On 3 May, 20:25, help bot wrote:
On May 2, 5:09
wrote:

Your choices are to add a few choice heuristics to the endgame
evaluation function when the number of pieces is 4 or less and/or to
implement basic tablebases for the classic 3 & 4 piece endgames.

These would be roughly (in bothKBBkandKBNk) to:

Make the winning sides king bolder and take the opposition with the
enemy king.
Restrict the mobility of the enemy king to 2 or 3 moves where
possible.
Restrict the region of the board available to the enemy king.
Winning side scores for forcing enemy king to edge of board, bonus for
corners.
Avoid stalemates (stating the obvious)

* For human players, the last heuristic -- that of
studiously avoiding stalemate -- might be moved
to the head of the pack, once the enemy King
is driven against the side of the board. *:D

Indeed. Although I would hope no computer would ever need it.

I tweaked my toy eval to test this idea for KBBk and a suitable weight
for king distance and piece centre weighting together with an 8 ply
search is just about adequate. That is these simple heuristics to
restrict mobility to 3 then 2 squares in a corner will get you to
within a mate in 4. It worked OK with a 5 ply search for the test
cases I have to hand. I haven't tried it for KBNk.

Regards,
Martin Brown