|Photo: US NARA via Wikipedia.|
|1920 census form - Click to enlarge"|
By 1910, some error checking was done by the tabulators themselves... "If the entire color, for example, was left unpunched the card would fail to register and therefore fail to ring the bell for which the operator was listening. Cards were also rejected if they were 'off gauge' or otherwide mechanically defective. These rejected cards were examined and the omitted items supplied — or new cards, conforming to the mechanical requirements, were punched. In addition to these defective cards, there were many cards containing items that were so far inconsistent as to raise questions with respect to the accuracy either of the punching or of the original returns from which they were punched. The verification run of 1900 was designed to take care of both types of defective cards, in advance of the use of the cards in final tabulations, through a special machine wired to reject not only incomplete and defective cards, but also those with inconsistent items, items otherwise frequently subject to question of accuracy, and a considerable number of items of infrequent occurrence, of which it was desired to be doubly certain." [Truesdell, p.94-95]
These verification runs were perhaps the first example of "programming by wire", made possible by Hollerith's invention of the wiring panel for his Type I Tabulator in 1906.