Awesome! I tested it locally, found no errors and merged the branch.

Looks ready to merge to me. Have you tested it (locally), just in case, e.g. with the CEGAR heuristic that now uses initially stopped timers?

I agree, we don't necessarily need the assertion and I removed it. Can I merge this?

I had a look. I think the case distinction in the constructor/use of NaN is more complicated than necessary (though I realize you need it if you want the assertion), but in general this looks fine.

I prepared a patch at https://bitbucket.org/jendrikseipp/downward/pull-requests/153 . Could you have a look, please?

This came up in issue781.