While I agree with you on all your remaining points, one could argue that logic in tests has a non-zero risk of being flawed or buggy on its own. Worst case, a

While I agree with you on all your remaining points, one could argue that logic in tests has a non-zero risk of being flawed or buggy on its own. Worst case, a bug in the test's logic and another in the logic under test cancels each other out. That's why I've made it a habit to write only strictly linear tests in my own code, and use fixtures and parametrization to keep duplication to a minimum. Ultimately, there's no one true best strategy. Everything is full of trade-offs all the way down.