We had a customer whose Unix environment didn't pass a stderr file handle to running programs. So if you tried to print something to stderr, it scribbled on whatever file happened to be open on handle number 2. This got reported to us as a bug. After looking at it, we told them to fix their insane runtime environment.
A sysadmin had (I can only assume accidentally) removed the SUID bit from the crontab binary causing an application we used that invoked the binary to not work. To even allow this to work we had a custom SELinux module that we suspected had the issue.
I ended up comparing systrace output between a working instance and the broken instance (in prod) to track it down.
It happened because the mvn on the build server ran out of memory exactly because of that 3 character addition.