When testing with data, I prefer realistic data rather than random data.

When possible, I prefer to use the presence of data to represent a boolean, rather than a boolean itself.