Poll-based outage monitors miss half of outages and inflate their reported duration

A software developer comparing a 60-second WiFi poller against a microsecond-resolution 802.11 management-plane listener found the poller recorded only 7 of 16 actual outage episodes over a two-hour window. All eight missed episodes were short disconnections lasting around 14 seconds — well under the 60-second polling interval — revealing a systematic censorship bias rather than random sampling error. The poller also overstated average outage duration by roughly 1.8–2x, because it captured only the longer events and because its own remediation attempts extended the outages it was measuring. Despite these distortions, total recorded downtime was 9.58% versus the true 12.28%, making aggregate availability metrics appear deceptively healthy. The author filed the finding as a defect and proposed flagging duration estimates only when no automated recovery rungs had fired during the outage window.