I suggest that you make duplicate copies of your blend-file several times a day (in addition to the backup system), numbering and/or dating each one. Lock it and keep it … forever. You just never know when you might want to go back to it. If you want to “crumple up the piece of paper, throw it away and start over,” archive it first.
There’s definitely such a thing as “analysis paralysis,” and also, “the land of Diminishing Returns.” To help counter this, I will usually start a project using simple geometric shapes … to scale. I will get an idea of where I might want to put the camera, how “the show” might pan out (story-beat by story-beat …), and therefore, “what is actually likely to matter, versus what isn’t.” This turns out to yield two very significant and time-saving points.
The first is, “if you don’t see it, it doesn’t have to exist.” When they made Western movies in the desert, there was no “back side” to any of the buildings. They were all facades. And, in a very real sense, so is every CG model and/or set that you create. If you’ll never see the back side, don’t build the back side. Every model should be correct for the shot(s) in which it is intended to appear.
The second is, “edit, then shoot.” Use OpenGL Preview renders (and simple block shapes) to work out the shot-by-shot flow of the show. Then, use this to figure what you need. Build that. No less, no more. As the show progresses, one by one the rough shots will be replaced by finals … and they will match, exactly. Instead of wasting(!) hours “lovingly rendering” footage that you decide not to use, flip the process end-for-end. Decide what footage you will use, then “lovingly render” only that.
If you pour yourself into that mode of thinking, you’ll be surprised at how different your initial assumptions were, from what you actually wound up doing. “Human imagination” is the greatest secret weapon in your arsenal.