Do all the heavy lifting (big corrections) in a well behaved RGB working space. IOW, make it look as you wish and save that as your archive. Then setup a custom soft proof for the output device. Edits at this point are based on this output alone (you may find there isn't anything additional you can do). But if so, do this on adjustment layers or on a copy as you're editing based on this single device. Note that there are things you simply can't recover based on the soft proof. For example, you're going from a much larger gamut working space to an output space. You see some colors desaturate as you soft proof. Nothing will really bring that back to this output device. The soft proof is your reality check.
The Simulate check boxes are useful in getting a more realistic soft proof but at a cost. First, they WILL make the simulation look uglier (as Schewe would say, reality sucks). But having the paper white and ink black in tow does provide a much better soft proof based on the dynamic range of the paper and ink. Do NOT view the simulation update for one. Or do the update and then walk away from the screen for a few seconds. When you come back, the image will not look quite as ugly. Watching the soft proof update, especially with respect to the mapping of white messes with your brain big time. Your eye has adapted to the white of the display and when it remaps to the more realistic white of the paper, it looks pretty ugly to watch. On that note, the simulate paper white means that you can't have any palettes or menu items being displayed because that white doesn't undergo the simulation. Your eye adapts to the whitest white it sees. So the white in the image is now dull (but accurate) surrounding by white from the palettes and so forth. Result is the image isn't being seen correctly. That means Unfortunately you need to work in full screen mode (hit the tab key once, the F key twice). Makes editing pretty hard! So use this for viewing only. This is the best way to show a client the image too. If you show them something like a saturated image in Adobe RGB (1998), that's what they will expect on print. Ain't going to happen.
Lastly, the accuracy of this soft proof is going to be based on not only the display profile but the output profile as well. Since output profiles have two tables (one affects the actual output, the other the soft proof), if the output profile isn't a really good one, it's possible that there's a disconnect in the two tables. You might get output that doesn't match the soft proof as you'd like. Of course you're all viewing the prints correctly under a 5000K light box of which the profile (most profiles) assume is the viewing conditions.
