Post History

What came to mind immediately on reading the question, limitation-wise, was that you actually need to understand the current probabilities and what contributing factors you want (notionally) changed.  Changing the fairness of a die, then, is (maybe) pretty easy, since you could nudge the bouncing cube away from seven minus the target number more often than an even chance.  You only have one piece and a lot of opportunities to shift things.  But (ignoring setting ideas) encouraging a car to stall, you need to know the relevant mechanisms, their failure modes, and the existing odds that they'll happen.  That could take *days* to work out, when you only have a few minutes.

Not everybody does it, because not everybody has the patience to prepare that kind of work.

And the cost also seems potentially straightforward, despite the question trying to avoid it at the end:  Arbitrariness.

That order imposed on events needs to "come from" somewhere, making other events more difficult to predict as their probabilities skew.  Sure, your pursuer got a stone in his shoe, but do you really want to risk climbing that fence with your heart condition after lucking out like that...?  Those upcoming actions used to be random, but now they very much *are* arbitrary, and nobody can guess how, because nobody knows the state of every system just prior to borrowing that randomness.