There are two ways you can view causality flowing in terms of RPG rules: from the game-world to the rules, or from the rules to the game world. Either game rules attempt to describe a game-world, or they define the game-world.
In the first view, game-world effects have game-world causes, and the rules are just a model or approximation of the factors and chain of events from cause to effect within the game-world. They’re there to make adjudication more consistent, predictable, or speedier, but they’re intended to be sacrificed whenever they don’t accomplish those goals. It’s taken for granted that the rules are only approximations, and they need not be consulted if the results are obvious to all the players (or perhaps just obvious to the GM), and they need to be overruled whenever they yield a result that doesn’t make sense in terms of the game world.
In the second view, the rules are in effect the physics of the game-world, and it’s impossible for them to yield a result that doesn’t “make sense” in the game-world. If there’s any flaw, it’s in the players’ improper grasp of the way the game-world operates and their invalid attempt to apply ordinary ideas of cause and effect or probability imported from our world into the game. The rules are there to tell the players what is and is not possible in the game-world.
The choice is a matter of taste, but the two views are mutually exclusive. Even if you switched back and forth from one view to the other, or used one view for certain rules and the other view for different rules, you can’t simultaneously hold both views of a single ruling. If they are temporarily congruent you might not be able to tell which you were using, but when they conflict you have to come down on one side or the other: conform to the rule despite the apparent illogic, or conform to the logic overriding the rule. (You might subsequently adjust the rule to try and make clashes less frequent, but at that moment, you came down on the side that the game-world trumps the rules.)
Game systems tend to favor one view over the other, even if they don’t make it explicit or apply it consistently across all decisions. Even the same rule often can be viewed one way or the other by different gaming groups. In original D&D, for instance, the game explicitly took the view that the rules were approximations but in every case the referee was the final arbiter; nevertheless there were rules such as Magic Users being forbidden to wear armor which weren’t explained in terms of game-world logic, leaving different groups on their own to either come up with explanations to justify the rule so that causality still flowed from the game-world to the rules (e.g. “armor is too restrictive, MUs can wear it but any attempt to cast spells will fail”), or to reverse the direction for that rule and say “Magic users can’t wear armor because that’s the rule. There is no why.” (Or perhaps by an appeal to a meta-game consideration, such as “MUs can’t wear armor because that would be unbalanced.”) Note that if the group followed the first tack, there would be further in-game consequences that flow from it, such as MUs having their companions carry armor around so that when they ran out of spells they could armor up. If the group took the latter tack, there’s often an awareness that the world is operating in strange and arbitrary ways. Much gaming humor (such as in Order of the Stick) comes from making the characters as aware of the flow of causality from the rules to their world as the players are.
Some games simply make no sense in terms of the first view: you cannot really regard the rules as an abstraction of game-world causality without it becoming a gonzo humor game. For instance, in the PDQ (Prose Descriptive Quality) system used by games such as Truth & Justice, when a character takes damage in a fight, the damage can be applied to a trait such as Accounting. So Spider-Guy getting hit by a truck thrown by the Blue Boar makes it more likely that some time later in the campaign, some complication will crop up having to do with his Accounting, such as being audited by the IRS. But even the most pronounced rules-first, game-world as a result system such as Truth & Justice, Dogs in the Vineyard, or D&D 4th Edition will have fairly large areas of the game that can be decided not by interactions of the rules, but consultation with the logic of the game-world, such as ordinary conversation between the PCs and NPCs (at least where the PCs aren’t trying to “win” a conflict with the NPCs or gain information that the NPCs do not wish to divulge).
On the other hand, unless you’re running system-less, there are probably no “rules as model” games where the rules never yield a somewhat implausible result that’s nonetheless taken as the actual game-world result, if for no other reason than to speed the game along and not make each ruling a source of debate.
Nevertheless, there is a clear distinction between games which aspire to the view of game-world causality as primary and those that take the opposite approach, and not understanding which approach the game is taking can lead to debates, frustration, and anger. Players and GMs may be seen as trying to twist or undermine the rules or even cheat in a group that expects causality to flow from the rules when they reason from game-world causes to game-world effects; in the opposite situation they may be seen as stifling creativity, being rigid, or power-gaming when they reason from rules causes to game-world effect regardless of game-world logic (another common form of gamer humor, as epitomized by The Knights of the Dinner Table, particularly Brian).
When you GM and when you play you should try to remain aware of which direction causality is supposed to flow in the game you’re playing, so that you can keep it clear whether the rules are the alpha and omega, or are they, as it were just guidelines…suggestions, really.