I find it most productive when the assumption is asked and granted for something uncertain, interpretable, subjective, etc. Or when we are just having fun :-)

If a) you are building a serious proposedmodel of reality and b) start with something demonstrably false, I'd have to be in a pretty specific mood to go with it :-D