This paper presents the TREAT match algorithm for
AI production systems. The TREAT algorithm introduces a new method of state saving in production system interpreters called conflict-set support. Also presented are the results of an empirical study comparing
the performance of the TREAT match with the commonly assumed best algorithm for this problem, the
RETE match. On five different OPS5 production system programs TREAT outperformed RETE, often by
more than fifty percent. This supports an unsubstantiated conjecture made by McDermott, Newell and
Moore, that the state saving mechanism employed
in the RETE match, condition-element support, may
not be worthwhile.
