I'm trying to be canon-compliant in terms of worldbuilding and all characters except Bella. So, yeah, Alice's power works pretty much the same way in the originals, although I've found excuses to bring it up more frequently. My (largely surface level, not physics-inspired) explanation for her limitations is that people's decisions are the only things liable to be affected by her advance knowledge. The weather doesn't change based on what she or anyone else does, and so she is a perfectly reliable meteorologist except at time scales so far in advance that butterfly effect type stuff starts adding up. However, when she sees things, even if she tells no one, her own decisions change, and so does what Edward can see in her head, etc. Alice's visions are: a) purely visual, b) do not tend to have vantage points from inside of objects, c) are only mostly under her control, and d) can be changed even after she's become very confident in them. She cannot see inside people's future minds. She can't perfectly decide what to see. The sorts of things that upset her visions after she's had them are reactions to that vision itself in her or others (like when she sees Bella telling her something that there is subsequently no reason for Bella to share), decisions that haven't been made yet being prompted by other factors (like when James's coven heard the baseball game), and people deliberately exploiting the nature of her visions (the way James did). In canon, it turns out that she can't see Quileute wolves or half-vampires, nor things they affect, at all - the initial supposition is that Quileutes are just too unpredictable because of how their shapeshifting is unexpected, but that's patently ridiculous. They often do predictable things and many have good control over their "phasing". In the last book she blames the gap on "half breeds" in general and says she can see humans because she was one and vampires because she is one. In neither case does canon indicate that she actual

If you imagine that the future is modeled by differential equations, and that Alice can discern the coefficients for these equations to within a certain value, there will be certain equations which are obviously determined (f~=f' and others which aren't (f~=.0001*f') In the same way that people feel that tables are solid, because the reaction to their hand is fairly determined, and pillows are soft, because the reaction depends on how you're moving your hand, it makes sense (to me) that predicting the future could easily have fuzzy areas. If this isn't how it works, then Alice should be able to do NP-complete problems in polynomial time, as in rationalist!Harry's experience with the time turner. Perhaps this will be explored?