In testing, I found that any keyboard event results in an atime update,
regardless of whether the event was captured by an xterm, a Thunderbird message
window, a Firefox URL bar, or simply a modifier-key captured by the window manager.
It's simple to test to see whether it works for yourself:
* In one terminal, run `watch -n0 stat /dev/ttyN` (where N is the the TTY
managed by your session, probably '7' -- check with `w` output.)
* Whilst watching the atime field in the terminal, experiment with different
classes of keyboard input..
With non-USB keyboard, Condor would catch that activity by atime changes
on /dev/keyboard and /dev/mouse, but atimes do not change with USB
devices on an unpatched kernel....