Ah, right, I see it now. I guess you can check the current-milliseconds after the fact, and force the default value in that case. But looks like this is going to be a problem if I try to safely simulate other agents... Actually, I suppose it's possible for the target thread to similarly not get returned to for a long time, causing any watchdog to overestimate the time it used. current-process-milliseconds might help with that, but I'm not sure if it can deal with nested threads usefully.