling any specific piece than Fairy-Max.
Fairy-Max is only a very simple Chess program, but despite its complete
lack of programmed knowledge (except the piece values) it plays
surprisingly strong in normal Chess, dominating over many engines that are
stuffed with Chess-specific knowledge. Apparently it does not need any
knowledge to handle its pieces intelligently enough to win; plain search is
good enough, even in the end-game.

From what people told me, I don't think Zillions would be a match for
Fairy-Max in Chess variants with FIDE-like Pawns that they can both play. I
have not tested this myself, of course, as Zillions is commercial software
that I don't have. So it is just based on what people that do have
Zillions told me.

I understand that there is an adapter that allows WinBoard engines to play
in the Zillions GUI. Isn't it possible to play Fairy-Max against Zillions
automatically, that way? Then we could handicap one of them by time odds
until they play equally strong in normall Chess, and then add or substitute
Guanacas on both sides, to see if this breaks the equivalence because one
side handles the Guanacas 'more intelligently' than the other.

The 2 Guanacas in the end beat the 2 Alpacas by 56.3% in 404 games. I now
gave them an additional Pawn odds for a new 400-game run. As predicted from
the previous result, the Guanacas seem to be losing that substantially.

I stopped the 2 Knights vs 2 Guanacas match after 300 gmes, with the
Knights leading by 77.8%. Such an extremely unbalanced result cannot really
tell me anything quatitative, So it makes no sense to try to reduce the
error from 2.5% to 2%. In stead I started a match of 2 Knights vs 3
Alpacas, where I put the third Alpaca on d2,e2,d7 or e7, advancing the
corresponding Pawn by 1 square. e.g.

rnbqkbnr/pppppppp/8/8/8/4P3/PPPPAPPP/RABQKBAR w KQkq - 0 1

where A = Alpaca. This seems to go pretty even (but only 20 games so far,
so can still easily be anything between 35% and 65%).

I remember I did some divergent pieces sometime ago (in particular mNcK,
mKcN, mQcN and mNcQ), and I thought I posted the rsults at CVP. But I
don't remember where, and I cannot find it. What I recall was that mQcN
was 450 and mNcQ 750, when N=325 and Q=950. I never did mQcK and mKcQ, but
snce K and N are pretty close in value, one might expect nearly the same.
Now R is about half a Q, and W is about half a K, so it does not seem that
silly if mRcW would be about half a mQcK, which would be about equal to
mQcN = 450. That would put mRcW around 225. Now the Guanaca is very similar
to mRcW: its lacks the odd-stride distant non-captures, but as a
compensation it jumps, and so it cannot be blocked on these odd-distance
squares. So actually something around 225 would seem a quite reasonable
value for something like the Guanaco.

It is truly remarkable that the Guanaco is only 0.25 stronger than the
Alpaca. But I believe now that it could be correct. It is easy to overestimate movability which lacks capture-capability.  Thanks for the investigating job! So I have some tweaking job to do in my Zillions programs, to lower its value. Tip: In Zillions the pawn's value can be increased somewhat by adding yet another rook or queen, etc. to the list of promotion pieces. It doesn't change the rule if it is allowed to promote to 'queen or queen'. I don't understand why you haven't purchased this wonderful software. It's only $20.
/Mats

I have now lowered the value of of the Guanaco in my Guanaco Chess, etc.
The immediate consequence is that Zillions starts to use it much more
actively. I wonder what those scientific piece value algorithms say about
its value? However, its value is such that it is really a misfit in the
Western piece array, so I am somewhat sceptical about it. The Alpaca is
different while Alpaca + pawn = light piece. 
/Mats

Mats: 'I don't understand why you haven't purchased this wonderful software. It's only $20.'

I guess the main point is that I aready have my own programs, which I
consider better, and for which I don't have to pay anything. That does not
only pertain to playing strength, but also to the things it can do. E.g.
automated self play. I suppose there is a reason why you only played 4
games with Zillions to play-test the Guanaca, and turn to me to play a few
thousand, in stead of having Zillions do that?

Another issue with Zillions is that it does not support a standard
interface for playing external opponents. I know there is an adapter that
allows WinBoard Chess engines to play in the Zillions GUI. But can the WB
engine play againt the native Zillions AI, in that case? Can two native
Zillions AIs (such as one of your hand-tuned versions against the standard
one) even be played against each other, to optimize the tuning? Does the
WinBoard-to-Zillions adapter work for any variant besides normal Chess?

