Service reliability report: uptime of 46 popular services

Key findings

1. Over the 18 days to 12 October 2026, 37 of 45 services stayed at or above 99.9% uptime; the lowest was Cloudflare at 59.3%.
2. We recorded 76 outage episodes (two or more consecutive failed checks) across 12 of 45 services; 33 had none.
3. The longest single outage was Cloudflare: 10241 minutes, starting 05 Oct 2026, 16:34 UTC (still ongoing at the last check).
4. By category, Communication had the highest combined uptime (100%) and Developer platforms & cloud the lowest (96.8%).
5. Across the 17 services we check over HTTPS, the typical (median) response time was 651 ms; the slowest was UIDAI (Aadhaar) with a median of 3511 ms and a p95 of 7060 ms.
6. Not counted above: Exotel — its own status page has shown an open incident for almost the whole window, which says more about how the page is maintained than about measured downtime.

Each sentence is computed from the table below at the time shown in the data window. Quote it with the citation at the end of this page.

By category

Combined uptime = all passing checks in the category divided by all its checks, over services with enough data.

Click a column to sort. Rows with insufficient data always sort last. * still failing at the last check. Response times are only recorded for HTTPS checks; a status-page check measures what the vendor reports, not speed. Open a service for its live status and recent outages.

Methodology

• What is checked. 46 services, each watched by one ordinary Vigil monitor every 5 minutes, around the clock, from Vigil's server in the EU (Europe). The monitors have no special treatment: they are the same checks any Vigil customer can create.
• Two kinds of check. Where a vendor publishes a machine-readable status page (GitHub, Cloudflare, Slack, OpenAI…) we read the overall indicator the vendor itself publishes. Those numbers reflect the vendor's own reporting: a problem the vendor has not acknowledged is invisible, and a “major outage” there may affect only some customers. A major or critical indicator counts as a failed check; minor issues and maintenance count as up. Where no status page exists (most Indian banks, brokers and government portals) we make one HTTPS request to a public page and record whether it answered successfully and how long it took.
• What an HTTPS check can and cannot show. It shows that a public page loaded from Europe. It does not log in, make payments or place orders, so it cannot see failed transactions, OTP delays, app-only problems or regional outages inside India; and a site that blocks or slows foreign traffic can look worse than it is for users in India.
• Confirmation retry. When an HTTPS request times out or cannot connect, Vigil retries once after 12 seconds before recording a failure, so a single dropped packet is not counted as downtime.
• Uptime = passing checks ÷ all checks in the window. Status-page checks where the vendor's page itself could not be read are excluded (neither up nor down). Percentages are truncated, not rounded up.
• Outage episode = 2 or more consecutive failed checks. A single failed check between two passing ones is treated as a blip and not counted as an episode (it still lowers uptime). An episode lasts from its first failed check to the first passing check after it, so durations are accurate to about 5 minutes.
• Response time: median and 95th percentile (nearest rank) of passing HTTPS checks, in milliseconds, measured from Europe, including DNS, connection and TLS set-up.
• Window. The last 30 days, or since checking began (2026-09-23) if that is shorter. A service with fewer than 50 checks in the window is marked “insufficient data” and left out of every finding and roll-up.
• Freshness. The page is recomputed at most every 10 minutes from the stored checks; the data window above shows the exact time of the latest check used.

How to cite

Vigil by PuzariCloud, “Service reliability report: uptime of 46 popular services”, 25 September 2026 – 12 October 2026, https://puzaricloud.in/research/service-reliability (CC BY 4.0)

Free to reuse under CC BY 4.0: quote, chart or republish the numbers with a link to this page. The raw table is in the CSV.