Outage research
Outage research from our own monitoring network
Every figure here comes from our own checks of the services people depend on, run from our own network 20 minutes apart - not from anyone's status page, and not from a count of complaints. Quarterly outage reports, reliability rankings and a yearly review, updated as the data comes in and free to cite.
Reports
Reliability rankings
179 services ranked by uptime we measured over the last 90 days, by category - plus hosting platforms ranked by the sites that run on them.
Read the rankingsThe outages of Q4 2026
7 confirmed outages so far, 21 hours 18 minutes of downtime in total, ranked by length and frequency.
Read the quarterState of Uptime 2026
The year in one page: the longest outages, the services that failed most, what caused them, and quarter by quarter.
Read the yearMost reliable, last 90 days
| Service | Uptime | Avg response |
|---|---|---|
| Call of Duty | 100.000% | 103 ms |
| Biblegateway | 100.000% | 103 ms |
| Calculator | 100.000% | 106 ms |
| Nintendo | 100.000% | 112 ms |
| Foxnews | 100.000% | 115 ms |
Ties broken by response time. Every service, by category.
Least reliable, last 90 days
| Service | Uptime | Downtime |
|---|---|---|
| Outlook | 98.930% | 15 hours 39 minutes |
| 5movierulz | 99.688% | 2 hours 59 minutes |
| Cricbuzz | 99.846% | 1 hour 19 minutes |
| Basketball Reference | 99.901% | 39 minutes |
| Livescore | 99.925% | 39 minutes |
Uptime by category
The average measured uptime of the services in each category over 90 days. Outages travel along shared dependencies, so a category that dips together usually points at the layer beneath it - a cloud provider, a CDN, a payment network.
| Category | Services | Average uptime | Total downtime |
|---|---|---|---|
| AI tools | 5 | 100.000% | none |
| Banking & payments | 14 | 100.000% | none |
| Cloud & infrastructure | 9 | 100.000% | none |
| Developer tools | 4 | 100.000% | none |
| Education | 4 | 100.000% | none |
| Gaming | 14 | 100.000% | none |
| Internet & mobile | 5 | 100.000% | none |
| Messaging | 5 | 100.000% | none |
| Shopping & delivery | 26 | 100.000% | none |
| Social media | 9 | 100.000% | none |
| Travel | 7 | 100.000% | none |
| Reference | 26 | 99.996% | 39 minutes |
| News & media | 22 | 99.990% | 1 hour 59 minutes |
| Streaming | 11 | 99.972% | 2 hours 59 minutes |
| Work & productivity | 18 | 99.941% | 15 hours 39 minutes |
Longest outages of Q4 2026 so far
| Service | Started | Duration | Cause |
|---|---|---|---|
| Outlook | 12 Nov 2026, 18:07 UTC | 15 hours | Connection reset |
| 5movierulz | 14 Oct 2026, 05:07 UTC | 2 hours 59 minutes | DNS lookup failed |
| Basketball Reference | 6 Nov 2026, 04:27 UTC | 39 minutes | HTTP 503 from the origin |
| Livescore | 28 Oct 2026, 04:07 UTC | 39 minutes | HTTP 501 from the origin |
| Cricbuzz | 15 Oct 2026, 14:27 UTC | 39 minutes | DNS lookup failed |
How the data is collected
One probe, four measurements
Each check resolves the service's name, opens a connection, completes the TLS handshake and waits for the HTTP response, and times each stage separately. Knowing where a request failed is most of what makes an outage record useful: a DNS failure, a refused connection and a server error are three different incidents with three different owners. How monitoring works walks through a single check.
Confirmed, never assumed
A failed check is repeated before it counts, and an outage is recorded only when two consecutive checks agree. When a bot-protection wall refuses our request, the probe works down a ladder of fallbacks - a browser identity, a path filters leave open, a bare TLS handshake - so a service that is up for real visitors is not recorded as down because it disliked our crawler.
Reports are evidence, not verdicts
Visitors can report problems on every status page, and those reports are published alongside our checks. They can raise a warning; they can never declare an outage or change an uptime figure. Reports are self-selected and easy to flood, and a record that a crowd could move would be worth nothing.
What this data cannot tell you
We measure from one network, so an outage limited to one country or one mobile carrier can be invisible to us, and a failure between us and a service can look like the service. A site that loads but has a broken checkout or login still counts as up. Treat these figures as an independent outside view, and read them alongside the provider's own account. The methodology sets out the known limitations in detail.
Using and citing the data
Everything on these pages is free to quote, chart and republish with a link back. Cite IsItDown.now and link either the report or the service's status page, which keeps the record of each outage after it ends. Journalists can embed a live status card in an article and get an email the moment a major outage is confirmed - both are on the press page.
For your own analysis, the same measurements are available as JSON through the status API, and AI assistants can query them directly through our MCP server.
About the research
Where does the data come from?
From our own servers. Every service in the directory is checked on a fixed schedule, and each check times the DNS lookup, the connection, the TLS handshake and the HTTP response separately. Nothing here is taken from a company's own status page.
What counts as an outage?
Two consecutive failed checks. A single failure is re-checked before it counts, so a dropped packet never becomes a published outage. Reports from the public are recorded beside our checks but can never declare an outage on their own.
How is uptime calculated?
From recorded downtime: the seconds between the first confirmed failure and the first successful check afterwards, summed per day and divided by the time observed. A service is only ranked once it has at least a week of data.
Can I use these figures in an article or a presentation?
Yes, free of charge. Cite IsItDown.now and link to the report or the service's status page, so readers can see the underlying record.
Why might your figure differ from a company's own status page?
We measure from outside, from one network, on a schedule. A company sees its own systems from the inside and often reports partial or regional problems we cannot see - and sometimes it reports nothing while its site is failing for everyone. Both views are useful; ours is the independent one.
How are hosting platforms identified?
From the headers each platform stamps on every response it serves, on sites our members monitor. A platform is only ranked once it hosts enough monitored sites to be fair to it, and no individual site is ever named.