I find that a training run where I start the sync in doors then start as soon as hooked can have terrible data at the start and when I look at the start point it is way off. However when I sync at a parkrun or race and have been standing around for a few mins at the start with the satelites hooked on then the accruacy is pretty good,
For me stopping the watch is almost cheating. If I stop in a race then the chip time is still time elapsed so do not see why I should stop the watch on a training run. If I am not running then I am having a benefit from resting so to me the splits without rests would be false anyway.
Trying not to sound to stupid on this as it has taken a lot of debate with myself to come to this rule. The only two exceptions I make are laces (which are rare and drive me nuts having to stop). and unexpected problems like floods, water logging etc. For everything else then the watch keeps ticking.
Sometimes if I am slowed, I will press the lap button and then press again after so that whatever problem I have had shows as a short / slow lap.
I should clarify that I do live in the country so hardly ever have to stop for traffic. Living somewhere like London would drive me crazy with crossings and such.