I hear your hypotheticals about testing. I think this comes up often, by way of example, when somebody has been in the hospital for a long time. Long stays aren't always medical, they're sometimes due to social issues (adult children are out of town until Tuesday, so grandma won't be safe elsewhere and stays in the hospital).

We don't have a prior that checking her labs daily will have a medical benefit after the time we decided she should have been discharged. And there are some abnormal values in daily labs that are, at once, not uncommon to encounter and also too abnormal for the day of discharge. So we sometimes end up keeping grandma an extra day or two to give her more potassium while we have no idea what the day-to-day variation of serum potassium would be for a 'healthy' person similar to grandma.

This is a clear example of when we shouldn't order testing: there's no expectation of marginal benefit to us while there's a risk that we'll be forced to act based on the result. This is also an example of a time where a person might say, "Why aren't you checking her? What if she develops [x], [y], or [z] and you don't see it until she has symptoms?"