The only consideration is "Does it make sense, in this world, for the situation at hand?"
But, as [MENTION=42582]pemerton[/MENTION] says, this must involve a good measure of personal aesthetics and expression of belief, since not every conceivable situation in the world can be imagined in advance, surely? This boils down to the players guessing what the GM's aesthetics and beliefs about the world are (the real world, that is, as that inevitably informs a good chunk of how we imagine fantasy worlds to work).
Really? I hadn't considered that someone might try to run an entire game, without so much as developing a theory about how magic actually works - at least well enough to answer anything that the PCs attempt.
Count me as another who has run many, many games with only, at best, a fairly cursory outline idea of how magic works in the world. In fact, from what I know based on conversation and correspondance with him, count Robin Crossby, the maker of HârnWorld as another, and one who designed game world elements professionally!
The DM creates the world. The PCs explore it.
I held to this as a theoretical paradigm for many years. I no longer believe it works (or even that it would be desirable if it did).
As for the rest of it, and genre conventions? The DM should know how the world works. You created the world, after all. You have responsibility for adjudicating these things. If the action makes sense to you, in your mental model of the world, then the players can do it within the world. If your vision of the world is not one where chaotic energies can be channelled into an item, then the world doesn't work that way. (And the characters, who have lived in this world for decades, should have a pretty good idea about whether such things make sense for that world; there should be no need to actually play out such an inevitable failure.)
You describe a situation where a player explores a complete and fully described world made up by the GM, but is that ever really a possibility? We do not have a complete and comprehensive understanding, collectively and with the full record of several hundred years of scientific enquiry, of this one world in which we live! We do not even have any very coherent explanation of how our own minds work, despite recent progress concerning the nature of memory and perception - it has become clear that we are barely scratching the surface, so far. So what is the chance that we can conceive of a fully realised world that is significantly different to this one? I would say the chance is slim indeed!
But I also think such a situation, at best illusory, is unnecessary. Other posts in this thread have crystallised one way that 4E gets around the problem, even, with its "say yes (to rolling the dice)" mantra. It seems to me that this can become quite analogous to the scientific method, but for a game world. Even better is that, as GM, I get to do some exploring of my own! As the participants in the game imagine potential possibilities, we do the equivalent of experiments. These experiments either work or they don't, as determined by the dice results, and just as with scientific experiments these results build the theories concerning how the game world works. You can think of the dice here representing not so much the variability of character skill or circumstances, but as the "voice of the world" telling us whether our theories concerning the "laws of nature" have weight. Just as with real-world science, a successful experiment on its own is not necessarily a deal clincher; we might say that future attempts to duplicate the experiment (or something similar) require a similar roll. Our grasp of the theory is still rudimentary and perhaps partial. But a failure of the experiment might say more; in 4E terms the initial experiment might involve a Moderate DC, but a second after a failed first becomes Hard - evidence is accumulating that our theory is wrong! Eventually, the DC for a duplication of what has worked before might become easy. In time it might become another established Power. Thus the "reality" of the world is established by all involved - and without any need to break immserion in the game world at all.
It's interesting that, in the context of such a system, DCs related to the "experimenter's" level actually have some logic behind them. After all, the roll is not intended to represent skill or power - it represents whether the world is really as the character conjectures it to be. Of course, the characters will, to some extent, appear to shape the world via this technique, since only those conjectures they choose to make will be tested - but is that not precisely the same with scientists in the real world?