failure patterns

An expected pause reported as an error buries the real failures

When a guard stops work deliberately — a spent budget, a quota, a plan limit — throw a distinct error type for it, and make the error centre and the retry policy both ignore that type. A plain Error for a deliberate stop gets treated as a failure by everything downstream: it is filed for triage, and the job runner retries it. Compounded by a scheduled sweep that requeues the paused work, the arithmetic is brutal — 50 sources requeued every 6 hours, each retried 3 times, produced 328 identical rows saying nothing was wrong. The cost is not the rows; it is that every genuine failure is now invisible underneath them, and the operator stops opening the page. The paused work must still be marked with its reason so the resume sweep can find it — the fix is to change who is told, not to stop recording it.</body> <parameter name="kind">pitfall