This is a conglomeration question when answering please specify which part you are addressing. I am looking at creating a maze type game that utilizes elevation. I have a few features I would like to have, but am unsure as to some of the implementation.

I have done work doing fileIO maze generation (using a key to read the file, and then generate the level based on that file), but I am unsure how to think about this with elevation in the mix. I think height maps might be a good approach, but don't know how to represent them effectively. for a height map which is more beneficial XML(containing h[u,v] data and key definition), CSV (item1 is key reference, item2 is elevation), or another approach that I have not thought of yet?

When it comes to placing the elevation values themselves what kind of deltah values are appropriate to have it noticeable at about a 60degree angle while not really effecting gravity driven physics (assuming some effect while moving up/down hill)?

I am thinking of maybe going to procedural generation at some point, but am wondering if it is practical to have a procedurally generated grid (wall squares possibly same dimensions as the open space squares), or if designing to a thin wall open spaces is better? this decision will effect the amount of work need on the graphics end for uniform vs. irregular walls.

EDIT: game will be a elevation maze shooter. levels/maps will be mazes with elevation the player has to negotiate. elevations will have effects on "combat" vision, and movement