A report with nowhere to appear
A weekly job, a database table, an endpoint that answers correctly, and no screen anywhere. From either end it looks finished — only the missing caller says otherwise.
Here is a failure that does not look like one from any single angle.
A job is scheduled. It writes its findings to a table. An endpoint reads that table and answers, with a carefully shaped payload and a comment above it explaining exactly why it is shaped that way — never a bare empty list, because the reader needs to know whether the job ran at all or simply found nothing.
Every one of those pieces is correct. Call the endpoint and it answers properly. And nothing in the entire front end asks for it. No button, no screen, no wrapper in the file where all the other calls live.
Why nobody notices
Look at the job: it works. Look at the table: it fills. Look at the endpoint: it answers. Look at the screens: they all do what they do. The absence is not in anything. It is the space between two things that both look fine, and nobody greps for a space.
The way it surfaces, eventually, is far worse than a bug report. The job runs, writes something useful, and the person who asked for the feature sees no change. The obvious conclusion is that the feature does not work — when in fact it worked perfectly and had nowhere to put the answer.
Asking the question the other way round
So we asked it as a census instead of a hunch: take every read endpoint the server registers, and check that something in the front end asks for each one. Anything with no caller either gets wired to a screen, or gets a written reason why nothing calls it — because plenty of endpoints are legitimately used by something that is not a screen. A printer polls one of ours directly. A customer's own integration calls a set of others. Those are not orphans; they are clients that are not browsers, and writing that down is the point.
What is left after that is the real list, and it is short. It is now a test, so the next one is caught the week it appears rather than the month somebody wonders why a feature seems dead.
The general shape
The engine computes the right answer and nothing asks it. It is one of the most common failures we find in our own work, and it is invisible to the usual defences. Tests pass, because the parts are right. Monitoring is quiet, because nothing errors. Only the question who calls this? finds it, and that question has to be asked deliberately, because nothing ever raises its hand to be asked.