As far as uniqueness goes, I tend to avoid using random numbers, since there’s no guarantee that in a given set of n tests run concurrently that there won’t be a numeric collision, I instead favor a timestamped approach. But I think you’ve given a very good synopsis on the pattern, and how it might be used. I think cleanup and generating of Ids, if you can manage that later, may save you some in run time, but I subscribe to a the notion that its best to cleanup when you can. Things work better if you’re in an environment where the new deployment environment, also resets up a base database for use, which IMO means you could ignore some of the cleanup parts.