Never Send A Salesman To Do An Engineer’s Job

Working for IBM on Unit Record systems (Punched Card equipment) was my introduction to Data Processing, and to this day I think it was one of the most fun jobs I’ve ever done. Here’s one little tale of those days:

One customer had a Series 50 Installation, full line of punch-card equipment: 026 Keypunches, 056 Verifiers, 604 Calculator, 086 Sorter, 77 Collator & 403 Accounting Machine. The user’s data files naturally consisted of trays of punched cards.

The customer had run one of his card files through the sorter to extract the records needed for a particular report. After feeding the extracted cards through the 403 and creating the report, he tried to collate the cards back into the full card file. At a certain point, one of the cards ripped in half. It was removed and a replacement card keyed up, only to have the same problem occur, repeatedly. IBM Customer Engineer went over the hardware with a fine-toothed comb but found nothing amiss. The salesman on the account asked me to investigate.

After ripping a couple of cards myself, I looked at all the cards which were being collated back into the card file. When collating two files, each must be in order. The original file was in order but evidently after the extracted cards had generated the reports, some of the cards got mixed up and put back out-of-sequence. I could see why the collate had failed and how to fix the problem but could not figure out why the card was cut in half.

Looking at the logic of control panel wiring, I found the bug. The panel had been wired by the salesman rather than by a Systems Engineer (me). He correctly tested for an out-of-sequence condition in either of the two files being merged. His logic was such that correct output went into the standard output pocket, an out-of-order card in one file was supposed to go to a 2nd output pocket and and out-of-order card in the other file was supposed to go to a 3rd output pocket. That logic was correct, but his implementation was wrong. The equipment has something called Selectors, basically a switch that could shunt a signal between one of two circuits. The position of a Selector is controlled by a ‘pick’ signal. The machine had two electrical ‘pulses’, one long and one short. The correct wiring would be for the ‘pick’ signal to be a long pulse, to open the Selector early in the machine cycle until late in the cycle and the signal going through the Selector to be a short pulse, beginning and ending within the long pulse time. The salesman’s wiring, however, had both the ‘pick’ signal and the ‘through signal as long pulses. The end result was that the ‘through signal’ went to one circuit as the Selector started to flip, then to the other circuit once the Selector had flipped completely. One of the activated circuits deflected a thin ‘blade’ to deflect the card to pocket 2, the other circuit to deflect the card to pocket 1. With both circuits live, one ‘blade’ deflected with the card half way past pocket 2 because pocket 1 was selected by the other circuit, only to trigger because circuit 2 was also live. It cut the card in two, half falling into pocket 2, half into pocket 1.

The remedy had three parts:
Rewire the control panel to pick the Selectors with a long pulse and take a short pulse through them.
Sort both card files before collating them.
Never let a salesman do a Systems Engineer’s job.

Leave a Reply