In February 1965, IBM’s Product Publications department in Poughkeepsie issued publication A24-3332, a 46-page reference manual for the machine most people came to call the 029 keypunch. The manual never uses the name. Its subject is the IBM 29 Card Punch, introduced in 1964, the year of System/360, and its declared reader is a card punch operator or trainee assumed to have an introductory knowledge of data processing. A comment form was bound into the back, with a mailing address supplied in case an earlier reader had already torn the form out. IBM kept revising the document into November 1971. Read now, it is a catalog of work that had not yet moved into the machine.

The description opens with a claim no data sheet would carry today: the machine is “easy to operate, quiet, and attractive.” After that the prose settles into numbers. Duplicating under program control runs at 20 columns per second, or 18 on models that print as they punch. Skipping and card release run at 80 columns per second. When one card passes column 80, the next is registered for punching a quarter of a second later. The hopper holds about 500 cards and the stacker about 500 more, and an automatically zero-filled field of eight digits takes about half a second to punch, during which the keyboard locks and an operator who starts keying too early can feel the keys refuse. The manual specifies that pause because the pause is part of the machine.

The most instructive sentences are the warnings, because each one describes a part that does not exist. The Model B can insert zeros to the left of a number automatically, holding keyed digits in storage until the operator presses a key to punch them out. Nothing in the machine prevents keying more digits than the field can hold; the manual states there is no interlock and assigns the operator to guard against it. Backspacing more than 20 columns disturbs the card’s registration, so a sentence prohibits it. Backspacing past column 78 will foul the following card unless the operator removes that card first, so a sentence prohibits that too. The sensing mechanism must never be lowered onto an empty program unit. Torn card fragments are to be pushed out of the stations with another card or a smooth-edged metal blade, and the manual takes the trouble to ban saw-edged blades, which suggests someone had used one. Every Caution names a failure the hardware permits and installs a person where an interlock would later go.

The longest chapter is titled Programming the IBM 29 Card Punch, and what it teaches is handwork. A program here is a punched card: a 12-punch in a column defines a field, an 11-punch starts an automatic skip, a zero starts automatic duplication, and a 1-punch shifts the keyboard from numbers to letters. Because those codes occupy only the upper rows, a second complete program can be punched into the lower rows, so one card carries two independent 80-column programs, selected by a switch or, in the middle of a card, by keys. Installing the program means wrapping the card around a metal drum, clamping one edge under a toothed strip and the other under a smooth one, seating the drum on its spindle, and lowering a lever that rests up to twelve spring-loaded starwheels against the card’s surface. The drum turns in step with the card being punched, and wherever a starwheel drops through a hole, the machine skips, duplicates, or shifts. The program is an object. It takes both hands to load, and heavily used machines were known for missing starwheels and dry ribbons.

Correcting a mistake is likewise a procedure rather than a feature. An operator who notices an error presses the release key, duplicates the correctly punched fields into the next card, and rekeys only the field that was wrong, with no need to know which column the error occupied. For cards that arrive partially prepunched, the manual gives a numbered procedure fifteen steps long, involving the stacker, two manual insertions, and the settings of three switches. The document also records its own revision history in passing. A note explains that the backspace key no longer resets a locked keyboard, because the error reset key now serves that function, a sentence addressed to operators whose hands still remembered the earlier machine. Throughout, the worked examples belong to an invented General Manufacturing Company of Endicott, New York, the town where IBM manufactured its punched card equipment.

The options list shows the checking already starting to move inside. A self-checking feature could verify code numbers as they were keyed, using a check digit computed by modulus 10 or modulus 11 arithmetic. When a keyed number failed the check, the machine punched a 12-hole over the check digit, lit a red light on the keyboard, and locked the keys. Verification that the base manual assigns to the operator’s attention is here performed by circuits, and the manual’s remaining job is to explain the light.

In 1971 IBM announced the 129, its last keypunch, a 29 with buffer memory added. An operator could key a full card, inspect it, and edit it before any hole was committed, and the machine stored six programs internally, with no card to punch and no drum to wrap. Most of the Cautions in A24-3332 stop being necessary the moment the data can be examined before it becomes holes. The 29 itself carried on regardless; Columbia University photographed its last public keypunch, an 029, still in service in 1986 beside a pair of video terminals.

Manuals have been thinning ever since, and the reason is visible in this one. The sentences moved into the machines, where they run as interlocks and validation and need no reader. What IBM printed in 1965 is the full text of an interface, from the last period when the only place it could run was a person.