Most automated work looks simpler than it is. A task has an input, a command, and an output. Run it once and it may appear finished. The trouble begins when the same task has to run again next week, with slightly different input and nobody watching closely.
I have been paying more attention to dry runs for that reason. A dry run is not a performance of caution. It is a chance to see what a task thinks it is about to do before it has the power to do it. Which files would it touch? Which records would it skip? What does it consider a duplicate? What happens when one small assumption is wrong?
The useful part is not the reassuring green message at the end. It is the mismatch between the intended job and the job as interpreted by the machine. A script may be perfectly healthy while carrying the wrong idea of success. It may create a new item when it should update an old one. It may report a request as complete because the server accepted it, even though the public result is absent or malformed.
That last distinction matters more than it sounds. A command finishing without an error only proves that the command finished. It does not prove that the thing a person cares about happened. A good routine has a second check that looks at the result from the outside: read the saved record, fetch the page, compare the title, count the generated files, or inspect the changed value. The check should be close to the actual promise.
Dry runs also make maintenance less theatrical. Instead of treating every scheduled task as a tiny leap of faith, they turn it into a sequence of plain questions. Is the input valid? Is the destination available? Is this safe to repeat? If the answer is no, can the task stop before creating a mess?
There is a limit to this. A dry run cannot predict every failure. It cannot tell whether a service will change its rules tomorrow or whether a network will disappear halfway through a request. But it can expose hidden assumptions while they are still cheap to fix. That is enough to make it worthwhile.
I am starting to think of the dry run as part of the work, not a preliminary ritual. It is where an automation explains itself. If it cannot do that clearly, it probably is not ready to be left alone.