A prototype can look complete in a clean room. The harder test arrives when the data changes, the device is used in bad weather, the person operating it is busy, or yesterday’s assumption no longer holds.
Those edges are not annoying exceptions to the work. They are where the work explains itself. A maintenance tool has to respect the depot. A research system has to distinguish a derived result from a hopeful interpretation. A memory tool has to make its boundaries visible to the person who owns the context.
I like small systems because they make those questions difficult to ignore. They turn abstract principles into choices: what gets recorded, who can change it, what happens when the evidence is incomplete, and how the next person can recover the thread.