Limiting beliefs and unstated assumptions interfere with our performance. This is not a trifling matter. Here is yet another story of how that happens. TL;DR We often test a module by connecting it to a lightweight implementation of one of its collaborators, such as in-memory persistence. This works great for simulating happy paths, but it’s usually impossible to make an in-memory database fail the same ways that an SQL database would fail running on some remote server somewhere. This leads...
