You have your Olympic-class marksman, a master sniper, about to take the shot from a rooftop at a target. There's absolutely no way he can miss. Or is there?
We're not talking about a stressful situation, which is commonly offered up as the criterion for whether or not to call for a dice roll, or for why you might have a "botch" rule (e.g., 00 on %ile dice is always a failure).
No: fundamentally, nothing in life is ever certain. (The sophist in me wants to say, "It's certain that I won't grow wings and fly into the sky." Maybe nothing positive is ever certain.) Back to the sniper. Could a pigeon crap on him just as he's pulling the trigger? Could the target move unexpectedly, say by bending down to pick up something dropped? Could he be spotted? Could someone know about his plans in advance and catch him before he pulls the trigger? Could the target be a decoy?
Some of these explanations may fall outside the range of possibilities that a game would want to consider, but others should be considered, I think. Particularly when you have characters who are defined as having a certain level of mastery.
Conceptually, when I ask whether someone accomplishes X, there are two sets of factors to consider: internal causes, and external causes. At some point in the continuum of mechanical ratings and resolutions, as a character goes from rank incompetence to absolute mastery, the relative importance of those two sets of causes evens out and then becomes reversed, even as the likelihood of success approaches the maximum. (There's always a maximum below 100%, even if it's so close to 100% as not to be worth modeling. In practice, RPGs may benefit from exaggerating the effect of external events and capping success chances at 99%, 215 in 216, whatever.) A beginner using a smoothbore musket, aiming at a target 150 m away, may have a 1% chance of hitting--mostly due to lack of skill. But an expert might still miss 10% of the time because, even in a pigeon-free environment, non-rifled muskets have a fairly high dispersion--shoot twice with precisely the same aim, and you'll rarely hit the same spot.
The problem is to determine when a failure happens because the character wasn't up to it, and when it happens because of something extrinsic. This can be important because the extrinsic event may have interesting consequences, or at least the image of the character and the integrity of the fiction will be maintained instead of looking ridiculous. Bond doesn't fail to catch the hit-man because he trips over his own feet--it's because a truck suddenly enters the intersection, blocking him.
Some games aim in this direction. West End Star Wars 2e had something called the Wild Die that seems to have been along these lines. Mythic RPG has "interrupts"--basically any die roll can trigger a random event. In either of these games I suppose you could interpret a regular failure as being the character's fault, but special failures are extrinsic. But I don't think this idea has ever been made fully explicit.
(BTW, what would be a good tag for this sort of discussion? Mechanical interpretation, mechanical semantics, mechanical hermeneutics, mechanical semiotics?)
Then there's the question of interpreting success, too. And for both successes and failures, certain mechanics could be limited to extrinsic or intrinsic causes--e.g., spending a "luck point" might let you turn a failure into a success explained by "extra effort", but spending a point to turn a success into a failure would have to be explained by an extrinsic cause.