Whenever I've "started over", I've sometimes longed for going back to pre-existing stuff and evolving it. Sometimes, but not always. Whenever I've 'evolved' existing systems, I've almost always wish we'd started over (or some variation thereof).
Recent-ish project: holding on to multi-year old legacy data that was simply bad/broken (but was assumed to be good because it was barely used/examined), then requiring every single new feature to work flawlessly with broken data, ensuring that legacy customers 'never' hit an error (when 95% of them have not logged in in years). That's a candidate for 'rewrite from ground up' - you can keep an old system going in maintenance mode, build the archetypal "version 2", then decide if/when/how to migrate old data over...
Knowing that you can 'evolve' a system towards improvements and knowing when you should do that is another sign. Something may be technically possible, but the time/effort/cost can outweigh whatever benefits there may be.
