Often the test doesn’t depend on specific data. And when it doesn’t, it can be good to share that same hard-coded data set so you don’t miss things like someone’s mutating it in the code that shouldn’t be (because it’s not in the right state for the next test).