On Jan 29, 1:36*am, help bot wrote:
On Jan 28, 5:16 am, Martin Brown
wrote:

20. Nc3-d5{88} e6-d5{62}

Why in heavens name is it swapping a N for a P ?

* *Two* pawns; White gets two pawns for
his Knight (which, as I keep telling Sanny,
is not nearly enough). *As everyone knows,
a Knight is worth around 3.1415927 pawns
-- about the same as a piece of apple pi... .

2 pawns and an attack or some structural damage to the kings defences
might have been OK, but otherwise 3 pawns for pure material. Sannys
program seems to like 2P for 1N swaps far too much....

A lot of real games tend to hinge on a swap that runs N,B for N,B,P
when the exchange is done in the right order.

The attack here is at
best illusory, at worst non-existent.
Kg2 is probably needed at this stage.

50. Kf1-g2{138} Rf7-f4{22}
51. Rd4-f4{126} g5-f4{6}

It has definitely got a death wish!

Keeping the rooks on the board and hoping that your opponent makes a
mistake out of boredom letting you snatch a pawn is the only chance
remaining here. Once the rooks are off the two passed pawns protected
by their king are invincible.

* No computer "thinks" that way; even strong
programs will object to hanging a pawn for
nothing here, and unless its algorithm allows
for full-pawn sacrifices just to keep a pair of
Rooks on the board, would do this sort of
thing. * GetClub does not see far enough

You can tweak the evaluation to favour equal swaps when ahead and try
to avoid them when behind. I agree that giving up a pawn to keep rooks
on would tax most chess engines (though not the current crop).

* Note also that near the end, the program
moved its King into the corner; it seems to be
programmed to believe this is the ideal spot
for King safety (even where the opponent has
no attackers left). *In simple endgames, the
program is severely handicapped when
compared to a typical human player.

The lack of basic endgame heuristics would make it easliy vulnerable
in the endgame.

Regards,
Martin Brown