We were field testing a new Unisys mainframe (V530) at a
banking customer in Brussels in 1987. The customer had a number
of older systems in their datacenter (which was on the top
floor of the bank in downtown Brussels with an excellent
view) including some V380 systems, upon which were were
to install the latest version of the operating system
and processor firmware. I started at 2300 hours by
updating the processor firmware and then coldstarting
(loading from tape) the new MCP. It failed during
boot, which after an hour or so, we tracked down to
an instruction fetch issue in the hardware.

The field engineers had an on-site spares kit, so I
started swapping out processor boards (not something
that an OS engineer does regularly) to attempt to
get the system working before the morning shift
arrived and the system was needed for production.

Being a fetch error, I started with the two fetch
boards, one at a time, testing inbetween. Didn't
help. Then tried the two XM (Execute) module cards.
No change. Tried two other processor related
boards. No different. Finally changed out
the memory read controller board and voila, the
MCP booted and the test apps worked. It was
now 0750, and the production crew started at 0800.

I also heard stories about a system installed in
a San Diego county building, next to the indoor
shooting range, where a deputy put a few holes
in one of the mainframes (intentionally was the
rumour).