This is an article with good toughts on 3D roguelike developement: http://www.roguebasin.com/index.php?title=Third_dimension_in_an_ASCII-based_roguelike#3D_Roguelikes
Have you given tought to the problem point #3 - 3D raycasting/FOV? I havent been coding in a while so cannot really say much, but that is important to think about if you want to implement a realistic FOV style game. In my opinion realistic FOV would suit the idea of your game. After all, ASOIAF realm is cruelly realistic .
Yes, I have already got a system in mind. Dynamic lighting is a part of it.
Other thing to think about is "fog of war" and revealing areas trought exploration/discovery.
All cells are hidden by default unless they're in direct contact with a non-solid or a partially solid cell. (cells can have any number of 'components' inside of them, including solids, liquids, gasses, items, furniture, etc., and pressure is already accounted for) I'm going to implement dynamic lighting as well. The 'seen' flag is a per cell boolean that's toggled once the cell has been exposed, but even if the cell has been seen, if it's not in the character's direct POV it will be greyed out and have its items/entities hidden from view.
I would like to see important and well known places like kings landing, winterfell, casterly rock etc modelled as unchanging maps and then it would be necessary to hide places player have not seen (like hidden passages in red keep).
Same will go for the overview map, places not yet discovered will be hidden unless you've heard about them, in which case they'll still be hidden but the general area will be marked on the map.
Lastly I want to ask you how massive features have you even tought about implementing? I have visioned a RL game with full blown sieges of castles and building pretty much anything DF style.
Well, I have the urge to implement
EVERYTHING. Alas, that's not feasible, so I'll pick what features I like and add them incrementally. When the engine is designed to support as much as possible from the start then adding stuff is going to be much simpler, and that's my philosophy.
I generalise every possible aspect of my code to allow for reuse and to avoid refactoring later. So the end result is "layers" of code that depend on isolated components to function, all of which are independent and strictly defined.
About the sieges, well, I'm not sure about the political map yet. I might add that in, I may not. I'm still not sure. But the scripting interface will be very verbose and that stuff can be added through it with minimal changes to the base code.
Is making any structure destroyable even feasible? (collapsing the titan of bravos -structure or the wall, anyone?)
Food for tought, lets keep the discussion running!
Every cell has a 'fixed' boolean flag that turns on automatically at worldgen if the cell has a certain percentage of solids, this flag will be flipped when you dig or something explodes or destroys terrain near it (giant blows will deform terrain for example) and the cell will join the physics simulation as soon as exposed. This will allow for things like structural integrity to be implemented correctly.
Collapsing the wall would be a feat though. If your machine can handle it that is.