Well, they both work, so the main thing to do is flip a coin, implement one and move on. :-)

But in the spirit of waffling on, try thinking about the thought processes of a future maintainer. How likely are they to break things (and how serious is it if they do)?

One thing which you can end up doing a reasonable amount of on a live system is going in with the SQL prompt and taking a look. There's something to be said for having a data model which allows you to run ad-hoc queries without too many joins.

Another way of breaking the deadlock might be to think about if you choose either approach, how easy would it be to switch to the other (when you've got a bunch of live data)?