I take a different tack to JimBob. It's not an issue of reactive vs. proactive or whatever. It's that "say yes or roll dice" doesn't work when players desperately want to be told "no". Even with the best of intentions on all parts, "no" is what gives a game a sense of verisimilitude.
That said, I'm currently inclining toward a default philosophy of not saying "no" unless I have a good reason in terms of a strong feel for the setting/situation, established precedent, or a specific detail that I've prepared. E.g., if there's a moat filled with crocodiles, it's there, period, and no the crocs aren't baby crocs or tame crocs or narcoleptic crocs who faint if you clap your hands. But if you want to try running across the backs of the crocs and that's roughly in the tone of the game, I'm not going to expend a lot of effort on whether such a thing is plausible--to the dice, I say!
What I think is perhaps overlooked in this whole thing is the way that dice mechanics can take the pressure off the GM's having to prepare or judge every detail of the game. A high variance in random outcomes, combined with a little flexibility in interpolating the cause behind what might seem to be an unlikely result, seems like it could be a good strategy, and I have to give Forge writers credit for leading me in what looks like a promising direction. On the other hand, remove too much influence from judgment and you have a game where how you do something doesn't really matter...and in that case I'd suggest that neither failure nor success should be allowed to produce extreme results very easily. (I think this may be the philosophy behind a few indie games and/or games that have influenced indie design such as Heroquest.)