...unlike the usual coding mistake which produces an easily noticeable discrepancy in nearly all of the results, an undetected overflow condition may only occur sporadically and perhaps not at all for the chosen test cases. Thus if the coder neglects to query the computer about alarm conditions his later results can be completely wrong. In fact one could wonder why he went to so much trouble, took so much time, and used such an expensive machine only to get some meaningless numbers. Needless to say the situation is aggravated if the numerical results are used as the basis of an article in a research journal.
— Coding for the MIT-IBM 704 Computer (1957), "Overflow, Underflow and You"