I suspect that is far less critical in Tinderbox than in most programs. One of the unusual features of Tinderbox (in my limited experience of using it) is that you can start with virtually no structure or setup to begin with, and allow the structure to emerge from working with the data. This is what I understand by the concept “incremental formalisation”.
Depending on the kind of work you are doing, it can be very useful not to have any structure at the outset. Indeed, in something like qualitative data analysis, having no preconceived ideas at the outset (apart from the inevitable cognitive biases that we all carry within us!) is probably critical to avoiding distorted results. But I’m wandering …
However, some time ago I accidentally started rather a long thread that touches on incremental formalisation User Attributes or Tags?. You might find it interesting to look over it. I seem to recall there was some very interesting input from James Fallows, regarding the usefulness of attributes rather than hierarchies (containers).