We present a new algorithm for selective search by iterative expansion of leaf nodes. The algorithm reasons with leaf evaluations
in a way that leads to high confidence in the choice of move at the root. Performance of the algorithm is measured under a
variety of conditions, as compared to minimax with α/ß pruning, and to best-first minimax.
Keywords Selective search - evaluation function error - misordering assumption - certainty - confidence - swing - evaluation goal - swing threshold - LCF - random trees - artificial time - Amazons - Othello
Acknowledgments This material is based on work supported by the National Science Foundation under Grant IRI-9711239. Rich Korf shared his
code for efficient indexing of a random number sequence. Gang Ding, David Stracuzzi, and Margaret Connell provided helpful
comments.