Graph History Interaction Problem

[kevinfat]

I am trying to understand the graph history interaction problem described in this paper by Kishimoto and Mueller.
http://webdocs.cs.ualberta.ca/~mmueller ... ci-ghi.pdf

I see that the problem has already been discussed but I couldn't follow their discussion.
viewtopic.php?f=5&t=1015
http://talkchess.com/forum/viewtopic.php?t=21343

Presumably the high level pseudo code below has problems. Can someone describe concrete examples that cause problems?

Code: Select all

Input Graph g, Position p in g, Move m, Evaluation e:
let p' be position after making move m in position p or create p' if it doesn't exist in g
add edge (p, p') with label m
set evaluation e for p'
call minimax on p'

[kevinfat]

The previous pseudocode is for implementing extending the graph by a move. And on a side note on further reflection I realize it doesn't handle for example an drawn endgame consisting of rook king pawn vs rook king. It would exhaust all the positions but if the approximate evaluation passed in was not 0 it would have no way to detect that all the positions should be set to 0.