Performance of the census UNIVAC system [includes discussion]

In June of 1948 the US National Bureau of Standards acting on behalf of the Bureau of the Census contracted with the Eckert-Mauch1y Computer Corporation for a UNIVAC System. This UNIVAC System, now generally known as the Census UNIVAC System, was accepted on March 30, 1951, and since that time has been devoted almost exclusively to tabulating results of the 1950 Census of Population and Housing. We will try to present here certain facts about the acceptance testing and about operating experience. We also will indicate the inferences we have drawn from these facts. We are aware that any given body of facts may be, and often is, interpreted in a variety of ways depending upon just what it is that the interpreter is trying to prove. We will try carefully, therefore, to distinguish between our facts and our inferences. Furthermore, we will try to present the extremes of the conclusions that might be drawn from the facts. One of the authors is an engineer with some familiarity with the difficulties of physically realizing the grand promises frequently made for, and not always by, engineers. The other author is in the business of producing statistics. He is interested in any tool that will increase the efficiency with which he conducts this business. For purposes of this paper, at least, each of them does his best to be a good advocate for his ´devil.´