A 403 From Netflix Is Not an Outage
My status page used to treat a homepage ping as health. Official vendor feeds say otherwise, and a probe that gets blocked is still proof the edge is up.
I rebuilt /status because a green tile that meant "this URL answered" is not the same thing as "this company says it is up." Most of the board now reads the vendor's own feed: Statuspage, Status.io, Better Stack, Slack, Google's incident JSON, Azure's RSS, PlayStation's region file. AWS Health was the rude one — the public events document is UTF-16 with a BOM, and res.json() quietly mangles it. Decode the bytes first, then parse.
Some products do not publish a feed. X, YouTube, Netflix, Steam, Xbox, Nintendo, Midjourney: those rows are reachability probes. HEAD first, GET if the server hates HEAD. A 4xx still counts as operational, because consumer sites routinely refuse datacenter clients. Only a 5xx or a timeout is degraded. The card labels that honestly: Official feed versus Probe only. Probe rows have no history bar. The drawer says so: a ping has no record of past incidents, and this page does not invent one.
The other landmine was Next's patched fetch. AbortSignal.timeout produced 500s with no log line on Vercel, so the scanner uses a manual AbortController and clears the timer itself. A scan opens about ninety connections at once; one retry covers the odd drop.
The lesson I wanted on the page: a homepage answering is not a status page, and a 403 from a CDN is not an outage.