The focus of this very well-written paper is the determination of a system whose expected performance measures are larger or smaller than a single system, with the goal of subsequently presenting the smallest and largest system.

After a brief, but interesting introduction, consisting of the paper’s organization and a history of such experimentation, Kim defines the problem and lists basic assumptions.

In the next section, the author lays out a generic fully sequential procedure, followed by a customized procedure. Section 3 summarizes the results of an experiment to compare the two-stage procedure with a fully sequential procedure. Section 4 details the statistical and mathematical theorems used in the study.

The conclusion briefly summarizes the findings. The author states that, with or without the use of the common random number (CRN) scheme, the estimated nominal probability of correct selection (PCS) of fully sequential procedures (FSP) was over .95. He indicates that overall experiments showed that that FSP is uniformly superior to NG (a generic procedure) under any configuration, in terms of the total number of observations. There is a short reference section.