Methodology
A status page is only worth reading if you can check its working. This page describes exactly what we measure, how often, from where, and where our measurements stop being reliable.
Most outage trackers use one source. We use three, and every status page says which one produced the reading you are looking at.
Where a service publishes a machine-readable status feed, we read it every 5 minutes. This is the strongest signal available: it comes from the people running the service, it is component-level, and it carries their own incident history. We currently ingest 24 such feeds, and we link each vendor's status page from its DownBits page so you can check us against them.
Every 5 minutes we make a real HTTP request to the service and record the status code and how long it took. A single failed check never marks a service down — we require repeated consecutive failures, because one dropped packet is not an outage. Redirects count as reachable. So do 401 and 403 responses: a service that refuses us is still a service that is running.
Anyone can report a problem without registering. Reports are compared against a rolling seven-day baseline for that specific service, so a spike means "unusual for this service" rather than crossing a fixed number that would be meaningless for a small portal and trivial for a global platform.
This is the part most status sites leave out.
A number of Indian government and exam portals refuse connections from outside India, or route so slowly that a probe times out, while working perfectly for someone sitting in Ranchi or Patna. A probe that treated those timeouts as outages would publish a portal as permanently down, which is worse than publishing nothing.
So for those services we report "unreachable from our vantage point" and say so plainly, rather than asserting an outage. When our probe and a cluster of user reports from inside India agree, that is a much stronger reading than either alone — and that is what we show.
Individual checks are kept for seven days at full resolution, then compacted into hourly availability and latency figures that we keep indefinitely. That is what the 7-day and 30-day charts read. Incident records are never deleted.
If DownBits is showing something wrong about a service you operate, tell us and we will fix the monitor or remove the reading. We would rather show nothing than show something false.