In short, based on the information I have, WinBoard + Fairy-Max are far
superior to Zillions, and they are free. That Fairy-Max has limitations in
the piece types that can be programmed into it through its .ini file could
be a problem for non-programmer users that are limited to altering the
piece definitions in that .ini file. But it isn't to me: if I would be
really interested in a particular piece falling outside the usual parameter
range, I would simply add a few lines in the C code of Fairy-Max to make it
support such pieces (or special winning conditions). This is what I did to
make a Xiangqi version of Fairy-Max (MaxQi), or for implementing the
Shatranj baring rule (Shamax).

Did you ever download Fairy-Max? That is only $0...

On the main topic: Some 'half-time standings': 2 Alpaca + Pawn are
leading 65% over 2 Guanaco. (Which is more than I would have expected;
usually deleting a Pawn on 8x8 only swings the result by 15-18%). 2 Knights
are leading 55% over 3 Alpacas.

Zillions cannot be compared to anything else, because of its amazing
geneality. But that does not mean that any particular thing Zillions does
cannot be done better by other means.

Zillions is basically a programming language, and so it can do anything.
But C is also a programming language, and it can also do anything. Although
some pieces are too complicated to handle for Fairy-Max at the level of the
configuration (.ini) file, because I only provided for true Chess pieces
(i.e. moving from one square to another in a translation-invariant pattern,
capturing by replacement). Side effects are rare and too specific to
generalize them efficiently. So they cannot be implemented by enabling pre-cooked 
options from the .ini file.

But that does not mean I could not handle such pieces if I wanted to. It
only means that in such a case I would have to handle them at the C level.
Being the author of Fairy-Max, it is very easy for me to add the necessary
code to implement peculiar properties at the C level. I would have little
difficulty to use pieces like BodyGuard or Coordinator in a Fairy-Max
derivative, like I have no problem limiting Kings and Guards to the palace
in a Fairy-Max derivative (MaxQi) that plays Xiangqi. It is very
questionable if this would take more effort than describing the piece in
ZRF. 

How I would do that, would depend on the case at hand. having to think
about it would allow me to choose an efficient implementation. For the
BodyGuard I could for instance test all surrounding squares on the presence
of an enemy BodyGuard, and abort the ray scan if there is one. that would
probably still leave Fairy-Max faster (in searched positions per second)
than Zillions, which always has to test for who knows what strange
propertis of pieces. But really efficient implementation would add the
lines

position[piece] = toSquare;
position[piece] = fromSquare;
if(neighbor[toSquare - position[32 - color + BODYGUARD] + 140]) break;

in the code for MakeMove, UnMakeMove and the ray scan of the move
generator, respectively, plus an initialized table

char neighbor[280] = {
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,
0,0,0,0,0,0,0,0,0,0,0,1,0,1,0,0,
0,0,0,0,0,0,0,0,0,0,0,1,1,1
};

Well, big deal, 3 lines of code and a trivial table. I doubt if you have
to tell Zillions less than that to implement the stimying effect of a
BodyGuard.

I don't understand your negativism towards brute-force tree searching; I
am sure that this also is exacly how Zillions comes up with its moves.
Except that the need to be general slows it down, so that it searches less
deep. And that means it plays on the average weaker moves. Deep searching
has proven the way to winning Chess. Why you would prefer to have an entity
that plays like crap just to be able to say that it was not a 'bean
counter' escapes me.

You would have to use C-language, yes. Anyway, Zillions is freeware, too,
to a great extent. There are many games for free, and you can construct
chess variants, too, and store them as Zillions games. There are many
pieces implemented which one can add by right-clicking.

The brute tree-search is deadening. Chess is much more fun when plans are
used. Zillions sometimes comes up with the plan to move the rook to f6 and
then out on the kingside (before his own pawns) in order to make the enemy
king nervous. Sometimes it's not good, but it's a plan. ChessBase Fritz
has always been a very bad program in this respect. It has only made
calculations and is lacki ng in ideas. Zillions suddenly starts a mad pawn
storm on the king's side. At least it's a plan, although, by calculation,
it is often bad. But it's much more interesting because it's the human
style of play.  Zillions, of course, has too little of this. If a planning,
thinking, chess program would be created, it would revolutionize computer
chess. Chessplayers are bored to death about sterile chess programs.
/Mats