Built after a 32-hour outage that every dashboard called healthy. A redirect rule sent
a production hostname to a domain that did not resolve — the origin was fine, the
uptime check saw a valid 301, and nobody was paged.
“The endpoint responded” is not “the endpoint worked.”
3×
independent DNS vantages, quorum before anyone is paged
5
timing phases split out — DNS, TCP, TLS, TTFB, transfer
24h/7d/30d
availability and latency percentiles, gaps reported honestly