I've never been a fan of "Big Requirements Up Front".
There is no better way for a person to figure out what they need than to just start building something and working with that. One figures out very quickly where the problems lie and what the needs really are.
Although painful to redesign the approach, some of what has already been built is very much still useful/pertinent. Regardless: one is getting much closer to the final product this way versus trying to figure it all out on paper before actually building anything.
You can follow progress via future blog entries here, via the "TW Architecture" tiddler in the "new" TW instance, and by comparing the "new" Game Night Score Keeper vs the now locked first version: