I'm in Lr. I'm not softproofing yet. I look at the photo, see what I think is needed based on the non-softproofed view and make my edits. Looks nice. Then I push "S" (for softproof, or perhaps a four letter word) and OMG, what happened to that nice photo I edited - I need to change this, change that, change the other thing. Editing twice over, whereas if I started the whole process under soft-proof it's once over, save perhaps for some tweeks depending on whether or not I decide to change the RI, which I consider to be an editing variable.
Think about what you've written above, then we'll move on. You state you look at the photo and see based on the non-soft proof view,
you need to make edits. So you should! You really think starting with a soft proof isn't going to cause you to apply just as many edits and with less options for that data? Some of the edits are based upon
one output and one RI. It can work, but it's rather inflexible. What you going to do when you want to reprint that image on your new (fill in the blank) printer who's profile and behavior isn't anything like the one you used previous. Or send it to a lab with a printer you can't afford. Sound like a good idea? Not to me. Now editing the image to appear as desired in an RGB working space which is
output agonistic makes far more sense to me. And Adobe. I can send that document to a printer without any edits
if the soft proof shows me I need no such edits with a specific profile and RI. I
may need some edits. And now, I make them
specifically for this paper, printer, profile, RI. I can go back any time to the non output specific edited image to any output device and I've lost nothing. Time or data.
You're often editing twice; one to get the image into an appearance within a working space (editing space) you desire. That's the big work. Output specific edits
may or may not be necessary. If so, it's pretty quick and easy compared to the first editing process. You have flexibility!
Starting with a soft proof and editing all the way through is a bit like the old workflow days where images were drum scanned, in CMYK, to output size. And it worked. But you found you often needed to get another scan, then another because the image needs output specific scanning (size, color space in CMYK etc). RGB workflows changed all that for the better, unless you were selling scans! Scan once, use many. Same applies to RGB working space: Edit that master archive first, use it for all future output needs. Apply specific output edits IF necessary when necessary. Some of this can't be put back into the toothpaste tube so workflows that are based on flexibility, future output needs and changes in output devices seems to largely suggest to me, a 'start with a soft proof' approach is inflexible, nor addresses future output needs and future changes in the output device.
RGB working space were designed from day one to be divorced from any device; your display or any printer. They were designed with the same idea in mind as scanning once and using many.