later the problem came up again. I asked a library assistant to show
　　　　  me another example. "But you said it was impossible to fix," she
　　　　  reminded me.
　　　　  In half an hour we had isolated the problem and fixed it. The system
　　　　  had received the information but did not know it. The fix was just a
　　　　  matter of sending the interface unit something that would satisfy
　　　　  its search for a second screen without adding anything to the
　　　　  record. With intermittent problems, the main challenge is finding
　　　　  out what the problem really is. Once you find out what is different
　　　　  about the thing that malfunctions, it is usually easy to fix.
　　　　  Accepting uncertainty
　　　　  Sometimes things will return to normal before you figure out what
　　　　  went wrong. I call this the "uncertainty factor" and have learned to
　　　　  just accept it. A problem is quickly forgotten unless the machine
　　　　  does the same again within a few days. I don't feel that I have to
　　　　  know every aspect of a problem--especially if the problem
　　　　  disappears.
　　　　  Often after quick fixes, people ask how I did it. I mutter about
　　　　  fooling with the connections and prodding the thing. "But I did
　　　　  that!" they exclaim, and then they wonder if my job has an element
　　　　  of magic. It is useful for people to believe this. There seems to be
　　　　  a greater chance of success if the person believes that the machine
　　　　  will work as soon as I look at it. However, I must guard against
　　　　  believing my own clippings.
　　　　  The solutions for long-term problems often prove to be simple--so
　　　　  simple that I could kick myself later. One of my favorites was a
　　　　  machine in the acquisitions department that could not be used for
　　　　  check-ins. When the check-in box appeared, the data would start to
　　　　  fall apart. We checked the settings repeatedly, and they were
　　　　  identical to the other machines in acquisitions, even though the
　　　　  balky machine was an earlier model.
　　　　  Swapping cables with the machine on the next desk didn't help,
　　　　  proving that the problem was with the machine itself. In
　　　　  desperation, I tried changing the terminal emulation. This seemed to
　　　　  affect the check-in boxes. The machine could not handle the terminal
　　　　  emulation it was running, but our acquisitions system couldn't run
　　　　  anything else. Finally, the terminal was swapped with one linked to
　　　　  the cataloging and circulation system, which can be set at VT100
　　　　  terminal emulation. It has worked perfectly ever since.
　　　　  When a piece of equipment is broken, the only procedure is to send