This is my kata for basic test-driven data modeling. I use shoulda on top of test-unit, no fixtures, and almost zero database hits. I also take heavy advantage of aliasing in the command shell, and snippets in TextMate to speed up my typing.
I think I have a good flow down, but I'd love to see what others do. Do you use RSpec, or vanilla test-unit? Maybe you use VIM or gedit instead of TextMate. Show your version of this simple kata!