[...]
One of the major problems with constant changes is that it invalidates past assumptions. [...] you [end up with] a cascading set of changes that grossly changes the scope of the original minor change.
[...] if a goal is not set it can go this direction.
Yes, I've seen that happen as well, though with software engineering projects I've been more fortunate than you: most of them have had requirements nailed down and regular reality/sanity checks with good management.
Marc spent a long time
thinking about T5 and toying with concepts before he started writing the draft. He had material and concepts from 1977 to 1998 to mull over, defining concepts to distill, and a local minimum of complexity to seek. So the changes are not in the assumptions but (1) error corrections and (2) refinements. And a lot of editing.
The revisions Marc has posted change nothing about the core mechanics, which was the one thing he took the longest to decide on and the first thing he nailed down. Then people like me started asking "why in the world are you doing X/Y/Z?". When the successive chapters started coming out, then I started to see how a particular decision, generalization, specialization, or mechanism fit his vision. Or, perhaps, I started to see what form his vision took.
In short, Marc nailed down his requirements the way he wanted them, then started to work. He's always worked slowly with T5, for whatever reason. I'm tempted to say he's working on a dozen other things at the same time, but I wonder, if he had all the time in the world if he'd go that much faster.