An internet-monitoring system flagged nine countries as “out of range” between 13:40 UTC on 2 September and the morning of 3 September: Indonesia, Thailand, Japan, Singapore, Australia, Vietnam, Mexico, the Philippines and Hong Kong. The alert came from IODA’s ping-/24 probes, which count how many /24 networks respond; the system marks a country out of range when that count moves five standard deviations or >50% from its 90-day baseline. Probe responses plunged in the same ten-minute window (Indonesia −52.3%, Japan −35.7%, Singapore −35.6%, among others) but three independent signals did not move: BGP route announcements held steady (routes did not change by even a tenth of a percent), traffic to Google followed normal daily patterns (Singapore was even higher), and the network telescope showed only minor noise. Aviation activity was unchanged. In short, networks stayed announced and serving users; only the share answering the probes fell.
The cause cannot be determined from these data - path problems, response filtering or probe faults are all possible - but investigation found an internal defect that removed the crucial context: BGP and telescope values were being stored empty because of a clock/timing mismatch, so the probe drop lacked the contemporaneous refutation that should have been recorded. The bug was fixed mid-September but earlier intervals cannot be recovered. The case shows that a signal crossing its threshold is a fact about that signal, not the world, that duplicating the same API is not independent confirmation, and that context signals must be both stored and monitored.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.