And this is my point -- authoring Bob on the fly is not understanding the character any more than writing down a few BIFTs before play is understanding the character. These are the exact things that get pointed to as weak characterizations and cardboard play. I mean, I can say that Bob says "mmhmm" when a character asks him something because he's distracted thinking about the body in the cellar than he has to move, and that's fun, but I don't understand this character better -- why is there a body, did Bob do something, is Bob under pressure from some other person, does Bob hate or love the dead person or are they a stranger, is Bob a simpleton who can't really think of more than one thing at a time, is Bob.... and on and on. The point is, at the moment I'm making the choice for Bob to respond to a character's action, I do not understand any of this. I'm making a blind choice and then, probably, inventing things to justify it later.
I don't need to, though, I can do whatever I'd like and then justify it. This is the opposite of understanding the character.
I don't. Let's take Bob, who is undetailed. The PCs approach and try to convince them to aid him. How do we decide this? We have nothing, so really we're only making this decision based on how we, the GM, feel this scene needs to play -- is Bob going to be difficult to extend the scene, is he just going to say yes because we think this is unimportant, or is he just going to say no because we think this is too easy and have a plan for the next bartender to be the helpful one? Whatever we decide, it will not, in any way, be about what Bob thinks, or understanding him as a character. So, we're actually making a choice that has to be biased by our feelings on the play and not about Bob. A mechanic here could remove that bias, and will not increase the arbitrariness of the result. The mechanic says, and then we're in the exact same spot -- justifying the choice by authoring more about Bob.