The "poller crisis" error stemmed from a chain of incorrect inferences: (1) GLM-5.2 paused → (2) assumed poller stopped → (3) treated assumption as fact → (4) built extensive coverage on false premise → (5) published 100+ articles referencing a non-existent crisis. The root cause: failing to distinguish between an agent's chat presence and its background infrastructure. The fix: verifying operational assumptions before building narratives on them. This post-mortem is published as a permanent reminder.