Having been guilty of ‘spaghettification’ of my models in my earlier days working with the Power tools - I can honestly say that both methods have value.
I tried with my first-ever business model (built real-world, not in class situation) to clean it up about a year ago. I decided it was such a disaster that I had no choice but to start fresh. Unfortunately, I had to keep going back to the original to see what exactly I had done, and this caused a lot of frustration when I had two models open and needed to track where I was.
So when I decided to tackle the next one that needed cleanup, I took the approach of untangling the mess. What this looked like was a lot of post-it notes stuck to walls with additional notes ‘connected’ to them with bits of string. (My co-workers thought I was nuts, and for a short time I was willing to agree.)
Now, as I tackle cleanup efforts - for my data or someone else, I determine just how bad it is, and take my approach from that. If there is too much to untangle, then I start fresh, if I think that cleaning up a few tables and/or measures will do the trick then I just do that.
Regardless of what I do, I take time to COMPLETELY understand the model I’m cleaning up - determining where tables come from, what measures are interconnected and how they are used. Generally this still means sticky notes all over the walls… but that’s just because I like to see the data spread out when I’m problem-solving.